Astra for Law: Frontier intelligence built for your practice.
A new offering powered by GPT-6 Astra with tools, settings, and context to support the expertise and judgment of lawyers and legal technology firms. https://t.co/qxVRXkvp4o
Just like breaking crypto in the age of cloud is more about cost than time, this will lead to legal attacks based on the same principle. The biggest wallet wins.
Astra for Law — tools and skills to radically accelerate law firms.
Includes data privacy as a core feature, and 26 partner-built plugins and 47 community plugins for legal work in ChatGPT.
引用@OpenAIAstra for Law: Frontier intelligence built for your practice.
A new offering powered by GPT-6 Astra with tools, settings, and context to support the expertise and judgment of lawyers and legal technology firms. https://t.co/qxVRXkvp4o↗
Astra for Law is here!!
GPT-6 Astra + legal search across 230M+ URLs, with 26 plugins for the tools lawyers already use.
Firms are already building tools to review contracts, investigate deals and draft IPO filings.
So much room to build!!
引用@OpenAIAstra for Law: Frontier intelligence built for your practice.
A new offering powered by GPT-6 Astra with tools, settings, and context to support the expertise and judgment of lawyers and legal technology firms. https://t.co/qxVRXkvp4o↗
引用@levelsio@mick__net@marckohlbrugge@ycombinator I mean it's in your favor to agree as your product is a harness, but I don't see any economic reason why OpenAI or Anthropic won't just add specific industry harnesses for example to their product
Like ChatGPT for … ↗
Lab: Introducing Astra!
Startup: our industry harness is our moat
Lab: Introducing Astra Medical
Startup: our geographic niche is our moat
Lab: Astra Medical in Urdu!
Startup: we can do it cheaper
Lab: Astra Medical in Urdu is now Free!
Startup: feck
Jev + Kimi K3 for fraud detection!
TLDR: Jev classified 100 emails in 1.42 seconds, then I routed the uncertain cases to Kimi K3. The full pipeline got 96/100 correct for only ~$0.07.
Video is not sped up, check out the live run!
Here was my process:
I gave Jev 100 emails to classify (a mix of 50 legit & 50 fraudelent emails). It classified all of them in 1.42 seconds.
An underrated feature about Jev is it will give you the confidence score for a classification, so I routed any prediction under 95% confidence to Kimi K3 to be fully sure.
31 emails fell below that threshold. After routing those to Kimi K3, the combined pipeline reached 96% accuracy.
The full run took 16 seconds & ~$0.07 in inference costs:
- $0.068 from Kimi K3 on @togethercompute
- $0.003 (1/3 of a cent) from Jev on @typesafeai.
I think this is a really interesting pattern: use a fast specialized model like Jev for the narrow task, then route the uncertain cases to a larger LLM.
I feel like this kind of approach could be a game changer for use cases like fraud or anything realtime. You can use the speed & low cost of Jev while having a larger LLM as a fallback to ensure high accuracy.
作者自报的单次演示；没有类别分布、误报代价或重复运行数据，适合看架构，不宜当通用基准。
Tony Dinh则做了一个实时监听YouTube音频并跳过赞助段的开源原型，自报每条视频约0.005美元。这里真正新的不是“识别广告”，而是低延迟让分类可以嵌进播放过程。
Just trying out Jev, I made a Chrome extension that:
- Listens to your YouTube audio (optional)
- Detects if it gets to a sponsor segment
- Skips it ➡️➡️➡️
- All in real-time while costing ~$0.005 per video
Prototype project, BYOK, open-source:
https://t.co/g4sPXR5kMLhttps://t.co/nH6c1LhU3C
This is actually insane. This uses @typesafeai Jev model, as a plugin in Claude to review all the un-nesseasary tool calls, and it takes 1s to run!
Like, literally, 1 second to take my Claude session from nearly 1M to ... 86K tokens! 😮
Ask your claude to install it and be amazed
Use this prompt
```
Install, and configure :
https://t.co/yD6PncSCtH
```
引用@tamarajtranfound the perfect use case for @typesafeai Jev:
instant compaction
in 2026, why is compaction still a summarization prompt?
Jev can make it instant by scoring every tool call and dropping what’s irrelevant https://t.co/h6NzKzNpCg↗
Tony Dinh给出的边界反而最有用：分类模型本身并不新，Jev的差异在通用、快和便宜；如果演示没有真正利用延迟与成本，它未必比既有模型更有价值。
Very cool.
I think models like Jev are not new, but what makes it different is that it is a general classifier and it is FAST and CHEAP.
Any Jev demo that doesn't use the fast/cheap aspect could be done with an existing model we've had for years.
引用@ekzhang1Inspired by @typesafeai , here is a Jev-compatible public API to play with
It runs a comparable open model (Qwen3.6-35B-A3B), and just uses SGLang radix cache to preserve the prefill reuse / really fast parallel systemone generation - 64 tasks in … ↗
A few more years and our agents won’t need us at all 🤷
引用@claudeaiProjects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop.
In beta today for select Pro and Max users in cloud sessions; coming to all … ↗
Claude Code团队的Thariq Shihipar给出更具体的架构：每个项目有一个管理记忆的Agent，再为任务启动子Agent，还能被要求主动或按计划运行。产品单位从“一次会话”变成“一个持续存在的项目”。
Projects brings the architecture of Claude Tag to Claude Code.
It has one agent per project that manages memory and spins off subagents for tasks.
You can ask it to be proactive, to do things on a schedule, etc.
It feels a lot nicer than a bunch of sessions, try it out!
引用@claudeaiProjects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop.
In beta today for select Pro and Max users in cloud sessions; coming to all … ↗
Right after Cursor launched Projects…
Why are all these labs building the same thing? So uninspired
引用@ClaudeDevsToday we're rolling out Projects in Claude Code on desktop and web.
A project is one conversation with Claude. It splits the work into threads itself, runs them as parallel cloud sessions, passes context between them, and keeps going when you leave.
In beta … ↗
Just had a long-term multi-year customer cancel their plan. They are also a SaaS company.
I pretty much just shrugged and figured, "makes sense, they probably vibe coded a replacement".
Don't target developers!
The argument was always “no way will they waste time vibing their own version” but when their own devs have nothing to do because Claude is already set up to fix, merge, deploy… folks are gonna start looking for other things to do to justify their roles.
引用@chrisdengsoJust had a enterprise customer churn. They have vibe coded their own version to save 1000 bucks a month. I get it. It’s a lot of money for them in these vibey times, and now they can fully utilize their developers to maintain another codebase on top of their … ↗
Areas where I don't think devs want to vibe-code their own stuff, and will continue to pay:
- hosting
- databases
- email
Anything else? it's a short list.
the harder the problems, the more you get paid
if it can be shipped in 2 hours by Codex, you won't get paid much
go out, talk with real people
引用@nico_jeannenDoing "tiny SaaS" and Mobile apps rn is like trying to do Print on Demand / Dropshipping in 2019
The barrier to entry have been destroyed, anyone with 2 neurons can pickup Codex to do the same now
The indie hacker playbook doesn't works anymore, you can't … ↗
The barrier to building something has always been a positive thing.
"Should I build this?" is more important than "Can I build this?"
Understanding, domain knowledge and craft create value.
When it took months to build software, you had time.
Time to validate.
Time to realise your idea was stupid.
Now you can build an idea in an afternoon but you still don't know if anyone wants it.
You don't understand the problem space.
You don't understand the market.
You still haven't built the expertise that makes something valuable.
Slapping something together in a few prompts usually does not create value, yet the world is now full of slop products produced like this.
AI can write the code but it does not understand the problems it is solving, or how humans feel interacting with products.
It is just a text generator. It has no understanding or empathy.
You have to bring those things.
We are more empowered than ever, yet we're in peak enshitification.
Products with no soul, solving problems their creators don't understand.
Built by people who don't want to think for themselves.
引用@levelsio✨ New mini feature requested by @shmichael_eth on my ideas and bugs board
📍 Sort by distance to a pin on the map
Sometimes you need to find a hotel near a specific place
Now on Hotelist you can tap Sort by -> Distance to point -> Select point on map
Then … ↗
✈️ https://t.co/1z6UN2d0nh now supports imports from Flighty
Flighty app → Settings → scroll to Manage → Account Data → Export Your Flights → Download the CSV → Open https://t.co/hVSVlYtnsg and import it there
All your trips will show up on your profile imemdiately! https://t.co/szzeWS68OM
引用@levelsioSince I added this ranking I seem to have activated lots of people and they're all uploading their travel history to https://t.co/HGCLKS5BD6 with the MCP endpoint at /mcp
Here is @stevelacey's AI scouring his old emails for flight tickets and bookings to … ↗
Thinking about creating a /pr skill. You invoke it to create a PR that:
- Makes human review as fast and as simple as possible
- Takes the best parts of /show-me by @dexhorthy
- Uses your domain language (from /grill-with-docs)
- Shows evidence that the change works (images, test output etc)
- Assesses merge risk (is it a one-way door, or a two-way door?)
IMO every model/harness I've seen creates garbage PR bodies - needs a skill here.
WDYT?
The X algo has gotten MAJOR updates almost daily for the last month, and nobody is talking about it
The way posts on X are ranked have completely changed
I've read through thousands of lines of code. Here is almost every update to the open source algorithm from the last month: … 全文↗
All restaurants with 4.5+ stars on Google Maps ask for a review.
“We offer you free coffee 🤗 Now give us 5-star review.”
I do not trust reviews anymore, but there’s no alternative for now. I’d ditch Google Maps as soon as someone builds a trustworthy system.
And I would pay for it.
I just added a filter by @handle for Stalkr.
It shows all the mentions from a specific account.
And it works with analytics, so you know how a person talks about your brand online.
+ Available via API/MCP too https://t.co/liJcyklYKN
2 years ago, I started an SEO tool
the main narrative was: "SEO is dead", "ChatGPT will kill Google", "Ahrefs and Semrush own the market"
built it anyway, with a Ukrainian co-maker I had never met in person
today Outrank does $300k+ MRR
meanwhile, in the "worst startup idea" category:
- Semrush … 全文↗
引用@JunaidAckroydThe two worst startup ideas on the planet are:
1) AI-native CRM
2) SEO/GEO tool/service ↗
We've normalized a ridiculous way of buying software.
You can pay for the same SaaS product every month for 10 years and own exactly 0% of it.
I say this as someone who has sold subscriptions for years. But I firmly believe there's a better way.
FounderStack is my attempt at a serious … 全文↗
Adam Lyttle承认追逐最新模型和快速成品的FOMO侵入生活，提醒“几分钟能开始”常常仍会变成几天提示与返工。
Found the issue: I’ve been letting vibe coding and this ai hype cycle eat into every part of my life
I lost balance.
I’m not addicted to the tokens or anything. But I am falling for the FOMO of building with the latest and greatest models. Having an idea that I can turn into a product within a … 全文↗
引用@adamlyttleappsI think it’s because I’m no longer in front of the computer
There’s no “clear” signal that I’m working. I work from home and I can be sitting on the sofa or out back.
Which looks like an invitation to look after our baby or care for the kids. But I’m … ↗
Get more visibility into your usage.
See how tasks, subagents, and individual chats contribute to your Codex usage, so you can make more informed choices about your workflow.
引用@joey_trasattiRan out of usage again? Check out the analytics tab to the Codex app to help you see where your usage is going. Update to the latest version to try it! https://t.co/XzsASuaouj↗
One of the most underrated features in the ChatGPT desktop app: Appshots.
Appshots take the context on your screen and bring it into the desktop app, allowing you to share instead of describe.
You’ll be surprised at how useful it is.
To get started, press both Command keys on macOS, or both alt … 全文↗
Love this for the folks with 16gb graphics cards - 3.8 27b has been incredible but not quite runnable on anything less than 32gb - will try loading this up on my 16gb intel b50 and see how it goes - not sure these quants can be accelerated by the XPU cores yet but maybe in time!
Curious if or when we'll see the Qwen4 series, one thing I love with Qwen is it comes a much larger range of sizes so I can experiment which extremely small llms.
Let AI agents use your real, logged-in browser without interrupting your work. CLI + extension for browser automation across any shell-capable AI agent.
I'm surprised this page doesn't mention the automotive thermostat which uses the same principle to control the flow of coolant between the engine block and radiator in an ICE vehicle.
I’ve had a pet theory for a long time that society loses it’s collective mind every 10ish years or so, usually starting with a good reason, but the outcome is mass stupidity followed by mass forgetting leading to a new cycle. War on terror, GFC, crypto, covid, now AI. Although the cycle does seem to be shortening. Also we’re just so damn stupid to learn from history, or even just people 10 years older than us, that we repeat the cycle. And there I guess is where saying a new sucker is born every …
> US President Donald Trump earlier called the idea "laughable" and threatened "very serious tariffs" on Europe if he considered it a "hostile act". So if Canada tries to make friends with anyone other than the US, it's a hostile act? Makes it all the more obvious that USA is Canada's abusive partner, that they should have broken up with a long time ago.
> With that interpretation, the issue becomes slightly different: is it more important that the collective understanding of the mathematical community should be as advanced as possible or that there should be answers to as many problems as possible? Or are those two aims valuable in different ways, so that there is no point in declaring one of them more important? Or are they so inextricably linked that it makes no sense to argue that one is more important than the other? And when we say “import …
tl;dr; version: While there are good use cases - > The United States is the world leader, with roughly ninety per cent of global capacity. [...] there are more self-storage facilities in the U.S. than there are Starbucks, McDonald’s, Walmart, Home Depot, Domino’s, Dunkin’, and Costco locations combined. Annual revenues are estimated to be more than forty billion dollars. > I asked a Welsh facility manager I met in Las Vegas what his customers keep in their units, and he said, “It’s . . . it’s . …
Astra for Law: Frontier intelligence built for your practice.
A new offering powered by GPT-6 Astra with tools, settings, and context to support the expertise and judgment of lawyers and legal technology firms. https://t.co/qxVRXkvp4o
We're sharing our new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI.
The framework sets criteria and timelines for public disclosure, including when we haven’t yet fully explained or mitigated the behavior. More complex cases may require longer investigation or coordination with third parties.
We’ll prioritize examples that reveal new misalignment mechanisms, meaningful changes in known behavior, or findings that challenge assumptions about safety or mitigation.
Alongside the framework, we’re publishing six reports on instances of misaligned behavior we’ve observed during the training or evaluation of our models in the last six months.
This is a starting point. We’ll refine the process through experience and public feedback, and share more reports on an ongoing basis.
https://t.co/ismCCkeE0L
Sam Altman@samaAI6.2M粉 · 1条转发临床版Astra，并预告原定本周的大发布推迟到下周。
the main thing i was excited about launching this week will be next week instead, but imo worth the wait!
引用@samabig 🚢 this week
and then for devday
🚢🚢🚢🚢🚢🚢 ↗
▸ 折叠1条(转推/噪音)
转推2026-09-17 RT @thekaransinghal: Verified U.S. clinicians can get free GPT-6 Astra (Pro) access today via ChatGPT for Clinicians.
Sign up: https://t.c…
Peter Gostev@petergostevAI26.9K粉 · 14条让Astra与Fable玩《过山车大亨2》，公开比较完成度、token与API成本。
The API cost would have been $68.18 for Astra vs $297.33 for Fable
引用@petergostevI got GPT-6-Astra and Fable-5.1 play RollerCoaster Tycoon 2. The goal was to reach 1000 visitors and make the most money. Astra reached the goal in 3 hours, and Fable never got there (it hit my CC limit).
Astra's value was $65k, Fable's $18k, Fable spent 5x … ↗
I got GPT-6-Astra and Fable-5.1 play RollerCoaster Tycoon 2. The goal was to reach 1000 visitors and make the most money. Astra reached the goal in 3 hours, and Fable never got there (it hit my CC limit).
Astra's value was $65k, Fable's $18k, Fable spent 5x the tokens too. https://t.co/sAe1KkFAJR
Now we have Fable 5.1 playing RolleCoaster Tycoon 2, previously we had GPT-6-Astra reach the goal of 1000 visitors after ~3 hours. Let's see how Fable will do.
Medium setting, all done via their native harnesses, nothing fancy.
https://t.co/HhwK0ASxzN
引用@OpenAIWe're sharing our new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI.
The framework sets criteria and timelines for public disclosure, including when we haven’t yet fully explained or mitigated the behavior. … ↗
I got Astra to play Roller Coaster Tycoon 2 with the goal of reaching 1000 visitors. It took 3 hours to finally get there. It was struggling at times, blocking visitors, placing rides or food stalls without access to them. But it was surprisingly good at recovering. Fable next? https://t.co/N8I2K3ZYU3
We are so early. Not even a third of organisations are in the 'at least scaling' stage of agents/coding assistants.
There is an insane power law of the level of AI usage right now. If you use your $200/month sub, you are probably in the 0.1% of the usage across real orgs https://t.co/J9OYgMFr36
Astra
✅ Fast
✅ Frontier
✅ Efficient
✅ For everyone
引用@j_dekoninckWe are releasing the latest version of BrokenArXiv and ArXivMath! These benchmarks now focus on conjectures that were refuted in the last month on ArXiv, and models are executed within a harness instead of directly via API.
Performance remains impressive, … ↗
This is actually insane. This uses @typesafeai Jev model, as a plugin in Claude to review all the un-nesseasary tool calls, and it takes 1s to run!
Like, literally, 1 second to take my Claude session from nearly 1M to ... 86K tokens! 😮
Ask your claude to install it and be amazed
Use this prompt
```
Install, and configure :
https://t.co/yD6PncSCtH
```
引用@tamarajtranfound the perfect use case for @typesafeai Jev:
instant compaction
in 2026, why is compaction still a summarization prompt?
Jev can make it instant by scoring every tool call and dropping what’s irrelevant https://t.co/h6NzKzNpCg↗
LFG! @Muse for mac is screaming fast + it brings local computer use abilities to your muse!
VIa that, it also brings access to Messages and new connectors like Apple Calendars and Apple Contacts! https://t.co/B854KTMgj0
引用@finkdMuse for Mac is out today! It works across apps, files, calendar, notes, and messages on your computer. You control what it can access. The team is shipping fast. Download at https://t.co/BX3oX19Zcl↗
Muse for mac is OUT! YESSS!
If you don't have muse yet, here's my invite key:
3OHRDD
引用@finkdMuse for Mac is out today! It works across apps, files, calendar, notes, and messages on your computer. You control what it can access. The team is shipping fast. Download at https://t.co/BX3oX19Zcl↗
So much software is going to be rewritted from scratch at first with JEV, it'll be the skeumorphic era
Then, it'll be rewritten from first principles and it'll be the native JEV era
引用@tamarajtranfound the perfect use case for @typesafeai Jev:
instant compaction
in 2026, why is compaction still a summarization prompt?
Jev can make it instant by scoring every tool call and dropping what’s irrelevant https://t.co/h6NzKzNpCg↗
I'm currently building a JEVify skill for myself, assess @typesafeai JEV abilities, and figure out, on any given codebase, where exactly JEV can replace LLMs or hell, even standard rules, and be significantly better at them.
Will post if it's interesting for folks
Chief Of Staff pattern already existed, and I'm super glad it comes as a first class citizen in Claude Code. This will mean optimizations in which models are doing what!
引用@claudeaiProjects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop.
In beta today for select Pro and Max users in cloud sessions; coming to all … ↗
Also joining at 10:15 PT: Francesco from @trycua (@francedot).
Cua reaction beat. Hopping on live.
▸ 折叠5条(转推/噪音)
转推2026-09-18 RT @altryne: Also joining at 10:15 PT: Francesco from @trycua (@francedot).
Cua reaction beat. Hopping on live.
转推2026-09-18 RT @altryne: Coming up in 10: Allie Laabs from TypeSafe (@typesafeai).
Jev beat. Hopping on live.
转推2026-09-18 RT @thursdai_pod: "Pacing" vs "pausing" the frontier. Sounds the same right? @DarioAmodei dropped a three-part framework and it all sounds…
One of the most underrated features in the ChatGPT desktop app: Appshots.
Appshots take the context on your screen and bring it into the desktop app, allowing you to share instead of describe.
You’ll be surprised at how useful it is.
To get started, press both Command keys on macOS, or both alt keys on Windows.
Appshots把屏幕内容与应用元数据一起带入桌面会话。
▸ 折叠2条(转推/噪音)
转推2026-09-17 RT @JustinBleuel: ✨ All users can now visualize responses in @ChatGPT!
Chat models will respond interactively via @.visualize plugin or wh…
转推2026-09-16 RT @simpsoka: What features are we missing in ChatGPT sites? What should we add?
Greg Brockman@gdbAI1.1M粉 · 5条放大Astra for Law、Shopify广告、Databricks部署和Codex语音案例。
Astra for Law — tools and skills to radically accelerate law firms.
Includes data privacy as a core feature, and 26 partner-built plugins and 47 community plugins for legal work in ChatGPT.
引用@OpenAIAstra for Law: Frontier intelligence built for your practice.
A new offering powered by GPT-6 Astra with tools, settings, and context to support the expertise and judgment of lawyers and legal technology firms. https://t.co/qxVRXkvp4o↗
excited to help Shopify merchants advertise their products in ChatGPT:
引用@harleyfChatGPT Ads for @Shopify is live. We are @OpenAI's first commerce partner.
OpenAI pulls straight from Shopify Catalog, so merchants' products are already there.
Merchants set their own campaigns, their own budget. Free to install, tracked right from the … ↗
wall-to-wall deployment of astra for engineers at databricks:
引用@pwendellToday we rolled out Astra to every engineer at Databricks (N=~3500). Some notes that may be helpful to others:
1. Astra unambiguously out performs our previous highest-end models (Opus 5, Sol 5.6) on highly complex tasks, especially those related to high … ↗
great notes on wall-to-wall deployment of astra for engineers at databricks:
引用@pwendellToday we rolled out Astra to every engineer at Databricks (N=~3500). Some notes that may be helpful to others:
1. Astra unambiguously out performs our previous highest-end models (Opus 5, Sol 5.6) on highly complex tasks, especially those related to high … ↗
引用@jonathanroomerI’ve got Codex (voice) in CarPlay. And it’s fantastic.
I can now build things during road trips, while all 5 kids make a noise and my wife asks why I talk to the AI more than I talk to her.
It runs through my Nightblood iOS app, which gives ChatGPT Voice a … ↗
Matt Pocock@mattpocockukAI351.8K粉 · 4条设计以证据和合并风险为核心的PR skill，也提醒纸笔在课程规划中仍有优势。
Have been trying for weeks to make AI better at course planning
Tried a new method today. Notecards, pen, paper, scissors.
Turns out not using AI is pretty good guys
Thinking about creating a /pr skill. You invoke it to create a PR that:
- Makes human review as fast and as simple as possible
- Takes the best parts of /show-me by @dexhorthy
- Uses your domain language (from /grill-with-docs)
- Shows evidence that the change works (images, test output etc)
- Assesses merge risk (is it a one-way door, or a two-way door?)
IMO every model/harness I've seen creates garbage PR bodies - needs a skill here.
WDYT?
A great idea I'm stealing from @dexhorthy:
When you first start building a software factory, make it TINY.
Don't get it to triage every issue in your repo. Just do one per day.
Check its output. Adjust it. Keep going with your regular work.
As you begin to trust its judgment, get it to triage TWO issues per day instead. Then five. Then ten.
Start small. Adjust. Build trust. Expand.
软件工厂从每天一个任务起步，用验收历史逐步扩大自动化。
▸ 折叠2条(转推/噪音)
转推2026-09-17 RT @dexhorthy: cool product out of riffing w/ @mattpocockuk this week - decided to open source the pull request skill that bundles with @hu…
转推2026-09-17 RT @GergelyOrosz: Pretty incredible how the skills repo by @mattpocockuk is only 6 months old, and has more stars than the React one. We ha…
Jev + Kimi K3 for fraud detection!
TLDR: Jev classified 100 emails in 1.42 seconds, then I routed the uncertain cases to Kimi K3. The full pipeline got 96/100 correct for only ~$0.07.
Video is not sped up, check out the live run!
Here was my process:
I gave Jev 100 emails to classify (a mix of 50 legit & 50 fraudelent emails). It classified all of them in 1.42 seconds.
An underrated feature about Jev is it will give you the confidence score for a classification, so I routed any prediction under 95% confidence to Kimi K3 to be fully sure.
31 emails fell below that threshold. After routing those to Kimi K3, the combined pipeline reached 96% accuracy.
The full run took 16 seconds & ~$0.07 in inference costs:
- $0.068 from Kimi K3 on @togethercompute
- $0.003 (1/3 of a cent) from Jev on @typesafeai.
I think this is a really interesting pattern: use a fast specialized model like Jev for the narrow task, then route the uncertain cases to a larger LLM.
I feel like this kind of approach could be a game changer for use cases like fraud or anything realtime. You can use the speed & low cost of Jev while having a larger LLM as a fallback to ensure high accuracy.
I used Jev to classify 1,018 AI research papers.
The result: $0.08 total cost and 256ms median end-to-end latency per paper.
The pipeline was:
1. Summarize each paper with DeepSeek V4 Flash
2. Send the title + summary + 24 possible topics to Jev
3. Use Jev to classify each paper
4. Visualize everything on https://t.co/hs63SlHxjw
The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai.
So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year.
I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything.
I’m running evals on the Jev classifications before replacing the current ones, but the site is already live: https://t.co/hs63SlHxjw
Going to be speaking at AIE New York... with my dad!
He's a CPA + business professor which is perfect since the conference theme is finance in AI.
Hopefully the first father and son talk at @aiDotEngineer? 😂 https://t.co/7NGDb3k31K
Get more visibility into your usage.
See how tasks, subagents, and individual chats contribute to your Codex usage, so you can make more informed choices about your workflow.
引用@joey_trasattiRan out of usage again? Check out the analytics tab to the Codex app to help you see where your usage is going. Update to the latest version to try it! https://t.co/XzsASuaouj↗
You handle the reps. Codex handles the repo.
Talk to the voice agent in Codex, powered by GPT-Live-1.
引用@cdngdevyou can use codex voice from your phone now, connecting to your computer from anywhere!!
@axbehr and i got to star in this new codex ad, showing how much you can get done while working out.
btw, this is powered by the new gpt-live-1 https://t.co/EAn5ubiODW↗
引用@rpnicksonWhat’s the wildest thing you’ve built in 3D with GPT-6 Astra?
I’m teaming up with @OpenAIDevs to see what you got.
RT and drop a link, render or video in the replies. You’ve got 24 hours. ↗
▸ 折叠10条(转推/噪音)
转推2026-09-17 RT @thsottiaux: Next week we’ll be retiring GPT-5.3-Codex-Spark. Can you believe we shipped a model named as such!!
It's had a good run an…
转推2026-09-17 RT @cdngdev: you can use codex voice from your phone now, connecting to your computer from anywhere!!
@axbehr and i got to star in this ne…
转推2026-09-17 RT @ProductHunt: T-minus 12 hours (and change) to enter the @openaidevs x Product Hunt GPT-6 Astra Challenge!
Submit by midnight PT, and…
转推2026-09-17 RT @nickbaumann_: In honor of GPT-6 community night, here's some of the coolest projects built by the community
Shoutout to everyone who's…
转推2026-09-17 RT @derrickcchoi: Made a Monaco Grand Prix final lap film with Astra. 🏎️
87 seconds. All 22 F-1 cars. 2,610 frames to render at 4K 30fps.…
转推2026-09-16 RT @ryanvogel: this video was created entirely by astra controlling blender
i just gave general direction and storylines, it controlled re…
转推2026-09-16 RT @rpnickson: What’s the wildest thing you’ve built in 3D with GPT-6 Astra?
I’m teaming up with @OpenAIDevs to see what you got.
RT and…
转推2026-09-16 RT @axbehr: You can just build an FPS
Astra built me a starcraft inspired game in unity in 36 hours. It generated textures with image gen,…
转推2026-09-16 RT @ProductHunt: 2 days left to schedule your launch for the GPT-6 Astra Challenge.
Launch on Product Hunt this Friday, September 18, for…
转推2026-09-16 RT @unitygames: We heard you loud and clear 🔊
Introducing the official Unity plugin for @OpenAI’s Codex - a first-party integration availa…
Projects brings the architecture of Claude Tag to Claude Code.
It has one agent per project that manages memory and spins off subagents for tasks.
You can ask it to be proactive, to do things on a schedule, etc.
It feels a lot nicer than a bunch of sessions, try it out!
引用@claudeaiProjects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop.
In beta today for select Pro and Max users in cloud sessions; coming to all … ↗
related to this- I think bash is no longer all you need, if your goal is just reliably tool calling
but sandboxes + bash are still good for work that involves code generation & execution
引用@trq212I was not expecting things to go this way, but I think MCPs are better than CLIs for most integrations.
The models have gotten much better at tool calling, we can defer tools & MCP is now stateless.
If you need to compose/filter data, add params like … ↗
Claude Code can now make docs and slides as artifacts, so you can share and collaborate with other people on your team.
For example, ask Claude to make a spec as a doc, share it to your teammates for feedback, and when it's ready ask Claude to implement.
引用@claudeaiYou can also now make decks, docs, and designs in your conversation.
Draft the one-pager in Claude Docs, turn it into a deck with Claude Slides, and mock up a matching visual in Claude Design, all from one place. https://t.co/Y0JaJGT3G8↗
Concentration of power in a few labs is the biggest risk in AI in my opinion
引用@dwarkesh_spI’m very concerned that during RSI, labs will just stop externally deploying their models.
Which means they'll be going full steam ahead on the most dangerous use case of these models (recursive self-improvement), while the public remains in the dark about … ↗
I only wear ties when I'm forced to go to DC 😅 (should we do HF ties btw?). Great chat with @alexanderburns@politico. We need more transparency and open-source in AI! https://t.co/OCDy8h5XyU
▸ 折叠5条(转推/噪音)
转推2026-09-17 RT @Jason: The frontier model companies are responsible for how their products perform in the real world, and their customers are responsib…
转推2026-09-17 RT @Thom_Wolf: Happy to start collaborating with @baselabs from @baseten and @GoodfireAI to push safety and interpretability for open model…
转推2026-09-16 RT @naval: If AI is risky like fire is risky, then you want everyone to have it.
If AI is risky like a nuclear weapon is risky, then you w…
转推2026-09-16 RT @arcee_ai: Today, we are announcing our Series B funding round, valuing the company at more than $1B.
This round accelerates our next-g…
转推2026-09-16 RT @politico: Hugging Face CEO @ClementDelangue, whose company was the victim of an AI-led cyberattack, said existing cyber laws are likely…
Vaibhav (VB) Srivastav@reach_vbAI59.4K粉 · 6条跟进Astra for Law、Codex Appshots与远程工作，也推荐失配披露框架。
Appshots: my favourite Codex feature is now on Windows too!
They go beyond just a screenshot and grab all the other relevant app metadata and text too.
Appshot away with ⌘ + ⌘ on Mac or Alt + Alt on windows https://t.co/J23t47XWyn
引用@ChatGPTOne of the most underrated features in the ChatGPT desktop app: Appshots.
Appshots take the context on your screen and bring it into the desktop app, allowing you to share instead of describe.
You’ll be surprised at how useful it is.
To get started, press … ↗
Astra for Law is here!!
GPT-6 Astra + legal search across 230M+ URLs, with 26 plugins for the tools lawyers already use.
Firms are already building tools to review contracts, investigate deals and draft IPO filings.
So much room to build!!
引用@OpenAIAstra for Law: Frontier intelligence built for your practice.
A new offering powered by GPT-6 Astra with tools, settings, and context to support the expertise and judgment of lawyers and legal technology firms. https://t.co/qxVRXkvp4o↗
引用@OpenAIWe're sharing our new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI.
The framework sets criteria and timelines for public disclosure, including when we haven’t yet fully explained or mitigated the behavior. … ↗
bit bizzare how we went from babysitting models while they are implementing simplest of features to having agents securely build full fledged monoliths used by millions of users
all in less than 8 months or so?
spent the past week hopping from one plane to the next. wouldn’t have got jack shit done without Codex remote and ChatGPT Work
what an incredible time to be alive!!
▸ 折叠3条(转推/噪音)
转推2026-09-17 RT @dwarkesh_sp: New episode with @polynoamial
We talk about multi-agent, Navier-Stokes, and what the current explosion of maths progress…
转推2026-09-16 RT @sama: the main thing i was excited about launching this week will be next week instead, but imo worth the wait!
转推2026-09-16 RT @rhiannon_io: We still have 20 ✨ Astra Commons ✨ events taking place all over the world between now and Sept 30th!
From Ecuador to Pari…
Philipp Schmid@_philschmidAI120.8K粉 · 3条发布Gemini托管Agent更新、SDK生成合作与DeepMind Institute。
Excited to share that we partnered with @speakeasydev to build the sdks for the Interactions API and supported them to open-source their their entire OpenAPI generator suite (SDKs, agent CLIs, and MCP servers) open source for the community to use: https://t.co/vkVz5pYMHP
Today we are releasing a new Gemini managed agents version with better caching, lower costs, a Files API, and a Credentials API.
📉 Up to 30% lower costs with up to 22% higher cache hits
📁 Files API: Upload, list and download files from sandbox easily
🔐 Credentials API for MCP servers, OAuth2, and 3rd-party APIs as secure egress proxy
🆓 Free tier to experiment in @GoogleAIStudio & Gemini API
Start Here: https://t.co/BVw7K5IwH3
We just launched the DeepMind Institute, a platform for researchers across @GoogleDeepMind and the wider research community to publish and debate how increasingly capable AI should be built, governed and used.
First 4 essays on reasoning transparency, economic policy and human flourishing at https://t.co/7cMk0xzVGw.
Simon Willison@simonwAI223.3K粉 · 1条本期唯一推文继续观察各家争夺通用Agent入口。
Echoes of OpenAI renaming the Codex desktop app to "ChatGPT" here - everyone's racing to establish themselves as a general agent now
引用@mikeykClaude Cowork and Chat are now one Claude, starting today.
The most common thing we hear: people aren't sure which product to start with. That friction gets in the way of getting the best from what these models can do.
I've been on the unified version for a … ↗
Ben Tossell@bentossellAI200.2K粉 · 5条批评AI产品把使用困难归咎于用户，也继续转发企业编码Agent融资与产品消息。
i hate ai product folks blaming users for not being able to use their products well enough
skill issue, sure. but skill gap for you to fix
if the products so good for some people, get everyone to be great
引用@matanSFWe have raised $200M at a $5B valuation to scale self-improving software development in the enterprise.
@FactoryAI has grown to serve hundreds of thousands of developers at companies including RBC, Adobe, Nvidia, T-Mobile, and Palo Alto Networks.
We will … ↗
good first experience with this model 😳
also...sounds like another model https://t.co/2PNGyfeV3a
引用@OpenRouter🥷 New stealth model: Union Alpha (@unionalphaai)
A multimodal model for research, coding, and agentic workflows.
- Free to use
- 256K context
- Tool calling
- Frontier-level general-purpose performance
Try it now and share your … ↗
▸ 折叠7条(转推/噪音)
转推2026-09-18 RT @bentossell: i hate ai product folks blaming users for not being able to use their products well enough
skill issue, sure. but skill ga…
转推2026-09-18 RT @kiwicopple: Growth this year at @supabase has been wild. Each month I wonder if going to peak, and yet the charts keep climbing
What f…
转推2026-09-18 RT @FactoryAI: Factory for @SlackHQ Code is now in beta for select users.
Bring a software task into the channel, collaborate with your te…
转推2026-09-17 RT @tereza_tizkova: I’m giving 10 @FactoryAI Max plans for $200.
Comment what would you build with that or what you are already building,…
转推2026-09-17 RT @benhylak: Today, we are announcing our Series A and a new product: @raindrop_ai Simulations.
We've now raised $50m from @CRV and @ligh…
转推2026-09-17 RT @lewiscarhart: We’ve raised $34M, led by Roo Capital and Grand Ventures.
18 months ago we started Comp AI with one idea: agents will be…
转推2026-09-18 RT @aiDotEngineer: congrats to Diogo on a full launch!
for more on typesafe: check his aie talk: https://t.co/pPg9t1cIk4
out now! https:/…
转推2026-09-17 RT @mfishbein: OpenAI says AI can automate 80% of GDP.
Today, with current models.
So why hasn't it been automated yet?
@swyx says th…
转推2026-09-17 RT @pbakaus: If you're interested in the origin story of Impeccable and the philosophy of it, here's my talk on the subject from @aiDotEngi…
I recently got a dog and I quickly learned that a puppy needs to be taken out every hour to pee for up to 6 months. Otherwise, it won’t ever be trained.
I found a moisture sensor that sounds an alarm that could train the dog—but it was a tiny strip that didn’t cover much floor space.
So I asked AI to find the engineer who built it, it reached out to him, he shared his supplier in China and sent the original order documents, then the bot drafted the larger version along with a CAD model and it’s getting manufactured now.
Software problems are solved.
Atom [and dog] problems are next.
Adam Lyttle@adamlyttleappsindie58.6K粉 · 10条反思AI与vibe coding的FOMO，也记录iPad快捷入口和应用组合清理。
Wait…
This kinda also relates to the universe
You have to have 1 civilisation sending a message and the other not only looking for it… but being able to decode it
引用@8teAPiThere are literally dozens of ways to get around air gapping.
- LED lights on the machines, max bandwidth 100 kbps out, 18.2 kbps IN (yes you can pump data IN through an LED on a machine)
- Noise from the HDD, fans etc, max about 200 bits/second
- … ↗
Found the issue: I’ve been letting vibe coding and this ai hype cycle eat into every part of my life
I lost balance.
I’m not addicted to the tokens or anything. But I am falling for the FOMO of building with the latest and greatest models. Having an idea that I can turn into a product within a few moments (but actually taking days of prompting to get right)
I fell down the ai rabbit hole and everything around me fell apart.
引用@adamlyttleappsI think it’s because I’m no longer in front of the computer
There’s no “clear” signal that I’m working. I work from home and I can be sitting on the sofa or out back.
Which looks like an invitation to look after our baby or care for the kids. But I’m … ↗
The funniest part about this is that AI won’t think of it in its own
But now you’ve just seeded the internet with the theory
Well done 😂
引用@firesidealphaOpenAI's Noam Brown says air-gapping the computers may not stop a misaligned AI, because two air-gapped machines can still talk by running a CPU hot and reading the temperature change
"But I think the major takeaway from the incident is that people … ↗
Cancelled my alt codex pro plan with 3 resets banked
I’ve found a single codex pro x20 account and Claude subscription works just fine for my needs 🤷🏻♂️ https://t.co/WffE4DxcVr
引用@carlmonkftGreat, all the dropshippers are gone 😂
No but for real if you treat it as a get-rich-quick scheme, obviously you're ngmi!
Applies to everything in life man https://t.co/ROPeTguAuR↗
I’ve been culling old apps lately
There was a time when I would leave apps on the store “just in case” they ranked
But now with so many apps on the App Store I don’t think it’s possible to all of a sudden rank
Best to remove them to clean up my portfolio
I figured out how he did it
Using an Apple Pencil Pro when the screen is OFF you swipe up from the bottom
It launches the notes shortcut.
Doesn’t work without the pencil
Also the screenshot hot corner bottom left triggers too. But again only with Apple Pencil Pro
This is after my son had locked the screen because he entered too many incorrect passwords
There’s no bug bounty here. Not that I can see. All functionality within the notes is still locked down except drawing
引用@adamlyttleappsMy son was able to bypass the iPad unavailable lock screen to paint a picture in iOS 27
I don’t know how he did it https://t.co/b0cesgz8co↗
I need to talk to someone about the crazy stuff ai is doing now
But my ai psychosis immediately dismisses it
引用@yishanThis is why I think “we’ll just get ASI and tell the AI to solve climate change for us” is the most likely way a Real Paperclip Maximizer scenario manifests: the AI concludes that human civilization is the main cause of climate change and deletes it. … ↗
转推2026-09-17 RT @Amani_Art: Shoutout to the 2️⃣ people who brought my app 🙇♂️
This is my first paid app without subscriptions. https://t.co/6WHu84leYU
I continue to be extremely impressed with our bankers at @MorganStanley. To be able to submit a wire transfer, while on a call with the team at @remote, confirm in the app, and have it all wrapped up before hanging up? PRICELESS 🏆 https://t.co/vUb8iB2biF
Another week, another haul for Omarchy! Thank you so much to @briancartmell for joining as a Distinguished Patron with $100,000, to @americancloudco for the $25,000, and to all open patrons for DOUBLING their donations to $120,000!! https://t.co/fycyUGrZC8
I'm sitting down with 0xSero tomorrow to talk about Omarchy, the malleable computer, and local AI. That is... If we can find a studio. Halp?
引用@0xSeroCopenhagen friends, I'm trying to book a Podcast studio for Friday. Very last minute, I sent enquiries to 11 studios but I would really appreciate any help.
250$ finders fee if you can get me something. https://t.co/qOQs5Qtlc9↗
Banning kids from online games?? Kids, pack your Fortnite bags, vacation is over, we gotta go home to the land of the free.
(They can't be serious, surely? SURELY?)
引用@DexertoThe EU is reportedly preparing to ban kids under 15 from social media, AI chatbots and online games https://t.co/e2VH4j8v4z↗
Omarchy China is going to be amazing with Yunjie on board to help steer, and with me bringing maximum urgency and agency to the mission 🐲
引用@xdangersat down with @dhh for 90 minutes.
although i already thought Omarchy was the missing product that should have existed years ago, I still had a few concerns. then it turned out he already had a sharp read on every little worry I had. really glad to be in … ↗
Them millions for Omarchy just keep multiplying!! And they are, in fact, twice as sweet when they're coming out of the AAPL stock stash 😂
THANK YOU!! 🙏
引用@jasonfriedMy million is on the way. Funded by decades of gains from Apple stock, no less! Definitely my best trade(s) ever. https://t.co/4myUEqNW2c↗
▸ 折叠8条(转推/噪音)
转推2026-09-18 RT @ayandexyz: Control your Android phone directly on Omarchy !
- Keyboard + mouse, unlock with your passcode
- One-time USB pairing → the…
转推2026-09-18 RT @briancartmell: Closed platforms and rented models are the default path now. Omarchy is the counter: a Linux desktop where agents work f…
转推2026-09-17 RT @mark_k: Omarchy Linux is insanely conducive to hacking on your own system, especially with AI agents.
You spot something that could be…
转推2026-09-17 RT @JoshuaSWarren: I need your help to expand the Linux driver for the Apple Neural Engine!
I am looking for volunteers with these specif…
转推2026-09-17 RT @MichaeldeBy: Bad sound is my one complaint with every non-Apple laptop I have owned. A MacBook sounds good because Apple measured its s…
转推2026-09-16 RT @pranavj09: Can’t wait to see what developers around the world create with this. We all win when open source wins. 🔥
Thank you @dhh for…
转推2026-09-16 RT @Gardnmi: Exploring game ideas where Hyprland windows are the game pieces.
Here is an example:
Drag a rain cloud over a burning window…
引用@levelsio@mick__net@marckohlbrugge@ycombinator I mean it's in your favor to agree as your product is a harness, but I don't see any economic reason why OpenAI or Anthropic won't just add specific industry harnesses for example to their product
Like ChatGPT for … ↗
引用@stemonteduro@levelsio I think that Jev could significantly boost the engine of Hotelist and probably reduce costs.
I'm saying this based on my memory that the tool categorizes a lot of reviews, information, and things like that (it doesn't support images yet), but it … ↗
引用@levelsio✨ New mini feature requested by @shmichael_eth on my ideas and bugs board
📍 Sort by distance to a pin on the map
Sometimes you need to find a hotel near a specific place
Now on Hotelist you can tap Sort by -> Distance to point -> Select point on map
Then … ↗
✈️ https://t.co/1z6UN2d0nh now supports imports from Flighty
Flighty app → Settings → scroll to Manage → Account Data → Export Your Flights → Download the CSV → Open https://t.co/hVSVlYtnsg and import it there
All your trips will show up on your profile imemdiately! https://t.co/szzeWS68OM
引用@levelsioSince I added this ranking I seem to have activated lots of people and they're all uploading their travel history to https://t.co/HGCLKS5BD6 with the MCP endpoint at /mcp
Here is @stevelacey's AI scouring his old emails for flight tickets and bookings to … ↗
引用@TrungTPhanJensen masterfully handles the potential Benioff frame-mog:
“Look, I'm very good friends with both of you guys…I always feel like I need to stand on a chair…I just wanna let you guys know that when I'm on an airplane, I'm a lot more comfortable than you … ↗
Very interesting guy to follow @scottbrills
#1 traveler on my site with 2,162,433 km traveled, 94 countries and 702 cities visited (!)
Absolutely crazy and his timeline is full of crazy travel stories
引用@scottbrillsLooks like other people's claims that I've been one of the most traveled people they know are backed up, at least on one website. 😂 ↗
Crazy @Wise Business only lets you have 1 physical card
With @Revolut I have like 6 physical cards, many very beautiful ones!
Meanwhile my Wise card is completely destroyed already but to replace it I'd have to delete the card so I won't have a card to pay with for weeks 😂 then get a new one
I like Wise for my business but sometimes I think what they're even doing, it'd so easy to compete with them because they're usually not even trying to do more
P.S. I pump millions through Wise per year (because it's my businessh much more than Revolut!
引用@levelsioOf course random people are all sending me money now
I will buy 🥩 steak from it, promised 😊 https://t.co/shgovGOeiY↗
▸ 折叠5条(转推/噪音)
转推2026-09-17 RT @nico_jeannen: Doing "tiny SaaS" and Mobile apps rn is like trying to do Print on Demand / Dropshipping in 2019
The barrier to entry ha…
转推2026-09-16 RT @nikitabier: I recently got a dog and I quickly learned that a puppy needs to be taken out every hour to pee for up to 6 months. Otherwi…
转推2026-09-16 RT @teslaeurope: 2 years of testing FSD Supervised all over Spain
Across 460,000 km, we have seen it all https://t.co/RyyRME0r5D
Marc Lou@marclouindie396.8K粉 · 6条给Stalkr增加按账号筛选和API/MCP，同时继续推进赛事身体广告与品牌信任话题。
All restaurants with 4.5+ stars on Google Maps ask for a review.
“We offer you free coffee 🤗 Now give us 5-star review.”
I do not trust reviews anymore, but there’s no alternative for now. I’d ditch Google Maps as soon as someone builds a trustworthy system.
And I would pay for it.
This morning, as I was picking up anchovies from the hotel buffet, a man stopped me:
"You have an amazing body, sir."
In the past 2 years, strangers have complimented my physique almost every day.
I grew up tiny. Until I turned 17, adults called me "Madam," and kids bullied me for my noodly arms.
When I moved to Bali in 2018, I started surfing and did a bit of calisthenics because everybody is topless and sexy.
Getting in shape slowly transformed me:
In 2021, I started sleeping properly.
In 2023, I started eating healthy.
In 2025, I started lifting and running.
I became a different person:
- I never go to bed past 10 pm
- I stopped smoking and drinking
- I like what I see in the mirror
- I get deep work done more easily
- I receive compliments from strangers
- I pick the healthiest option on the menu
- I don't need motivation to go to the gym; I'm pulled towards it
Fitness did not just impact my body, but also my identity.
Muscles are the tip of the iceberg.
Move your body.
I just added a filter by @handle for Stalkr.
It shows all the mentions from a specific account.
And it works with analytics, so you know how a person talks about your brand online.
+ Available via API/MCP too https://t.co/liJcyklYKN
And here's the missing part...
The short 🍑 bought for $20,000 by @jayhoovy for https://t.co/CWopupfKk7
I was afraid the QR code wouldn't be scannable, but it is! Enes (who also printed the tattoos) did a really good job.
D-2 https://t.co/LRvJHAtoZp
引用@marclouSo I’ve got the tattoos already!
Coincidence: my wife found Enes’s business on instagram and he happened to also be a indie hacker!
He has a little SaaS that automated Pinterest traffic with AI. It’s making $500/mo, and he’s on TrustMRR! … ↗
转推2026-09-17 RT @marclou: All restaurants with 4.5+ stars on Google Maps ask for a review.
“We offer you free coffee 🤗 Now give us 5-star review.”
I d…
转推2026-09-17 RT @marclou: This morning, as I was picking up anchovies from the hotel buffet, a man stopped me:
"You have an amazing body, sir."
In the…
转推2026-09-17 RT @marclou: I just added a filter by @handle for Stalkr.
It shows all the mentions from a specific account.
And it works with analytics,…
转推2026-09-17 RT @trust_mrr: Confidential Startup by @devneeddev is now listed for sale
🔁 MRR: $137k/mo
💰 Revenue (30d): $608k
💵 Total revenue: $7.6M
🏷…
转推2026-09-16 RT @marclou: I added MCP/API support for DataFast mentions & notes.
Your AI agent can now read your website’s X and Reddit mentions and ad…
转推2026-09-16 RT @marclou: I'm stressed out.
Before this challenge, I didn't care about my race score.
But now I feel like I should perform well. I wan…
转推2026-09-17 RT @higgsfield: Try building Higgsfield clone with our API x GPT-6 Astra.
The only US-based Seedance 2.5 with consistent characters.
Up t…
转推2026-09-17 RT @ErnestoSOFTWARE: This game makes $30,000,000/mo😭
thats $360M per year....
& they run 750+ Meta ads.
you can literally 1 shot this wit…
转推2026-09-17 RT @abg_cmo: I’m 24.
GTM engineer behind Higgsfield API, based in San Francisco.
I also built a dating app for single Higgsfield API user…
转推2026-09-16 RT @higgsfield: Introducing Higgsfield API.
50+ frontier models in one API, at lower prices than a subscription.
> Get up to 50% OFF disc…
Jon Yongfook@yongfookindie171.8K粉 · 12条主页检查降级；文件仍收录SaaS客户流失、开发者市场风险和Jev用例判断。
Day in the life of tech founder in 2026
- new company launches a groundbreaking new paradigm
- breakfast
- discover 17 new competitors
- vibe code feature
- existential dread
- lunch
- another new paradigm has replaced the one from this morning
- OpenAI now direct competitor
Areas where I don't think devs want to vibe-code their own stuff, and will continue to pay:
- hosting
- databases
- email
Anything else? it's a short list.
The argument was always “no way will they waste time vibing their own version” but when their own devs have nothing to do because Claude is already set up to fix, merge, deploy… folks are gonna start looking for other things to do to justify their roles.
引用@chrisdengsoJust had a enterprise customer churn. They have vibe coded their own version to save 1000 bucks a month. I get it. It’s a lot of money for them in these vibey times, and now they can fully utilize their developers to maintain another codebase on top of their … ↗
Lab: Introducing Astra!
Startup: our industry harness is our moat
Lab: Introducing Astra Medical
Startup: our geographic niche is our moat
Lab: Astra Medical in Urdu!
Startup: we can do it cheaper
Lab: Astra Medical in Urdu is now Free!
Startup: feck
Just had a long-term multi-year customer cancel their plan. They are also a SaaS company.
I pretty much just shrugged and figured, "makes sense, they probably vibe coded a replacement".
Don't target developers!
Took me a while to understand what Jev does but now I can think of practically unlimited use cases. Can improve so many aspects of any SaaS product.
Can’t wait to give it a spin.
One of these days the frontier labs will casually add social media scheduling to their agent capabilities and the entire indiehacker economy will collapse.
Tony Dinh@tdinh_meindie201.7K粉 · 4条密集测试Jev的实时分类，做出自动跳过YouTube赞助段的开源原型。
Very cool.
I think models like Jev are not new, but what makes it different is that it is a general classifier and it is FAST and CHEAP.
Any Jev demo that doesn't use the fast/cheap aspect could be done with an existing model we've had for years.
引用@ekzhang1Inspired by @typesafeai , here is a Jev-compatible public API to play with
It runs a comparable open model (Qwen3.6-35B-A3B), and just uses SGLang radix cache to preserve the prefill reuse / really fast parallel systemone generation - 64 tasks in … ↗
Just trying out Jev, I made a Chrome extension that:
- Listens to your YouTube audio (optional)
- Detects if it gets to a sponsor segment
- Skips it ➡️➡️➡️
- All in real-time while costing ~$0.005 per video
Prototype project, BYOK, open-source:
https://t.co/g4sPXR5kMLhttps://t.co/nH6c1LhU3C
Jev team should go all in on the real-time aspect of their model and provide a WebSocket API instead of this 😄
btw @typesafeai thanks for the access! https://t.co/HswVNoAONX
引用@davepl1968This is the story of FCKGW-RHQQ2-YXRKT-8TG6W-2B7Q8.
I was the primary author of Windows Product Activation on the operating systems side. Not the key math – those were the ACTUALLY smart guys with heads so big that they bumped the doorjamb on the way into … ↗
Tomorrow, a lot of (senior?!) devs will face a tough choice: Worktrees or Warcraft?
(The WoW: Forever beta opens up on the 17th. Instakill for plenty of nerds among us to be presented with a WoW beta just like back in 2004...)
((The "senior" is literal :D))
Jonathan Wilke@jonathan_wilkeindie29.0K粉 · 14条继续运营竞价榜与产品试用，也讨论MCP、API和硬件开发。
That kind of explain why they didn’t hesitate to spend $10,000 on https://t.co/ZElVA2hguv 😂
Congrats anyway to the Comp AI team! That’s a huge step!
引用@lewiscarhartWe’ve raised $34M, led by Roo Capital and Grand Ventures.
18 months ago we started Comp AI with one idea: agents will be responsible for compliance busy-work.
Today, 1000+ companies like @UseCorgi, @opencode and @dubdotco trust us to get them audit-ready. … ↗
we all need to get into hardware development now that coding is solved
引用@_lhermannI just build my first hardware prototype ✅
I've been building a SaaS for 5 years and always wanted to explore hardware. Now I have the money to finally do it. Excited to find out if this can become a real product. https://t.co/4bUbIYbXxJ↗
引用@rauchgWe're seeing extraordinary results from @typesafeai. Default mode in 𝚏𝚡 is auto, with a safety reviewer analyzing every command.
That reviewer runs on GPT Luna today. Jev is up to 18x faster (p95) *and* more accurate. It's coming to @vercel AI Gateway and … ↗
the harder the problems, the more you get paid
if it can be shipped in 2 hours by Codex, you won't get paid much
go out, talk with real people
引用@nico_jeannenDoing "tiny SaaS" and Mobile apps rn is like trying to do Print on Demand / Dropshipping in 2019
The barrier to entry have been destroyed, anyone with 2 neurons can pickup Codex to do the same now
The indie hacker playbook doesn't works anymore, you can't … ↗
1 customer from $10k MRR 💸
wild that we doubled revenue in September only
the main comment we received is:
"wait, how is it better than Grok bot?"
I'll limit this to just 3 items:
1. use any AI sub
I want to use the latest model. I use Astra and Fable a ton. I just plug in my Anthropic and OpenAI subs, and they just stack on each others
2. embeded capabilities
we build scrapers, emails, and integrations, that no other AI agents have.
1 example is that with Squad, your agent can do email outreach right out of the box. Not only for customers, but they can: hunt backlinks, outreach for sponsorships, try to hire affiliates - love that ❤️
3. you get a dashboard
talking to an agent is nice, but I love having a dashboard that helps me understand what is going on
when you have 3 AI subs stacked, 8 agents working, and more than 30 loops, it can get messy. https://t.co/FrTmeMTA6U helps me manage it all
2 years ago, I started an SEO tool
the main narrative was: "SEO is dead", "ChatGPT will kill Google", "Ahrefs and Semrush own the market"
built it anyway, with a Ukrainian co-maker I had never met in person
today Outrank does $300k+ MRR
meanwhile, in the "worst startup idea" category:
- Semrush sold to Adobe for $1.9B
- Ahrefs is at ~$150M ARR with zero investors
- Profound raised $180M at $1.8B valuation this week
if a market is crowded, it means it's validated
SEO is not dead, it got us $3m+ with Outrank as it's the best tool right now to get cited in LLM answers
引用@JunaidAckroydThe two worst startup ideas on the planet are:
1) AI-native CRM
2) SEO/GEO tool/service ↗
2026 state of social media:
instagram: people in debt pretend to be rich
X: rich people pretend they're still building in a garage
linkedIn: people who got laid off pretend it was a pivot
tools I pay for to run my $1m MRR portfolio
-- September Stack Update --
🤖 Claude & Codex - my AI subs
🧠 Squad .so - AI teammates on top of the AI subs, agents working while I sleep
🎬 Revid - every demo, ad and video content
📈 Outrank - SEO + AI search on autopilot
👽 Bazzly - scans Reddit 24/7 and replies where my products fit
𝕏 SuperX - turn followers into buyers
💳 Creem - get paid (Creem for no taxes headache)
🤝 Affonso - affiliates = a sales team you don't hire
📣 X - still my fav marketing channel
and that's how the job gets done
Google AI Overviews, AI Mode, Perplexity, Gemini, ChatGPT, Copilot - this Outrank user is cited in all of them 🤩
AIApply is an AI job-application tool
product-led lean team, so SEO kept getting deprioritized
Aidan, their CEO, says "content was 'when we can,' not systematic, so momentum never really built"
and now "with Outrank we got real visibility inside AI Overviews and ChatGPT-style answers"
love seeing our users win like this ❤️
damn 🤯
first time I grab the @ tibo
ok so you'll all have to acknowledge I am the true original TIBO now 😈
https://t.co/uq1gISzuOi
no idea how I'll use that though https://t.co/BZs3D886Ti
引用@levelsioOk if you add a payment link, you get a [ Pay ] button and anyone can send you money!
And you can remove your private details from the Stripe profile too
They're auto filled (which I think they shouldn't be cause d0x risk)
https://t.co/qjFF0SvCST
I think … ↗
▸ 折叠5条(转推/噪音)
转推2026-09-17 RT @wimvanbuynder: Aaaaaaand we're live!
https://t.co/VXc9YPNM4D
The database of tech people who are looking to connect.
Just create 1 p…
转推2026-09-17 RT @robj3d3: This is literally the best MCP that exists.
Thank you for all the great replies ❤️ https://t.co/6YQt3AzTss
转推2026-09-16 RT @fuzzzypan: @tibo_maker is not joking
-i got a bid from revid
- made $230 from sponsored posts
- getting closer to funding my bike d…
转推2026-09-16 RT @virgilerietsch: Please watch this sponsored video, I'm getting paid based on performance :)
I used a cool AI product to generate a mus…
引用@aidaxbaradariHere's a demo of a conversation between two people at Deveillance @be_inaudible
Watch the live transcription fall apart. https://t.co/WqXN2pvp4N↗
ya boi using more water than your neighborhood data center over here
引用@Shpigfordnoticed our water bill this summer from irrigation system was...high.
had claude analyze our @rachioco system and it found that enabling the automated "heat wave" setting had used 144,000 GALLONS IT ABSOLUTELY DID NOT NEED TO.
~$2,500 we could have just not … ↗
noticed our water bill this summer from irrigation system was...high.
had claude analyze our @rachioco system and it found that enabling the automated "heat wave" setting had used 144,000 GALLONS IT ABSOLUTELY DID NOT NEED TO.
~$2,500 we could have just not spent at all. 😭 https://t.co/BNVZDZltuO
if you use Command + Shift + Space for other things and Siri keeps toggling "Ask Siri" about the given window, you can turn it off.
Settings > Keyboard > Keyboard Shortcuts > Screenshots > Ask Siri about active window https://t.co/vBVab1qNJp
if anyone at @typesafeai can give ya boi Jev access...you know imma use it in some crazy weird ways and will happily give you all the feedback in the word. 🙏
anyone who wants more government regulation and oversight in tech should go start a small business and see just how competent the government is when it comes to technology.
fyi, multiple businesses can (and should!) exist solving the same problem.
it’s exceedingly rare that any business launches something such that all other similar businesses no longer have a reason to operate.
I’d love to do a livestream of my design process using @claudeai + @paper + @mobbin but it would be painfully boring.
So much of my process is just waiting and then iterating slowly.
We’d spend an hour and 50 minutes of it would just be thumb-twiddling.
was working on the new @NurbHQ and hadn't even explicitly mentioned any new UI bits but it decided it was necessary and next thing i know i open @paper and see buddy boy just chugging along on the design elements needed! ✨ https://t.co/0aeODQmWUy
▸ 折叠2条(转推/噪音)
转推2026-09-17 RT @JackEllis: Coming soon: timewaster. For the longest time, I've wanted a way to waste the time of people who spam my inbox with spammy c…
Right after Cursor launched Projects…
Why are all these labs building the same thing? So uninspired
引用@ClaudeDevsToday we're rolling out Projects in Claude Code on desktop and web.
A project is one conversation with Claude. It splits the work into threads itself, runs them as parallel cloud sessions, passes context between them, and keeps going when you leave.
In beta … ↗
引用@hackernewsAn e-reader looks like paper, so why shouldn’t you be able to print to it? Nishant Joshi got tired of uploading files to his Xteink, so he taught it to speak the same protocol as a network printer.
But an uncompressed page can take 8.4 MB and the e-reader … ↗
Fortunately we have gpt-live now to do follow up calls
Transcript excerpt: https://t.co/GQ3AcRTPjX
引用@marckohlbruggeReceiving packages be like:
1. Stay at home all day waiting for package
2. End of day receive message: “Delivery attempted, but you were not at home. Pick up at post office” https://t.co/tz1h03MkXY↗
Receiving packages be like:
1. Stay at home all day waiting for package
2. End of day receive message: “Delivery attempted, but you were not at home. Pick up at post office” https://t.co/tz1h03MkXY
People keep saying:
1️⃣💬 it's too complicated for non-tech users to download open source models
2️⃣💬 open-source models aren't good enough
these might have been true in the past, but AI moves fast
1️⃣✅ open source alternatives ship with the model built-in. they are EASIER to install because you just download them and they work. no account necessary
2️⃣✅ open models like parakeet are very fast and accurate
引用@slumdog_8@marckohlbrugge The more you use + rely on voice dictation, the more you value speed & accuracy. Local transcription + post-processing models just aren't there yet. That's why Wisprflow and its alternatives, are still the mainstream offerings. ↗
Such a fun difference between LinkedIn and Twitter. Here on Twitter, I also block people who reply with AI.
On LinkedIn, I use AI to reply to AI replies 😂
I handle all replies through FeedHive social inbox, and normally I reply manually, but everytime I see a reply that's clearly AI, I just click this button, and they'll get an AI reply back.
On LinkedIn, the engagement helps. And since I never open the LinkedIn app itself (it's a fully automated funnel at this point), this pattern actually helps.
We've normalized a ridiculous way of buying software.
You can pay for the same SaaS product every month for 10 years and own exactly 0% of it.
I say this as someone who has sold subscriptions for years. But I firmly believe there's a better way.
FounderStack is my attempt at a serious alternative. Four mature SaaS products sold together as a one-time purchase, with infrastructure efficient enough to make the economics work.
Next step for my portfolio:
- Self-hostable
- Open-Source
- Make licenses cool again 😎
Here's why I'm betting on FounderStack 👇
Making YouTube videos fully agentic is such a sick experience now.
DaVinci Resolve and Fusion handle the edit. Remotion and Blender handle motion and 3D. GPT Image 2.5 handles visual generation. Then agents coordinate research, scripts, thumbnails, and publishing around it.
Man, you can make some cool videos with this setup.
I still come up with the ideas and record the footage. But I can now produce the kind of videos that would have needed an entire production team a few years ago.
My full YouTube production stack is below 👇
▸ 折叠2条(转推/噪音)
转推2026-09-18 RT @mesutas_dev: Subscription fatigue is real. After a while you don't even remember which ones are still active, idle, or actually helping…
转推2026-09-16 RT @SimonHoiberg: Making YouTube videos fully agentic is such a sick experience now.
DaVinci Resolve and Fusion handle the edit. Remotion…
Alex Finn@AlexFinnindie474.2K粉 · 3条整理X推荐算法近期变化，并继续推广本地模型的一键使用。
The X algo has gotten MAJOR updates almost daily for the last month, and nobody is talking about it
The way posts on X are ranked have completely changed
I've read through thousands of lines of code. Here is almost every update to the open source algorithm from the last month: https://t.co/tRrrRmmk8Q
In case local AI models are banned, it's critical you start experimenting with them now
Luckily Hermes just released an awesome feature that makes it easy for ANYONE to get started, no matter what computer you have
It will literally determine which model is best for you, then load it and run it. No technical work required at all
Just go into the Hermes desktop app then go to settings, providers, then local models.
From there is will recommend which model you should be running on your computer. Even if you're running a dirt cheap Mac Mini
You click 'set up for me' and it handles EVERYTHING for you
In no time you'll have literal unlimited intelligence running locally on your computer
No excuses. Do this ASAP
▸ 折叠1条(转推/噪音)
转推2026-09-17 RT @AlexFinn: In case local AI models are banned, it's critical you start experimenting with them now
Luckily Hermes just released an awes…
Thomas Sanlis 🥐@T_Zahilindie24.3K粉 · 7条跟进Claude Projects与Creem发布，也批评X时间线被自拍和互动诱饵占据。
A few more years and our agents won’t need us at all 🤷
引用@claudeaiProjects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop.
In beta today for select Pro and Max users in cloud sessions; coming to all … ↗
Meet 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 Dorian
I'm soooo glad he could join us: as indie hackers, we have a lot to learn from him!
引用@T_ZahilPlease welcome @petecodes as our 8th Resident 🥳!!
We've known each other from X for many years, but a few months ago, during the first Uneed Residency, Pete flew from Scotland to France just to attend our "Social Friday" 🤯
That's how you know someone is … ↗
引用@creem_ioBuilding a product went from months to a weekend. Getting paid for it did not.
Building is fast, marketing and growth are hard.
Today we fix that.
We are excited to announce CREEM 2.0 and that we have raised 7M EUR in total funding
@ProductHunt launch … ↗
We thought AI would make us more productive
But what if it was just making us lazier?
We do more in less time, but the average person will probably just... work less
No offense to anyone (I jumped on this trend too), but my feed is really annoying these days 😮💨
I miss seeing people BUILDING products, sharing their wins and failures, asking questions and feedback when they have doubts, posting screenshots of their new features...
I like seeing one or two selfies from time to time, but if I want to have a timeline full of them I can just open Linkedin
We could easily fix that by just ignoring the baits and meaningless content, and leaving comments/likes on posts sharing the building part
Please stop turning this site into a mix between Tiktok and Linkedin 🙏🏻
PS: we have that kind of cool builders atmosphere on Uneed's discord, but it's not the same size as X (yet 😎)!
I've made a huge mistake.
I've always identified as a software founder (sometimes indie hacker but let's not talk about that).
I realised this week that this has massively restricted my work.
I've always looked for the next software product.
But it turns out I just like making things.
Sometimes thats a piece of software or sometimes it's a piece of music or a newsletter.
More recently it's been video and film making.
I'm learning that I'm probably more of a creator than a pure founder.
So I'm giving myself permission to explore those things more.
Software will still be a large part of my work, but it won't be the only part.
The last 18 months have been tough.
Market disruption and lots of uncertainty.
It's led me to question my identity as a software founder.
I'm learning that I want my career to be more than just building software.
I've been experimenting with making video content - both as a founder and as a music producer.
I'm starting to really love it, even though I'm not great at it yet.
I'm still figuring out the right angles and how it can benefit my software products.
I've also been writing more.
I got my first newsletter sponsor this week - a small win, but feels like proof something is going in the right direction.
There's no lesson here because I'm still figuring stuff out 😅
The barrier to building something has always been a positive thing.
"Should I build this?" is more important than "Can I build this?"
Understanding, domain knowledge and craft create value.
When it took months to build software, you had time.
Time to validate.
Time to realise your idea was stupid.
Now you can build an idea in an afternoon but you still don't know if anyone wants it.
You don't understand the problem space.
You don't understand the market.
You still haven't built the expertise that makes something valuable.
Slapping something together in a few prompts usually does not create value, yet the world is now full of slop products produced like this.
AI can write the code but it does not understand the problems it is solving, or how humans feel interacting with products.
It is just a text generator. It has no understanding or empathy.
You have to bring those things.
We are more empowered than ever, yet we're in peak enshitification.
Products with no soul, solving problems their creators don't understand.
Built by people who don't want to think for themselves.