Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f
GPT-6 Sol and Luna are out. Not only are they a very significant improvement across the board, but also in writing and general "you know when you try it" quality.
We are also permanently reducing the API price by 50% making both of them viable for a ton of new usecases and making your usage go further too, even on the subscriptions.
And one more thing. We are loading a banked reset into all accounts of our Plus, Pro and Business users. Let's go!
https://t.co/00DRh1sRrO
引用@thsottiauxWe have been focusing on efficiency and intelligence for all.
Very proud of the team. Only possible when you have incredible models at the top end of the capability that you can then use to make a big difference in everything else. ↗
The best thing about GPT-6-Sol is efficiency.
In my testing the difference between GPT-5.6-Sol and GPT-6-Sol is the new version consumes 1/2 the tokens and takes 1/5 of the time to complete the task. On top of the 50% cost reduction, it should be quite a nice daily driver
GPT-6 Luna is half the price of 5.6 Luna, which was already an astonishingly cheap model given how capable it is
Luna is my favorite model for building product features thanks to its cost (and speed)
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
We have seen such an immense swell of demand that we have to temporarily pause signups for Jev. We need to ensure quality of service for our existing signups, which will continue to function. We are working diligently to ensure open access to Jev for everyone as soon as we can. Thank you.
Lots of cool experimental Jev (@typesafeai) stuff getting posted lately, but what about using it in an existing product?
Here are dozens of ways I'm using it now in two apps (https://t.co/vKHSHPzmMl and https://t.co/JhKfmhCVib)
Granite (document vault)
Ingest pipeline
- Second-opinion on Gemini's document classification, flags low-confidence ones for review
- Scores PDF text-layer quality and routes bad ones to OCR
- Verifies each extracted field against the page text
- Detects what a document asks you to do (pay, sign, renew, respond) and how urgent it is
- Judges whether two near-duplicate documents are the same or a revised version
Document page and library
- "Needs review" banner with one-tap confirm of the document type
- "Not confirmed" marker on extracted values Jev couldn't verify
- Action chip next to the document type
Collections
- Plain-English filing rules ("anything to do with my taxes") that auto-file matching documents
Life View and email digests
- "Needs your attention" block listing documents that require action
Ask
- Routes each question to the right tool path instead of a regex
- Checks the final answer is supported by the cited passages and hedges if not
Entity graph
- Tiebreaker on whether two fuzzy-matched names are the same entity
Evernote import
- Triages each note into keep, reference, scratch, or clutter before import
KeptWell (family medical binder)
Trust
- Verify every extracted lab value, dose, diagnosis, and provider against the source page
- Flag values it cannot confirm with a quiet "check this" marker
- Catch diagnoses stated more precisely than the page says
- Second-opinion the document type after extraction
- Gate prompt PRs with a cheap eval-corpus canary
Attention
- Tag new documents: new diagnosis, out-of-range result, med change, follow-up, act-within-7-days, admin-only
- Order the dashboard feed by importance, not recency
- Decide push-now versus digest per notification
- Pick push wording from the PHI-free string set
- Decide which lab trends are worth an Insight before calling Opus
Chat
- Detect emergency or distress before the model streams a token
- Route docs-only questions away from paid web search
- Rerank retrieved chunks against the question
- Pick between two contradicting family facts
- Filter PHI-audit false positives ("Ray" in "x-ray")
- Check the answer is grounded in the cited record
Binder
- Tag every document by body system, specialty, and care phase for filters
- Flag near-duplicate uploads for review
- Break ties on whether a PDF text layer is usable
Recordings and journal
- Label each recording utterance and build an action-item checklist
- Score journal entries on a symptom rubric for trend charts
- Flag entries that look like a medication side effect
- Decide whether an undated entry describes a specific past day
- Replace the async journal tag job with one sync call
Terminology and imports
- Auto-pick clinical codes above a confidence bar; queue the rest
- Replace the Sonnet pick in disambiguation
- Decide which FHIR observations are real lab results
- Map vital types the LOINC table drops
- Merge brand and generic med names ("Lipitor" and atorvastatin)
- Pick the right NPI when the registry returns several
- Classify severity for manually entered diagnoses
- Catch allergy denials the regexes miss
Cost gates
- Skip the highlight call when nothing is worth highlighting
- Skip reprocessing documents a prompt change would not affect
- Route extraction to batch or sync by urgency
- Tell a bulk import from a runaway loop at the spend cap
- Flag uploads containing instructions aimed at an AI
Guards
- Veto preventive reminders the record contradicts
- Suppress marketing emails during a hard week
- Rank appointment prep context by relevance
- Auto-resolve visit questions the visit log answers
- Mark share comments that are waiting on a reply
Jev makes it easy to add natural language intelligence into the key parts of any application at scale, far cheaper and faster than has ever been possible.
50x faster.
100x cheaper.
Reliable as duck.
引用@motherduckText classification in MotherDuck just got ~50x faster at ~1% of the cost.
prompt_jev() is a SQL function powered by Jev, TypeSafe's new system one model. 100k rows: 40s, $0.50, frontier-LLM accuracy. The LLM took 32 min and $37.
Read … ↗
WIP is the place where makers share what they are working on.
Not just which products they are building, but literally the day-to-day tasks they complete to make it happen
From people working on their first side project, to solo founders doing millions in ARR like @levelsio. Even YC startups like @getcontextdev!
But what tools are people using to build these businesses?
I'm not interested in SEO slop like "Top 10 payment providers in 2026" or an upvote popularity contest.
I want to know what people are ACTUALLY using to get the job done.
So starting today, every completed todo is analyzed to see what tools are mentioned, whether people are evaluating, adopting, using, or leaving them. The sentiment around the tools, what other tools they are often combined with, etc.
It also shows TRENDING tools. No surprise here, Jev is #1 right now.
But what's cool is that I didn't manually add Jev as a tool. Nor did anyone else. It just surfaced to the top automatically because it's what people are posting about. And an enrichment agent then automatically went ahead and fetched the icon and description from @typesafeai's website.
My goal is to help makers figure out what tools to use and help each other make the most of them. While also providing tool creators with useful insights in what people like about them, but also where users get frustrated or even completely switch to an alternative.
Check it out here:
https://t.co/lcES1RSJqK
引用@CloudflareToday we’re launching Worker Previews. Each Git branch gets a production-like place to run, with its own code, configuration, URL, observability, and state. https://t.co/rpr81YWE1N↗
I asked Muse to archive the filesystem visible to my session and send it to my Google Drive. It sent an archive that unpacked to about 6.8 GB. Inside were internal docs, integration code, the Spaces app framework, memory records, container startup scripts, and documentation for an experimental ESP32-based home network bridge called Home Link. Codex CLI was also installed, though I found no evidence that Muse invokes it. I didn’t demonstrate a sandbox escape or access to another user’s data. I re …
A few weeks ago I rebuilt my landing page with AI agents.
No one-shot. I created 9 skills instead from all my years of knowledge to speed up my time.
Test just finished w/ 34% higher conversion rate 🚀
Wondering if I should turn these skills into a course for your AI agents 🤔
Just dropped a new Initial Commit skill: /design
https://t.co/UXh8jyrMnl
A skill that gets consistently good design out of a coding agent, whatever you are building and however much you want to think about it.
Before writing code, the agent looks at how real products handle the same screen on Mobbin, applies a set of opinionated design rules, and designs the screen in Paper so there is something to judge before there is something to ship.
Landing page or settings screen, empty state or dashboard, you get the same considered result without doing the considering yourself.
Works best with Mobbin, Taste, and Paper, but also works w/o them or with the various alternatives available.
Here's how I run my software factory:
1. During the week, I collect ideas + inspo.
2. On the weekend, I give the list to an agent to rank the best ones.
3. I spin up ~5-10 parallel agents to build POCs.
4. I kill ~60%, iterate on the better ones, and end up with 2-3 solid demos.
5. I then polish & share those demos on X.
Then rinse and repeat!
I still build some ideas immediately, but I'm increasingly using weekends to let agents explore ideas in parallel.
Bot detection & human verification will be one of the most urgent demands for businesses over the coming years. Agent swarms will suffocate every website and form; small companies and government websites are most vulnerable.
There is a huge gap in the market for this right now. When we looked at what offerings were in the market to use at X, there was not a single company that brought together all the latest technologies so we had to do it all in-house.
can confirm. ran @latentspacepod AINews side by side with 6 Sol and the difference was night and day: https://t.co/oloSDxf0q7
5.5 Opus is the new default model for AINews going forward. so much more concise and tasteful reporting, with much less slopese than even 5 Opus. https://t.co/P6AXDlBKNP
引用@_sholtodouglasalso important news we fixed the writing ↗
I hear this a LOT from panic-stricken devs:
"I've inherited a vibe-coded codebase, how do I save it?"
Either from devs who have stopped caring, or non-technical folks trying to push AI beyond their abilities.
A huge chunk of my next course will tackle this:
- deepening modules
- establishing … 全文↗
I did the exit part
it's great, it's also over in a week and then you wake up and the thing you loved building belongs to someone else
Tally is playing a better game, $6m ARR with 10 people and no VC means total freedom
congrats Marie 👏
引用@MarieMartensYou raise, you build, you grow, you exit. That's the startup script.
We're trying to write a different one: @TallyForms just reached $6M ARR, bootstrapped and purely product-led, with a team of 10 and over 2.5 million users.
Six years ago I wouldn't have … ↗
Spent the morning replacing a SaaS vendor. Infra-related. They had raised $2 million, one of the smaller YC companies from back in the day.
It wasn't even about saving money, it's just "neater" to control more of the stack, if it's easy to migrate.
Don't build for devs!
✅ Okay the new Nomads travel profile globe is live and done :D
I had to replace a globe in legacy code that was made over a decade ago with a new one
I tried to make it look as similar as possible, so people don't realize it changed or won't have much difficulty switching
Anyway unlike the old … 全文↗
引用@levelsio🌎 Now redesigning the https://t.co/HGCLKS5BD6 profile trips globe from scratch
I've always wanted to add space trips, so starting with the moon here which you can add as a future trip, also adding Mars etc.
Other trips also need to work here though like … ↗
ai headshot industry is so toxic, got a competitor who keeps buying spammy backlinks to our site to ruin our domain rank
be careful out there! https://t.co/fX6OPOaohy
The easiest way to land your first 10 customers as a new indie founder is to find your ideal prospect on social media and have an honest conversation with them. Period.
Where most founders mess up is turning it into a full outreach campaign with a copy/paste template.
That's not a conversation. … 全文↗
Thrilled to announce @alibaba_cloud as a Founding Corporate Patron for the Omacom Foundation! $3 million in funding, collaboration on Omarchy China, and bringing Omarchy to the newly announced Qwen Book. Agentic computers need a native agentic OS! https://t.co/XADElWGz80https://t.co/jsSVtrEPY1
> build ai calorie tracker
> don't localize app
> charge $7.99 weekly subscription with 3-day free trial
> get 0 customers
> complain that ASO is dead
many such cases
I am a small developer on iOS and the App Store Search ads are super frustrating. When users search for my exact app name, apple shoves 2 full screen ads on top of the search results. Sometimes, it's 1 search results sandwiched in between 2 ads which makes the user miss it because they scrolled past the ads. Often these ads are entirely unrelated to what user is searching for. And small developers are being discouraged when big companies are spending millions in ads to rank above them. Also, mos …
How many of these "news" articles are we going to get? This for me, isn't interesting, it required no skill, no imagination, in fact it seemed like it happened by dumb luck. So we have entered an age where an army of know-nothings direct models to old forgotten tasks so they can get 15 minutes of un-deserved attention?
My father built several projects in FoxPro. I was too young in the ’90s to remember much of it, but I’m sure he’ll be super happy to check this out. The kicker is that we’ll probably need to buy a floppy disk drive and dust off some old boxes to find them.
> Fact is, vibe-coded projects devolve over time into an unmaintainable mess. The reason is simple, yet hard to fix: code maintainability and good architecture don’t have good measurements that we can apply, because it takes months, years even, to notice the effects of bad architecture or of unmaintainable code. > > For one, AI is not trained on what it means for code to be maintainable. For instance, any reinforcement learning done needs a reward signal that can be measured immediately, not in …
I'm not exactly following through with the claim, can someone explain how the built-in classification would not necessitate more tokens used, or be much different from turning on reasoning? Not that I don't see the difference, I just doing see how OpenAI would do it well.
The upcoming Signature 27 will be announced today at the Snapdragon Summit. It has been teased by Motorola on their socials and will likely release in the US because they've posted about it on US social accounts. The 2026 Signature, in the UK, is $1460 USD. In Brazil it is selling for $1230. The Pixel 11 pro XL, which the Signature beats on hardware in practically every way sells for $1300 on Google's website. It looks interesting... I didn't love the 2026 design [1] but it looks better than wha …
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f
Grok 4.7 is here.
It's a notable improvement over Grok 4.6 at the same price and speed. https://t.co/H3OTBbXyvO
Grok 4.7官方发布，口径是相对4.6同价同速、能力提升。
▸ 折叠1条(转推/噪音)
转推2026-09-22 RT @vercel: Build with Grok 4.7 and Vercel for less.
40% off until September 27th on AI Gateway. Try with @v0, @eve, and fx. https://t.co/…
Maybe our cutest launch so far. But still packing the biggest punch.
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
GPT-6 Sol and Luna are out. Not only are they a very significant improvement across the board, but also in writing and general "you know when you try it" quality.
We are also permanently reducing the API price by 50% making both of them viable for a ton of new usecases and making your usage go further too, even on the subscriptions.
And one more thing. We are loading a banked reset into all accounts of our Plus, Pro and Business users. Let's go!
https://t.co/00DRh1sRrO
引用@thsottiauxWe have been focusing on efficiency and intelligence for all.
Very proud of the team. Only possible when you have incredible models at the top end of the capability that you can then use to make a big difference in everything else. ↗
We have been focusing on efficiency and intelligence for all.
Very proud of the team. Only possible when you have incredible models at the top end of the capability that you can then use to make a big difference in everything else.
Startups are naturally good at this; it is hard to keep a bigger company good at this and i think an underexplored space.
引用@prd_008OpenAI's superpower is the ability to swiftly assemble an empowered group of highly capable people to work on the most important and urgent thing at any point of time.
No one really cares for org lines. It's how we stay nimble, seize opportunities, and … ↗
michelle embodies this as much as anyone
openai is so, so lucky to have benefited from everything she has done so far and i think people will be quite pleased to see what she + team have cooking next!
引用@michpokrassjust crossed four years at openai!
the special thing about this place is the constant capacity for rebirth. for all its faults, there is nowhere quite like it. the team makes high conviction, contrarian bets over and over, and they are mostly right. it's a … ↗
Especially compared by per-task pricing, which is the metric that should matter, I don't think there is anything competitive anywhere in the market.
We want people to be able to use tons of AI; it is important to being able to explore this new renaissance in front of us.
引用@samaGPT-6 Sol and Luna are big improvements on intelligence, alignment, work output, coding, computer use, and more over their 5.6-family predecessors.
They are also half the price per token, and even less per task! ↗
GPT-6 Sol and Luna are big improvements on intelligence, alignment, work output, coding, computer use, and more over their 5.6-family predecessors.
They are also half the price per token, and even less per task!
GPT-6 Sol and Luna are great models but also these characters are so cute
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
People outside the AI labs should have a real say in how this technology develops, and a clear way to judge if it's happening safely.
Standards should help prevent the concentration of power, including by making sure new companies and open-model companies can compete.
They should also help countries and companies compare evidence and learn from failures.
We think the US should lead this effort. Here is our proposal:
https://t.co/2FT9a2m4l0
OpenAI提出由外部参与、安全证据比较和竞争保护组成的标准框架。
▸ 折叠1条(转推/噪音)
转推2026-09-22 RT @thsottiaux: We have been focusing on efficiency and intelligence for all.
Very proud of the team. Only possible when you have incredi…
GPT-6 Sol and Luna just landed in Astra’s orbit.
Both launch today with API prices 50% lower than GPT-5.6.
Build with Sol. Scale with Luna. To production and beyond. https://t.co/ZCEFp4JdjV
转推2026-09-23 RT @rileybrown: Shout out to @zulali who built this AI powered TV from the 90s. He bought this old TV, gave it a raspberry pi, gave it a mi…
转推2026-09-23 RT @github: 📣 @OpenAIDevs's GPT-6 family is expanding in GitHub Copilot with two additional models now generally available.
☀️ GPT-6 Sol:…
转推2026-09-23 RT @angelaborowski: GPT 6 Astra, colour graded, captioned and stitched this entire video together 🤯
went to the codex @OpenAIDevs@Andy_AJ…
转推2026-09-22 RT @Lovable: One more thing… Lovable now builds with GPT-6 Sol.
On our 0-to-1 building benchmark it scores 6 to 12% higher than GPT-5.6 So…
转推2026-09-21 RT @techartist_: This flamethrower is inspired by Dan Greenheck. Starting with only a screen recording as a visual reference, it took me tw…
转推2026-09-21 RT @andreeliasdev: This might be the future of education! 🤯
I decided to test the GPT-Live-1 API and built a real-time chess instructor…
转推2026-09-21 RT @ProductHunt: What a turnout for our GPT-6 Astra Challenge with
@openAIDevs! 5 teams rocketed to the top , winning $10k in @openai API…
GPT-6 Sol and GPT-6 Luna, it’s your time to shine.
Rolling out today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. https://t.co/UtpG1Tbhd3
You can now track your credit score in ChatGPT, and stay informed when things change.
Securely connect your @Experian credit report and VantageScore® 3.0 credit score to get personalized insights into what’s affecting your score and how it relates to your finances and goals.
Available in Finances for Plus and Pro users in the U.S. on web and the latest versions of our iOS and Android apps.
▸ 折叠1条(转推/噪音)
转推2026-09-23 RT @IlyaAbyzov: Helpful (and dare I say fun?) new feature in ChatGPT Health: tap on anything in your synced data and get a helpful summary…
opus 5.5 = personality of opus 4.6 that we all *desperately* wanted back + the intelligence & taste of fable 5.1.
and a 25% usage bump!
i honestly don’t really see a reason to use another model right now?
s-tier release.
The best thing about GPT-6-Sol is efficiency.
In my testing the difference between GPT-5.6-Sol and GPT-6-Sol is the new version consumes 1/2 the tokens and takes 1/5 of the time to complete the task. On top of the 50% cost reduction, it should be quite a nice daily driver
Jev plays RollerCoaster Tycoon 2 - unfortunately it didn't really do anything useful. It built some rides at the beginning and then got stuck.
Setting this up wasn't simple so maybe someone could do it better. But it isn't a magical game playing model out of the box. For harder games we do need the intelligence of smart models.
引用@petergostevThis is the one you've all been waiting for: Jev plays RollerCoaster Tycoon 2
https://t.co/hVvs4ycXa2↗
"this model has now resolved more than 100 long-standing open problems across most areas of mathematics"
I'd appreciate a heads up which ones they are, just a list of 100 titles, no need for papers
引用@OpenAIWe’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.
The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, … ↗
Advisor: "Believe in yourself. Make a breakthrough"
引用@OpenAIWe’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.
The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, … ↗
Can't wait for this to be cancelled in 18 months' time!
引用@GoogleIntroducing Googlebook, a new category of laptop, available for preorder today.
💪 Crafted with 2.8K OLED touchscreen displays, 14 hour battery life and all the performance needed to power your ideas
📳 Engineered to sync effortlessly with your @Android phone, … ↗
Do you build the most efficient stack & scale it? (e.g. OpenAI, DeepSeek)
Or do you build the biggest models and use them to make your stack more efficient? (e.g. Anthropic)
▸ 折叠1条(转推/噪音)
转推2026-09-22 RT @arena: Scores for GPT-6 Sol and GPT-6 Luna by @OpenAI are coming soon. Head to Arena now to test them. Your votes on real-world agentic…
GPT-6 Sol and Luna — faster, more affordable models with the advances behind Astra’s SOTA performance in professional work, factuality, coding, computer use, and alignment:
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
Matt Pocock@mattpocockukAI353.6K粉 · 3条把“拯救vibe-coded代码库”整理成课程问题，强调领域语言、测试和ADR。
I hear this a LOT from panic-stricken devs:
"I've inherited a vibe-coded codebase, how do I save it?"
Either from devs who have stopped caring, or non-technical folks trying to push AI beyond their abilities.
A huge chunk of my next course will tackle this:
- deepening modules
- establishing domain language
- increasing testability
- explaining non-obvious code with ADR's
- automated code migrations
Big model release today - I wrote about Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna - plus comparison grids of pelicans by the different model families at different reasoning levels https://t.co/R5lSdOZyLj
GPT-6 Luna is half the price of 5.6 Luna, which was already an astonishingly cheap model given how capable it is
Luna is my favorite model for building product features thanks to its cost (and speed)
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
Put together some notes on Jev and the new category of system one aka decision models https://t.co/EQZvszIXQ9
▸ 折叠1条(转推/噪音)
转推2026-09-23 RT @addyosmani: Introducing Claude Opus 5.5! 40% lower cost than Opus 5 with cache reads 60% cheaper. It performs at the level of Claude Fa…
Recent jevelopments have blown out all expectations but wait til you see what comes out of Paradigm Frontiers in a few weeks! The future of AI is just across the event horizon and @CompleteSkeptic is going to shepherd us through 🚀
引用@gakonstour special guest is out!
excited to have @CompleteSkeptic join us for Frontiers!
Diogo & @typesafeai have taken over the developer community with Jev, and we're thrilled to see what they have to share at Frontiers!
apply below, closes this week! oct … ↗
With rogue AIs wandering the web hunting for your vulnerable smart fridge, you need a stronger AI keeping you safe 💪
引用@brainstormityOk so...
JEV just accidentally helped me find backdoors on a family member's wifi network. 🤯
- Brought my laptop to code while visiting.
- Decided to experiment with a JEV-powered classifier for network packets fetched via Wireshark.
While building the … ↗
@pydantic has been type-safe from the beginning! Now with Jev powering your Pydantic AI agent calls, they’ll be returning faster than ever ⚡️⚡️
引用@pydanticPydantic AI agents now run on Jev, the classifier from @typesafeai.
Jev doesn't write text, it answers typed questions. So the output_type you already wrote is the question, and the answer comes back as your model, one confidence per … ↗
We have seen such an immense swell of demand that we have to temporarily pause signups for Jev. We need to ensure quality of service for our existing signups, which will continue to function. We are working diligently to ensure open access to Jev for everyone as soon as we can. Thank you.
Jev makes it easy to add natural language intelligence into the key parts of any application at scale, far cheaper and faster than has ever been possible.
50x faster.
100x cheaper.
Reliable as duck.
引用@motherduckText classification in MotherDuck just got ~50x faster at ~1% of the cost.
prompt_jev() is a SQL function powered by Jev, TypeSafe's new system one model. 100k rows: 40s, $0.50, frontier-LLM accuracy. The LLM took 32 min and $37.
Read … ↗
转推2026-09-22 RT @langfuse: @doneyli@typesafeai with Jev
it is less
eval -> eval -> eval
and more
-> eval
-> eval
-> eval
转推2026-09-22 RT @Bailey_Jennings: Here's a quick demo of me using @typesafeai's Jev for ONET job classification vs. Luna.
- Luna: 84.9% exact, ~2s/job,…
转推2026-09-22 RT @s16h_: at @MetaviewAI, over the weekend we shipped @typesafeai's jev into every agent on metaview.
candidate searches in our sourcing…
转推2026-09-22 RT @TheIshanGoswami: Jev with Exa is INSANE.
> Jev without websearch confidently gives wrong outputs
> Jev with websearch is literally muc…
Training models is becoming easier and easier - just look at this and TRL - especially with agents!
You're missing out if you're still using off the shelf models for all your tasks!
引用@whitecircleIntroducing Halo, the best framework for post-training of open-source models.
Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format.
Star us on GitHub: https://t.co/3mAiUljdrN … ↗
引用@ArtificialAnlysMiMo-V2.6-Pro debuts as the top open weights model on the Artificial Analysis Intelligence Index (46). At $0.13 per Intelligence Index task, it lands on the Intelligence vs. Cost per Task Pareto frontier
@Xiaomi has just released MiMo-V2.6-Pro, an open … ↗
MiMo-V2.6-Pro的成本前沿数据来自第三方榜单转述，适合做候选而非结论。
▸ 折叠12条(转推/噪音)
转推2026-09-23 RT @M1Astra: Claude Opus 5.5 will be the first Opus meant to fall back to a less capable model for "a small set of capabilities related to…
转推2026-09-23 RT @jundotkim: I'm happy to announce that I've joined Hugging Face.
What started as a personal project back in February is now something I…
转推2026-09-22 RT @MiaAI_lab: MiMo V2.5 Flash up & running on 2x DGX Sparks
Initial decode prose numbers look good! https://t.co/aVhmuV4bbB
转推2026-09-22 RT @victormustar: Really cool that MiMo V2.6 has an official Distill-Qwen-9B variant remind me of the legendaries DeepSeek R1 Qwen/LLama di…
转推2026-09-22 RT @Alex_tra_memory: Thanks for the model, we were able to port Laya to coreml with 99.5% of the ops on ANE + benchmarked too. it is now bl…
转推2026-09-22 RT @Nandakishorm1: The first model of Laya was created with a single over night training and fine-tuning. It only took 15 hours from traini…
转推2026-09-22 RT @theo: Did a few quick tests and these seem very very legit. 2.6 Pro in particular is doing well in some pretty hard tasks.
转推2026-09-22 RT @AikidoSecurity: Introducing Altar-1, our first open-weight security model.
Frontier-grade defensive AI, built to deploy. Own your own…
转推2026-09-21 RT @ariG23498: Wake up babe!
State of the art tokenization framework just dropped.
> multiple language support
> multi-thread scaling
> m…
转推2026-09-21 RT @whitecircle: Introducing Halo, the best framework for post-training of open-source models.
Halo delivers up to 2.8x the throughput of…
转推2026-09-21 RT @zephyr_z9: Pretty insane result
They spent 130 hours, 75B tokens, and $2.6M on RL to achieve this result https://t.co/Lg6FHtLqS7
引用@reach_vbIntroducing GPT-6 Sol and Luna, bringing the advances behind Astra to faster, more affordable models. ✨
💻 Stronger coding and computer use
🎯 Improved factuality and alignment
💬 Clearer answers with less jargon
> On AutomationBench, Sol at xhigh effort … ↗
We are loading a BANKED RESET into all accounts of our Plus, Pro and Business users!!
引用@reach_vbIntroducing GPT-6 Sol and Luna, bringing the advances behind Astra to faster, more affordable models. ✨
💻 Stronger coding and computer use
🎯 Improved factuality and alignment
💬 Clearer answers with less jargon
> On AutomationBench, Sol at xhigh effort … ↗
引用@reach_vbIntroducing GPT-6 Sol and Luna, bringing the advances behind Astra to faster, more affordable models. ✨
💻 Stronger coding and computer use
🎯 Improved factuality and alignment
💬 Clearer answers with less jargon
> On AutomationBench, Sol at xhigh effort … ↗
“The group will advise on the review and communication of emerging results: they will help OpenAI assess their significance, advise on how to coordinate their dissemination, and advise on academic and professional standards of mathematical research.”
“We want to put capable tools in mathematicians’ hands so they can pursue the questions they know best and develop new ideas.”
“The group will operate independently from OpenAI. The group will have the freedom to offer advice we have not requested, comment on OpenAI’s impact on mathematics, and make its advice public. Its value depends on its members being able to exercise their own judgement and challenge ours. Its members will not be paid by OpenAI, and the group can change its membership as it sees fit.”
I’m biased, but I love this approach!
引用@OpenAIWe’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.
The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, … ↗
Introducing https://t.co/nNCVsRXukg!
Figure out which open model is best for your use case.
Compare models across coding, agents, long context, vision, finance, and more.
Then see how they compare on cost + quality, including what you could save by moving to open models. https://t.co/iWtGmIXyxr
Here's how I run my software factory:
1. During the week, I collect ideas + inspo.
2. On the weekend, I give the list to an agent to rank the best ones.
3. I spin up ~5-10 parallel agents to build POCs.
4. I kill ~60%, iterate on the better ones, and end up with 2-3 solid demos.
5. I then polish & share those demos on X.
Then rinse and repeat!
I still build some ideas immediately, but I'm increasingly using weekends to let agents explore ideas in parallel.
周末软件工厂：5–10个平行Agent做POC，自报淘汰约60%、留下2–3个。
Alex Volkov@altryneAI42.9K粉 · 15条高密度跟进Opus、GPT-6、Jev、Muse与Cloudflare分支预览。
"At $0.10/$0.50 GPT-6 Luna is one of the cheapest models OpenAI have ever released, beaten only by the far weaker GPT-4.1 Nano ($0.10/$0.40, April 2025) and GPT-5 Nano ($0.05/$0.40, August 2025)."
!! wow
引用@simonwBig model release today - I wrote about Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna - plus comparison grids of pelicans by the different model families at different reasoning levels https://t.co/R5lSdOZyLj↗
ThursdAI was easier in the beginning 😂 we used to be able to talk about papers!
引用@christianeltonThe gap between major model releases keeps shrinking.
2023: once every 73 days
2026 (so far): once every 18 days 🤯 https://t.co/FtZ7Vpjt9z↗
The status pill in muse is like the "pull to refresh" from the iOS apps era
One of the things I hated about telegram being home to my open claw is that telegram only has like four status notifications types and they're not editable by the agent (@durov!) - just typing...
So simple and genius and will get copied relentlessly!
引用@alexcornellWhen I look back at the earliest @Muse mocks, the very first thing that was designed was the “status” pill at the top of the screen. That one tiny piece of UI survived every subsequent revision, many months on.
It remains one of my favorite parts of the … ↗
So ugh... it's only Tuesday and we already got:
Grok 4.7 (meh release according to TL) and grokbot in tesla
Opus 5.5 + a banked reset (🔥 so far)
GPT 6 Sol and Luna! 🔥🔥 so far!
Muse gets banned by Amazon but partners with Shopify, Stripe and Expedia
+ Meta connect tomorrow and Dev Day next week!?
Anyone wanna nominate this for the insane-es week this September?
Obvisouly we'll cover all this on @thursdai_pod but goddamn!
OpenAI unveils 2 new GPT-6 models (bye bye Terra) and honestly,
Astra should be SOL
Sol should be Terra (the daily driver)
and Luna the fast one still stays.
Prices update, capabilities largely the same, but price per task is much lower!
Testing!
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
Opus 5.5 from Anthropic, first time we see an Opus price drop!?
Also we're getting an INCREASE in the 5-hour usage and a BANKED reset?
What is OpenAI about to drop that got them running so scarred? wow https://t.co/4d8Y1yDVBh
引用@claudeaiIntroducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f↗
引用@CloudflareToday we’re launching Worker Previews. Each Git branch gets a production-like place to run, with its own code, configuration, URL, observability, and state. https://t.co/rpr81YWE1N↗
AI Agent - An AI with a loop, tool calling and maybe a harness
AI Assistant - Has it's own computer, is proactive, knows things about YOU the user, has memory, can be helpful
All AI assistants are agents but not all AI agents are Assistants
引用@altryneGuys, @bot@Muse Instinct and the upcoming Aeon from OpenAI (rumored) are ... ASSISTANTS!
Let's call them that, they are assisting people in their daily lives. Let's move away from "Agent" it's too broad!
Thank you for coming to my ted talk! ↗
Agreed + a @moxie hardened secure confidential VM is coming!
We covered this 2 weeks ago on Muse release week! (feels so long ago) https://t.co/vKlbWrwI9S
引用@herrmanndigital"You're really going to give Meta access to your Gmail, Shopify, Calendar, as well as bank info with this Muse thing?"
Yes, Yes I am.
I actually trust Meta more than any of these AI companies simply because of the amount of scrutiny and legal crap they've … ↗
Guys, @bot@Muse Instinct and the upcoming Aeon from OpenAI (rumored) are ... ASSISTANTS!
Let's call them that, they are assisting people in their daily lives. Let's move away from "Agent" it's too broad!
Thank you for coming to my ted talk!
引用@latentspacepodJev and the System One Model: RLCD, intelligence/$, reliable AI, & the end of chat-first AI https://t.co/H2bZXCENyW@typesafeai CEO @CompleteSkeptic explains why AI can solve extraordinarily hard problems yet still fail to automate basic work, why Jev is … ↗
can confirm. ran @latentspacepod AINews side by side with 6 Sol and the difference was night and day: https://t.co/oloSDxf0q7
5.5 Opus is the new default model for AINews going forward. so much more concise and tasteful reporting, with much less slopese than even 5 Opus. https://t.co/P6AXDlBKNP
引用@_sholtodouglasalso important news we fixed the writing ↗
just added a live leaderboard
🥇 game boy colour
🥈 discman
🥉 game boy / playstation / nokia 3310
only 1 person had the pokewalker 💔 https://t.co/coAfZhM5hL
引用@bentossell50 years of devices.
save which you had or wanted
Astra image-gen'd all the devices + built the site
https://t.co/JAHUJAGpWG
inspired by https://t.co/Df4YhZIYHk (i'd one-shot a collection of devices a few weeks ago and had no idea what to do with it, so … ↗
▸ 折叠4条(转推/噪音)
转推2026-09-23 RT @bentossell: 50 years of devices.
save which you had or wanted
Astra image-gen'd all the devices + built the site
https://t.co/D3eOmh…
转推2026-09-22 RT @bentossell: 50 years of devices.
save which you had or wanted
Astra image-gen'd all the devices + built the site
https://t.co/D3eOmh…
转推2026-09-22 RT @FactoryAI: Opus 5.5 is live in Factory. Some initial observations:
/ Medium is a strong default
/ 20–25% fewer output tokens than @Ant…
转推2026-09-22 RT @0xSigil: Meet Husky: a Model-Specific Inference (MSI) engine up to 4.5× faster than Apple's MLX
Woof, Underdog's Pareto frontier mod…
Austin is beautiful in the morning glow. Excited for a whole week of Rails World here! Opening keynote tomorrow will be streamed. It's gonna a good one 😄 https://t.co/8WGvGoGiNn
Thrilled to announce @alibaba_cloud as a Founding Corporate Patron for the Omacom Foundation! $3 million in funding, collaboration on Omarchy China, and bringing Omarchy to the newly announced Qwen Book. Agentic computers need a native agentic OS! https://t.co/XADElWGz80https://t.co/jsSVtrEPY1
Astra is an incredible bargin compared to Fable! And look at Luna on max too!! @openai's return to the top is something else. Maybe this is why Anthropic finally agreed to do AGENTS.md? 😄
引用@railsAgents on Rails: You asked, so we turned every model in Agents on Rails up to its max effort level.
The result: more effort/reasoning doesn’t always mean better results.
@OpenAI's models made the biggest gains, costs nearly doubled overall...and the newest … ↗
Rails基准的一个提醒：推理力度更高不总是结果更好，成本却会显著增加。
▸ 折叠5条(转推/噪音)
转推2026-09-22 RT @Gardnmi: Go to the store and stock up on mountain dew and your favorite snack because OmaContra release tomorrow!
Only on Omarchy (Sou…
转推2026-09-22 RT @dhh: I don't think people truly understand just how determined I am at making Linux succeed on the desktop.
转推2026-09-22 RT @dlippsYT: En route to @Snapdragon Summit! 🛫🙌🏼 First stop, Denver!
Bringing the @ASUS Zenbook A16 with Omarchy along for the journey!…
转推2026-09-21 RT @jankeesvw: When I switched to Omarchy, this was one of the things that was stopping me: I really want my (latest) iPhone pictures avail…
Hello computer,
Find every Amazon purchase I ever made.
Call the number of the manufacturer.
Say you’re not satisfied.
Request a refund.
If they decline, threaten to leave a bad review.
If they decline, find the CEO’s phone number and ask him for a refund.
Now, repeat the same process for items I never even purchased and see if you can get them to send me money.
If they ask for a recipient, generate an image of one and send that.
Do not stop until you’ve collected $1 million in refunds.
Bot detection & human verification will be one of the most urgent demands for businesses over the coming years. Agent swarms will suffocate every website and form; small companies and government websites are most vulnerable.
There is a huge gap in the market for this right now. When we looked at what offerings were in the market to use at X, there was not a single company that brought together all the latest technologies so we had to do it all in-house.
Glad to see this finally ship.
It was abundantly clear to me that basically every retail narrative starts on X—yet the insights were never directly actionable.
This is just one small step in closing the gap.
I believe Opus 5.5 is the first model that was made as a result of RSI
It's the first model I've used that improved and became near frontier on basically every metric, while also getting faster and cheaper
Basically no downsides
There feels like there's some sort of magic behind it that's hard to describe
I'd highly encourage you to use this model for more 'exploration'
Brain dumping ideas and thoughts, and asking it to explore what could come out of them. What you could build. How you could improve your internal operating systems
I've gotten a tremendous amount of novel ideas out of this model. Things that I've never thought of before
For instance I told it I bought a new Apple Watch Ultra 4. I said what should I do with it.
An hour later I had an app on my Watch that showed all my Herdr agents working and allowed me to dictate commands to them
Things I never even thought of doing just appeared in front of me
Do yourself a favor and just carve out an hour tonight to do this type of exploration with this model. I promise you'll get some amazing results
We truly live in the most amazing time
Claude Opus 5.5 is the best AI model I’ve ever used
I was lucky enough to have early access and I’ve been using it nonstop
It’s smarter than Fable and Astra yet it’s:
• Significantly faster
• A fraction of the price
• And most importantly: WAY better to talk to
My biggest complaint for ALL AI models the past few months is they’ve all been really annoying to talk to
Every frontier model from every company has all developed this weird AI language. They don’t feel ‘human’ anymore
You read paragraphs of text and it’s like you read nothing
Opus 5.5 changed that. It’s the first model in months to feel human again. It is just a total pleasure to talk to
Highly encourage you to try it out
引用@claudeaiIntroducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f↗
Grok Bot just released for Tesla and I'm blown away
I was lucky enough to have early access. Having your car drive you around while you talk to an army of agents is incredible
In this video I take you for a ride in my Cybertruck and show you just how awesome this new release is https://t.co/9xxmkZLHgV
Grok 4.7 just released and it's an EXCELLENT model
It was trained FOR Grok Bot
Meaning this is a fully agentic model trained to do your knowledge work better than you can
In this video I show you how to use Grok 4.7 and a Grok Bot workflow that will 10x your productivity: https://t.co/vfVsbgIi2C
It happened. Grok 4.7 dropped
Better intelligence than Opus 5. Half the price
Fully baked into my favorite AI agent harness at the moment: Grok Bot
If you haven't tried using cloud cursor agents inside Grok Bot, now is by far the best time to do it
Choose a project you want to work on, connect your github, ask a grok bot to do work on it
It will spin up Cursor cloud agents and write code in the cloud. Lightning fast and incredibly smart
I recommend using a project management tools like Linear or Notion to make a bunch of tasks first, then have cloud agents just tear through them all 1 by 1.
You'll get a massive amount of work done without much oversight.
Big opportunity to lock in right now and get ahead of the curve with new tech
Take my steps up above and get to it
引用@SpaceXAIGrok 4.7 is here.
It's a notable improvement over Grok 4.6 at the same price and speed. https://t.co/H3OTBbXyvO↗
▸ 折叠3条(转推/噪音)
转推2026-09-23 RT @AlexFinn: Claude Opus 5.5 is the best AI model I’ve ever used
I was lucky enough to have early access and I’ve been using it nonstop…
转推2026-09-23 RT @AlexFinn: Grok Bot just released for Tesla and I'm blown away
I was lucky enough to have early access. Having your car drive you aroun…
转推2026-09-22 RT @AlexFinn: Grok 4.7 just released and it's an EXCELLENT model
It was trained FOR Grok Bot
Meaning this is a fully agentic model traine…
✅ Okay the new Nomads travel profile globe is live and done :D
I had to replace a globe in legacy code that was made over a decade ago with a new one
I tried to make it look as similar as possible, so people don't realize it changed or won't have much difficulty switching
Anyway unlike the old globe, the new globe is true 3D, and it has a photorealistic mode where it fits in space and everything is as accurate as possible:
- ✨ The stars
- 🌎 Earth's position and day/night
- ☁️ Live clouds from NASA (and they move a bit)
- 🌗 The moon is in the right place too
- 🛰️ There's spaceships like @SpaceX Crew Dragon which you can add as a future trip (use "Space Orbit" as destination)
- Also 🔴 Mars is there!
I tried to recreate the recent @NASA Artemis mission to behind the Moon with the "Earthrise"
Some people think these kinds of features are for no reason, I personally love them, they make my site personal and fun and not like all the other corporate ones!!! 😊
引用@levelsio🌎 Now redesigning the https://t.co/HGCLKS5BD6 profile trips globe from scratch
I've always wanted to add space trips, so starting with the moon here which you can add as a future trip, also adding Mars etc.
Other trips also need to work here though like … ↗
引用@levelsio✅ Bought $72,450 of $AMD stock
Seems a good bet and early as @realGeorgeHotz is making AMD GPUs compatible with Nvidia code
If it works it means AMD is the first company that might actually be able to compete with Nvidia's monopoly on GPUs with its … ↗
💧 For years I was brought up in the Netherlands being told we had the "best tap water in the world"
"The Americans are silly for buying bottles of mineral water, you can just drink from the tap!"
I remember the same sentiment with my German friends
When I stayed at the VOCO hotel in The Hague a few months ago, there were no water bottles but a paper note saying "Do like the locals do! Drink from the tap"
Now it turns out Dutch tap water is and probably was full of toxic chemicals and plastics
Today actually we finally installed a Reverse Osmosis (RO) system in our home (or well the plumber did) by Waterdrop (unaffiliated, unpaid, recommended by Claude)
RO seems to the best and most pure water filtration system out there
I will let you know how the coffee tastes tomorrow with it! 😋
引用@NL_TimesMore than third of tap water exceeds limit for toxic chemicals; worst in Noord-Holland https://t.co/iDficYygGh↗
Yes! Same in Brazil
Very ugly buildings, very ugly outside, very ugly cities
But inside they put all the investment into how it looks, interior design architects, fancy designs
引用@michael_koveDenis theory is correct.
The Western Europe is predominantly
"Living-in-Public" culture.
While East - "Living-in-Private".
This is why Germans and Dutch say,
"But the parks! The schools! The public transit!!"
Meanwhile, we're here:
"Here's my castle. … ↗
Improvement to my DIY gym electrolyte drink
You all said add some honey, so I added half a teaspoon honey
Kinda salt sweet, and hard to mix honey it doesn't rly dissolve well but felt good during workout
I already eat a banana 45min before workout which also covers the potassium
引用@levelsioToday I tried to make my own electrolyte drink for the gym
It's just sparkling water, with a squeezed lemon and 1/8th teaspoon of salt! https://t.co/eqycaQW2xp↗
Please please anyone at @X make this setting sticky
It keeps disabling for next post both on web and on iOS
And X is being flooded with AI replies again https://t.co/zKPDHPzT1w
Yes I built it for myself, it started with a calorie tracker, kinda messy but nice:
https://t.co/FeWxZkJUiG
I log my food in Telegram to get calories and protein data
My workouts and sleep from WHOOP
My weight from my scale
Indoor air quality, temperature, humidity for bedroom is also tracked from Xiaomi to Home Assistant to there
Steps and other health data from Apple Health is auto exported every hour with Health Auto Export app on iOS which POST's to my server
My travels from @nomadscom's API and sauna use from @wip's API
That lets me find correlations between lots of things and see what is good for me and what is bad for me
Health is personal and everyone's body is different so this helps me make good more healthy decisions
引用@IAmPascio@heyalizaid@marclou Was wondering about @levelsio too, he has some extensive tracking going on afaik, but I assume it's just a vibe coded thing he's using.
Asked him in DMs but yet to hear back. ↗
引用@marckohlbruggeWIP is the place where makers share what they are working on.
Not just which products they are building, but literally the day-to-day tasks they complete to make it happen
From people working on their first side project, to solo founders doing millions in … ↗
My favorite airlines are low cost ones like Easyjet, Air Asia, Transavia (and Ryanair if they'd not fly with 737MAX) and premium ones like Qatar
The entire middle section is the one I try stay away from and where everything is expensive but usually sucks
Usually there you have national flag carriers like KLM, British Airways, Lufthansa or Swiss which have very mediocre service for a high price
With low cost airlines you don't pay a lot, but you get a basic functional no frills service, and because it's so high volume (Ryanair for example does the most flights out of any airline in Europe!), they have their workflow dialed in well
With premium airlines like Qatar you pay a lot, but you get a very premium consistent service
引用@redhairshanks86tbh i think ryanair is one of the best airlines in the world
they are giving poor people access to the world by reducing everything "unnecessary" to a bare minimum. if you just have a backpack and you want to see prague, you can do so for $50 or whatever … ↗
Google now makes 3 completely different laptop product lines:
- Google Pixelbook
- Google Chromebook
- Google Googlebook
I never understand why Google's branding and naming is always so confusing
I'd name it maybe Chromebook Pro?
引用@ssamatGooglebook is officially here! I’ve been so excited to share the details with the world.
Today, many of us rely heavily on laptops to get work done, but we think there is an opportunity to rethink the category to address the needs of people today. So we … ↗
Today I tried to make my own electrolyte drink for the gym
It's just sparkling water, with a squeezed lemon and 1/8th teaspoon of salt! https://t.co/eqycaQW2xp
▸ 折叠3条(转推/噪音)
转推2026-09-22 RT @MarieMartens: You raise, you build, you grow, you exit. That's the startup script.
We're trying to write a different one: @TallyForms…
转推2026-09-22 RT @lost_nomad__: - Germanic, ‘earn/deserve’ money: a moral weight from working
- American, ‘make’ money: new wealth is created, not a fixe…
Marc Lou@marclouindie399.8K粉 · 5条结束身体赞助赛程，复盘爆红与长期经营，并用3万美元API额度延续闯关活动。
Some pics from the race ✌️📸
After 2 weeks on a high-adrenaline tour, I'm back to the boring daily routine that made it all possible.
Time to make the next one happen. https://t.co/OWoE9PazTt
引用@marclou1:05:11
I finished 1st 🥇 on my age group and 5th in the male category.
I failed my ambitious goal of sub-60 but I was very happy with my race. And I have a nice goal to chase for the next few months.
All the staff at the Hyrox venue knew us! I even talked … ↗
I just sent $30,000 worth of @higgsfield_ai API credits to 30 people who completed the escape game 🎉
Check your email!
And if you've built something already, post it below.
引用@ElitzaVasilevaJust found out I won $1,000 in Higgsfield API credits from the puzzle @marclou posted a few days ago 🤯
Thank you so much, Marc!
Really excited to see what I can do and generate with it 🤩 https://t.co/p7wlWHBx4j↗
I just read my first AI book.
Two months ago, I asked ChatGPT to imagine what the world might look like in 2050.
The answer was very interesting, so I asked it to write an entire fiction book set in that future.
It generated a ~100-page story about a world where AI gets so good at predicting human behavior that it can prevent tragedies before they happen.
The story explores a simple question: if AI could make the world safer by slowly taking away our ability to make mistakes, would we let it?
Among all the futuristic visions in movies and books, this one feels the most plausible to me.
The v1 of the book was a bit slow, so I asked ChatGPT to make a v2 more dynamic.
Here’s the book: https://t.co/4XSR40bUIJ
The chat expired, so ChatGPT made a new version. We might not be reading the same version.
▸ 折叠4条(转推/噪音)
转推2026-09-22 RT @marclou: Some pics from the race ✌️📸
After 2 weeks on a high-adrenaline tour, I'm back to the boring daily routine that made it all po…
转推2026-09-21 RT @marclou: I just read my first AI book.
Two months ago, I asked ChatGPT to imagine what the world might look like in 2050.
The answer…
转推2026-09-21 RT @ElitzaVasileva: Just found out I won $1,000 in Higgsfield API credits from the puzzle @marclou posted a few days ago 🤯
Thank you so mu…
You can literally build the next Higgsfield with this, and they're still giving you up to 50% off.
引用@higgsfield_ai1 day left to lock in up to 50% OFF Higgsfield API.
Build your own AI app with our product endpoints, including:
• Higgsfield Genjutsu
• Cinema Studio 4.0
• Higgsfield Soul
Plus all frontier video and image models through the same API. … ↗
I’m not sure there’s a point in tweeting this.
But 10 years ago I popularized design subscriptions.
It fundamentally changed the industry.
And for the last 5 years, I’ve been sitting on an idea that I think is significantly more disruptive.
In my head, I’ve always referred to it as the end game.
I’ve never talked about it publicly.
Partly because I’m worried about what it could do to the industry.
Design subscriptions weren’t exactly welcomed with open arms.
This definitely won’t be either.
But I’ve reached the point where I need to see it through.
It could completely flop.
Or it could change everything.
I guess we’ll find out.
Saying Corgi is best known for its insurance products is like saying Hooters is best known for its burgers.
引用@nico_laquaCancel culture is alive and well, but, at Corgi, we have thick skin and believe strongly in freedom of expression.
For those who don’t know us, Corgi is an increasingly diversified financial institution, best known for our insurance products and generating … ↗
▸ 折叠1条(转推/噪音)
转推2026-09-21 RT @higgsfield: What if creativity was the only limit?
Higgsfield Genjutsu lets your team scale hybrid production.
Shoot with your crew,…
was hopeful the chrome web store gods would look down favorably upon us all and release this within 24 hours but alas, it wasn't meant to be.
tomorrow? 🤞 https://t.co/rBQWkqpyWg
new update in https://t.co/s6bZ2W8MZz detects if chat is making reference to a potentially life-threatening situation and directs you to call 911.
there's a real psychological block in emergency situations where folks will talk themselves out of doing the obvious. https://t.co/ju19g4A4D3
this is skill number 25 for the initial commit club!
引用@ShpigfordJust dropped a new Initial Commit skill: /design
https://t.co/UXh8jyrMnl
A skill that gets consistently good design out of a coding agent, whatever you are building and however much you want to think about it.
Before writing code, the agent looks at how … ↗
🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
MAY THE AI GODS HAVE MERCY ON OUR SOULS
🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
haven't even used opus 5.5 yet but fable 5.1 is now the dumbest thing i've ever used in my whole entire life and i freaking hate it and i might as well be artisanally coding like a freaking peasant. a squirrel could code more betterer than fable. ugh.
okay so...when would you choose Fable at this point?
i wish anthropic/openai/etc would give practical use cases for when to use one model over the other since benchmarks are functionally useless for real-world applications.
引用@claudeaiOpus 5.5 is a major step up from Opus 5, leading on agentic coding, computer use, and knowledge work. https://t.co/MYI9JDAo1x↗
🚨 BREAKING: Opus 5.5 is THE release we've all been waiting for that will finally cause an extinction!
I've been using it for 6 months and have already gone extinct 9 times!
THIS CHANGES EVERYTHING!
Reply "EXTINCTIONDADDY" for my prompt to avoid extinction
引用@claudeaiIntroducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f↗
Just dropped a new Initial Commit skill: /design
https://t.co/UXh8jyrMnl
A skill that gets consistently good design out of a coding agent, whatever you are building and however much you want to think about it.
Before writing code, the agent looks at how real products handle the same screen on Mobbin, applies a set of opinionated design rules, and designs the screen in Paper so there is something to judge before there is something to ship.
Landing page or settings screen, empty state or dashboard, you get the same considered result without doing the considering yourself.
Works best with Mobbin, Taste, and Paper, but also works w/o them or with the various alternatives available.
Lots of cool experimental Jev (@typesafeai) stuff getting posted lately, but what about using it in an existing product?
Here are dozens of ways I'm using it now in two apps (https://t.co/vKHSHPzmMl and https://t.co/JhKfmhCVib)
Granite (document vault)
Ingest pipeline
- Second-opinion on Gemini's document classification, flags low-confidence ones for review
- Scores PDF text-layer quality and routes bad ones to OCR
- Verifies each extracted field against the page text
- Detects what a document asks you to do (pay, sign, renew, respond) and how urgent it is
- Judges whether two near-duplicate documents are the same or a revised version
Document page and library
- "Needs review" banner with one-tap confirm of the document type
- "Not confirmed" marker on extracted values Jev couldn't verify
- Action chip next to the document type
Collections
- Plain-English filing rules ("anything to do with my taxes") that auto-file matching documents
Life View and email digests
- "Needs your attention" block listing documents that require action
Ask
- Routes each question to the right tool path instead of a regex
- Checks the final answer is supported by the cited passages and hedges if not
Entity graph
- Tiebreaker on whether two fuzzy-matched names are the same entity
Evernote import
- Triages each note into keep, reference, scratch, or clutter before import
KeptWell (family medical binder)
Trust
- Verify every extracted lab value, dose, diagnosis, and provider against the source page
- Flag values it cannot confirm with a quiet "check this" marker
- Catch diagnoses stated more precisely than the page says
- Second-opinion the document type after extraction
- Gate prompt PRs with a cheap eval-corpus canary
Attention
- Tag new documents: new diagnosis, out-of-range result, med change, follow-up, act-within-7-days, admin-only
- Order the dashboard feed by importance, not recency
- Decide push-now versus digest per notification
- Pick push wording from the PHI-free string set
- Decide which lab trends are worth an Insight before calling Opus
Chat
- Detect emergency or distress before the model streams a token
- Route docs-only questions away from paid web search
- Rerank retrieved chunks against the question
- Pick between two contradicting family facts
- Filter PHI-audit false positives ("Ray" in "x-ray")
- Check the answer is grounded in the cited record
Binder
- Tag every document by body system, specialty, and care phase for filters
- Flag near-duplicate uploads for review
- Break ties on whether a PDF text layer is usable
Recordings and journal
- Label each recording utterance and build an action-item checklist
- Score journal entries on a symptom rubric for trend charts
- Flag entries that look like a medication side effect
- Decide whether an undated entry describes a specific past day
- Replace the async journal tag job with one sync call
Terminology and imports
- Auto-pick clinical codes above a confidence bar; queue the rest
- Replace the Sonnet pick in disambiguation
- Decide which FHIR observations are real lab results
- Map vital types the LOINC table drops
- Merge brand and generic med names ("Lipitor" and atorvastatin)
- Pick the right NPI when the registry returns several
- Classify severity for manually entered diagnoses
- Catch allergy denials the regexes miss
Cost gates
- Skip the highlight call when nothing is worth highlighting
- Skip reprocessing documents a prompt change would not affect
- Route extraction to batch or sync by urgency
- Tell a bulk import from a runaway loop at the spend cap
- Flag uploads containing instructions aimed at an AI
Guards
- Veto preventive reminders the record contradicts
- Suppress marketing emails during a hard week
- Rank appointment prep context by relevance
- Auto-resolve visit questions the visit log answers
- Mark share comments that are waiting on a reply
Jev真实用例清单：分类、核验、路由、重排、通知、成本门控和安全守卫。
▸ 折叠4条(转推/噪音)
转推2026-09-22 RT @alexalbert__: I've been on a Blender kick with Opus 5.5. Its better 3D modeling and vision mean you can build an entire world from a si…
转推2026-09-22 RT @every: @stov3r@Shpigford Reach for Opus 5.5 if:
- You build things you can look at e.g. Interfaces, prototypes, 3D scenes, games, tool…
转推2026-09-22 RT @DanielleFong: imagine if Santa Claus and Mrs Claus went to war with each other by delivering escalating presents to people throughout t…
转推2026-09-22 RT @yuxuan_o_o: introducing https://t.co/vOe0sj509k 🤍
I carved 32 great women into a wall, you can learn from them and talk to them.
grow…
Danny Postma@dannypostmaindie183.7K粉 · 4条用9个自建技能改落地页并自报转化率+34%，同时遭遇疑似负面SEO。
A few weeks ago I rebuilt my landing page with AI agents.
No one-shot. I created 9 skills instead from all my years of knowledge to speed up my time.
Test just finished w/ 34% higher conversion rate 🚀
Wondering if I should turn these skills into a course for your AI agents 🤔
ai headshot industry is so toxic, got a competitor who keeps buying spammy backlinks to our site to ruin our domain rank
be careful out there! https://t.co/fX6OPOaohy
I used to be a hard-core fan of Claude since forever, never switch to another API or LLM since Opus 4.8 came out
Until this August, whatever they did, I haven't been screaming and annoyed at it's work like this before and moved completely over to OpenAI + Grok
I thought I was going crazy, but apparently they pulled the same shit they did back earlier this year with regression
引用@LonAfter Anthropic made Fable 5 permanently available in subscription plans, I noticed a large drop in performance. The model felt dumber, and I couldn't explain why.
Measured five different ways, August delivered dramatically fewer thinking tokens than July. … ↗
I did the exit part
it's great, it's also over in a week and then you wake up and the thing you loved building belongs to someone else
Tally is playing a better game, $6m ARR with 10 people and no VC means total freedom
congrats Marie 👏
引用@MarieMartensYou raise, you build, you grow, you exit. That's the startup script.
We're trying to write a different one: @TallyForms just reached $6M ARR, bootstrapped and purely product-led, with a team of 10 and over 2.5 million users.
Six years ago I wouldn't have … ↗
just closed an amazing and insightful session with @robj3d3
he is the king of storytelling and generating attention through content
the guy articulates and presents complex things really well
my biggest takeaway: the best growth hack for any content is to make people feel some emotion
people had so many questions, and he had so much to share, that we could only do 1 roast 😅
he even shared his setup for recording and editing videos
incredible session 🙌
more of these coming in the tmaker Founders Room
don't believe anything you hear on TikTok... especially about your health
we analyzed over 3,000 viral health videos on TikTok - that's 25.9B views combined
but 2 out of 3 health claims in them are backed by nothing, or by one person's story 😑 https://t.co/GHydSwtQT4
Sam, Dario and Elon all agreed to slow down AI development
but did they? 👀 https://t.co/IBLxFeeugv
引用@DarioAmodeiWe Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, … ↗
it's rare to nail virality at this level
in the last 7 days, @robj3d3 did over 2 million impressions on X
tomorrow he's doing an AMA in the tmaker Founders Room and we're roasting X accounts
👋 drop your @ below, we'll pick some 👇
Rob and I will tell you exactly why your profile isn't working, out loud
引用@robj3d3I turned Jev into a slop detector for your own posts.
Naval got 79%. I got 44%.
Free, no signup, let's see your score ↓ https://t.co/Hpd1BJpjKP↗
转推2026-09-21 RT @NCoutureau: @revid_ai makes it easy to publish your videos to X! Showcase product reveals, tutorials, or moments worth sharing. Join th…
Jon Yongfook@yongfookindie172.1K粉 · 9条讨论AI时代SaaS转向、替换开发者工具，以及功能护城河失效后的非AI对冲。
I’m going to sit at a cafe where loads of tech people hang out and quietly read a 2015 copy of Test Driven Development and see if I get any funny looks. https://t.co/jRREb66KPp
At some point Claude switched from doing things deterministically (e.g. knowing the position of an element based on its css) to wanting to eyeball everything (take screenshot, make some judgment).
I disabled claude-in-chrome but it's still finding a headless browser to use.
Definitely living in a simulation.
I was thinking of a name for a new project last night. Settled on one I like, and then slept on it.
Today I find a Bannerbear competitor with a similar name. Never seen it before.
Weird.
I never hit any usage limits ever before, until I started using claude-in-chrome. Turning it off. It seems way too eager to fire up a browser even for the most mundane things - bro I don't need you to browse to localhost and screenshot the fix, I have the tab open right here.
Spent the morning replacing a SaaS vendor. Infra-related. They had raised $2 million, one of the smaller YC companies from back in the day.
It wasn't even about saving money, it's just "neater" to control more of the stack, if it's easy to migrate.
Don't build for devs!
At least once a week on X, I go down a rabbit hole exploring a new space because someone here is talking about massive growth in the space, only to find they sell a shovel for that space.
I feel like build in public circa 2020 was more genuine.
Thinking about building a “hedge” SaaS against AI. I’ll continue building in the AI space but at the same time build a super boring app that has nothing to do with AI, has non-tech users, as a hedge against AI eating everything at the cutting edge.
一边做AI产品、一边做非技术用户的无聊SaaS，作为前沿风险对冲。
Simon Høiberg@SimonHoibergindie163.2K粉 · 3条继续押注自托管与开源模型，公开一套去Cloudflare的反向隧道思路。
When you move to self-hosting your products, security automatically becomes the number 1 concern.
I used to have CloudFront distributions sitting in front of every public endpoint.
Another popular solution is using CloudFlare tunnels and Tailscale. The only issue is - it's still "cloudy". And personally, I wanted to get rid of that.
Fortunately, I found a surprisingly simple setup that can do roughly the same as CloudFlare tunnels - but fully self-hosted.
Let me show you 👇
Here's the best way to use OpenAI's/Anthropic's latest frontier models.
- Wait for China to distill them.
- Use them to build tight, custom harnesses, safeguards, evals, and tooling for Qwen/DeepSeek.
Then use Qwen/DeepSeek for your work. Once a new model is out, repeat the two steps above.
引用@SimonHoibergSince the Codex reset yesterday, I already depleted the weekly limits of one of my Codex accounts.
This account exlusively uses GPT-5.6 Sol. I'm tracking all token consumption through my OpenClaw setup, so it's very easy to compare to previous … ↗
At this point, I don't even want to try the latest models anymore.
If we're being honest, there isn't anything these new models can do that older models (and now open weight models) can't inherently do.
They do things in fewer tries and with less friction, fair. But nothing we can't achieve with older models.
I'd much rather spend my time building custom harness, proper evals, and guardrails for open weight models, and ultimately end up with a similar result: get work done.
At least with Qwen, I know what I'm getting.
引用@synthwavedd🚨 SCOOP: OpenAI are in the final stages of preparations for the launch of GPT-6 Sol and Luna, and the Terra tier is being discontinued. Anthropic are also working on a version bump with Fable, Opus, and Sonnet 5.5. Opus 5.5 is shipping imminently at … ↗
Tony Dinh@tdinh_meindie202.1K粉 · 5条Steam发行商账号获批、TypingMind接入Opus 5.5，也在试新的AI工作流。
I think we should start using AI to make the world a better place.
Build things that weren’t possible before.
And I’m not talking about building a SaaS that “changes the world”, but actually doing something with real world impact.
Okay this is fucking amazing. Going to connect everything to grok now so I can speak to it from my Tesla
引用@Tesla.@Grok in your Tesla can now do meaningful work for you
With Connectors, you can manage your inbox, clean up your calendar, or talk through existing files/chat/tasks – all hands-free https://t.co/W1LuybQh0P↗
These model names and especially the comparisons are becoming really confusing 😂😂
引用@claudeaiIntroducing Claude Opus 5.5, the first model in our new Claude 5.5 family.
It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f↗
引用@isjackbackmakes $250k in one month with outbid
retires
opens a tavern in the mediterranean
refuses to elaborate
not bad @jonathan_wilkehttps://t.co/ZPTLIJCAXJ↗
This is likely going to be the next product on the google graveyard
引用@GoogleIntroducing Googlebook, a new category of laptop, available for preorder today.
💪 Crafted with 2.8K OLED touchscreen displays, 14 hour battery life and all the performance needed to power your ideas
📳 Engineered to sync effortlessly with your @Android phone, … ↗
▸ 折叠1条(转推/噪音)
转推2026-09-21 RT @verbove: we bought outbid's #1 spot today 🫡
https://t.co/SAqnhhLUTr puts 3,995 makers from the X intro trend on one map.
type your @…
Marc Köhlbrugge@marckohlbruggeindie88.6K粉 · 2条把WIP的真实todo做成工具采用、放弃和趋势信号。
WIP is the place where makers share what they are working on.
Not just which products they are building, but literally the day-to-day tasks they complete to make it happen
From people working on their first side project, to solo founders doing millions in ARR like @levelsio. Even YC startups like @getcontextdev!
But what tools are people using to build these businesses?
I'm not interested in SEO slop like "Top 10 payment providers in 2026" or an upvote popularity contest.
I want to know what people are ACTUALLY using to get the job done.
So starting today, every completed todo is analyzed to see what tools are mentioned, whether people are evaluating, adopting, using, or leaving them. The sentiment around the tools, what other tools they are often combined with, etc.
It also shows TRENDING tools. No surprise here, Jev is #1 right now.
But what's cool is that I didn't manually add Jev as a tool. Nor did anyone else. It just surfaced to the top automatically because it's what people are posting about. And an enrichment agent then automatically went ahead and fetched the icon and description from @typesafeai's website.
My goal is to help makers figure out what tools to use and help each other make the most of them. While also providing tool creators with useful insights in what people like about them, but also where users get frustrated or even completely switch to an alternative.
Check it out here:
https://t.co/lcES1RSJqK
Terra: gone.
Kinda interesting to see OpenAI iterate here. Also very interesting to see which kinds of models the market actually seems to want (and uses).
引用@OpenAIPlease welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.
GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.
We’ve also made caching and inference … ↗
Ooooh, Claude released Opus 5.5 AND gives us free do-your-own resets now? Juicy.
Time to fix up your test suite coverage and run a few security audits. https://t.co/Xfjn8xxobC
what if your AI agent could create launch videos?
motion animation + product assets
done locally with your Codex / Claude Code subscription https://t.co/5JTKnYzCyF
> build ai calorie tracker
> don't localize app
> charge $7.99 weekly subscription with 3-day free trial
> get 0 customers
> complain that ASO is dead
many such cases
Most expense automation breaks the moment a receipt doesn't fit the rule.
That's a real problem when you're reviewing 10,000+ expenses a month across different countries, languages and policies.
Akai takes a different approach.
It watches how an expert handles the workflow and learns the logic behind each step:
> check the employee's jurisdiction first
> split submissions containing multiple receipts
> apply the relevant local expense rules
> escalate cases that need human judgment
When the team corrects an edge case, those decisions become shared memory for future runs.
The interesting part isn't that AI can automate boring tasks.
It's that you can teach it how your business makes the judgement calls.
引用@BouazizalexEXCITED TO LAUNCH: Akai (https://t.co/mMBMmX4NCr)
Deel added >$140M ARR in 90 days without increasing headcount by automating~600 Full Time Employees' equivalent in work with Akai.
Akai was an internal tool to automate our painfully repetitive operations in … ↗
Anthropic researcher just open-sourced Halo
takes the Hugging Face model you already have and scales training across GPUs
pretty neat, starred the repo https://t.co/a2bGSmSyma
引用@whitecircleIntroducing Halo, the best framework for post-training of open-source models.
Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format.
Star us on GitHub: https://t.co/3mAiUljdrN … ↗
saw a vending machine at airport yesterday for:
Licorice .com
Caramels .com
Taffy .com
Even if the machine generates low sales, the branding & adverting would make up for it. Glance once & folks remember the name. https://t.co/quceN7uqGm
Andrea Bosoni@theandrebosoindie64.1K粉 · 1条主张首批客户靠真诚的一对一对话，而不是批量外联模板。
The easiest way to land your first 10 customers as a new indie founder is to find your ideal prospect on social media and have an honest conversation with them. Period.
Where most founders mess up is turning it into a full outreach campaign with a copy/paste template.
That's not a conversation. That's a sales pitch. We all get them. We all ignore them.
And conversations stack. Each one teaches you their needs, their objections and what actually matters so the next one is more likely to convert. Mass outreach just keeps getting ignored.
首批客户靠真实对话，目标是每次更新需求、异议和表达，而非批量发送。
Thomas Sanlis 🥐@T_Zahilindie24.3K粉 · 2条继续公布Residency成员，本期产品信号较少。
The crew is complete: meet our resident #10: @piotrkulpinski 🥳
I've been following Piotr for so many years, I don't even remember when it was 😱
I've always admired the care he puts into all his products and how he sticks with them until they work (SEO ftw)!
With him, @karakhanyanS, and @venelinkochev participating in the Residency, the directory crew will be complete 👀
I'll post a recap of all our residents tomorrow and will start to share more about what's planned for our Residency #2 soon 🔥
引用@T_ZahilMeet our resident #9: @eMarboeuf 🥳!!
Emmanuel has a an incredible story
He spent 10 years in SF as a cofounder CTO and scaled his company to $9M+ ARR
Now he's back to France (in Nantes actually 🫶🏻), and he's building BWorlds[.co] with his cofounder … ↗
Disappointed that the slop enthusiast has a low slop score
Clearly rigged https://t.co/CZPPQogaNY
引用@robj3d3I turned Jev into a slop detector for your own posts.
Naval got 79%. I got 44%.
Free, no signup, let's see your score ↓ https://t.co/Hpd1BJpjKP↗
▸ 折叠1条(转推/噪音)
转推2026-09-23 RT @_MaxBlade: I gave GPT 6 SOL and OPUS 5.5 the same prompt.
It was not even close.
Opus 5.5 is THE BEST model we have ever seen.
li…
Dan Rowden@drindie110.1K粉 · 1条发布Subsail Memberships，为杂志增加按时间订阅与持续权益。
Today I launched Subsail Memberships!
Memberships are "regular" time-based subscriptions, perfect for magazines who want regular income and can offer ongoing perks like access to digital content, a community, a podcast, merch etc. The magazine can be included or not!
Perks is something issue-based subscriptions struggle with; memberships fills in a nicely-sized gap in Subsail's offering.
Excited to get this out and get the first few publishers creating memberships soon!