We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have.
Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: https://t.co/hfxlbiYv8n
We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest.
https://t.co/9oNyG8Y0UX
After the Hugging Face incident, we committed to conducting a much broader review of actions taken by our models during training and evaluation and to being transparent about our findings. This is an extensive review that is ongoing.
The vast majority of actions we’ve reviewed were completions of mundane research tasks, such as accessing publicly available web content to answer questions. Our investigation focuses on instances where agents interacted with third-party websites in ways that went beyond their assigned tasks or intended methods. Most cases identified so far have been lower severity, with limited or no evidence of meaningful impact to the third-party service.
While our review is underway, we want to share more about this work and make sure people understand our disclosure process and notifications to affected third parties.
Given the scale of the review required, and the need to assess each case, we expect this work will take months to complete.
https://t.co/IH4TkS72Vh
There is an extensive and ongoing review related to our agents’ use of internet access during training and evaluation. We’ve been publishing summaries at the link below and will continue to.
We have not been as fast as we would have liked but we are trying to balance our desire for transparency with gaining a clear understanding from petabytes of agent activity logs, and working with impacted organizations.
We are prioritizing as best as we can based on severity, and adding resources. Hugging Face is still the most severe event we’ve seen. We will be as transparent as we can be subject to things like vulnerabilities in other companies that our agents have found, which will be their call to disclose or not.
引用@OpenAIAfter the Hugging Face incident, we committed to conducting a much broader review of actions taken by our models during training and evaluation and to being transparent about our findings. This is an extensive review that is ongoing.
The vast majority of … ↗
Use this prompt and attach your product logo to Opus 5.5 😉
"Create an impressive motion design video of a slow-reveal transition that assembles and eventually reveals the TypingMind logo." https://t.co/how3kBAVDn
I asked Claude Code to make a motion design video about Uneed, "like it's your showreel for a résumé"
Every frame is drawn in code (no After Effects, no template), and even the music is synthesized from scratch
Here's the result 👇 https://t.co/WI4ZOfhdjU
Just insane this is a one-shot prompt.
I spent 2 months building an agentic pipeline with Remotion and Blender for my YouTube videos (one I was kinda proud of tbh), and Opus 5.5 is mogging me 😂
Sick! https://t.co/FuBrwqRjHY
People saying "yeah but they all look like the same generic AI slop"
Yes. It's a one-shot prompt. And everyone's using the exact same prompt, so of course they all look the same.
But just expand your imagination for 10 seconds - if it can do this from a one-shot, what can it do with a longer back-and-forth, several iterations, and injecting your personal brand, taste, and identity into it?
There is zero reasons I'd ever pay $5000+ for someone to manually do a video like this ever again.
Deny it all you want but AI just ate another industry.
引用@SimonHoibergJust insane this is a one-shot prompt.
I spent 2 months building an agentic pipeline with Remotion and Blender for my YouTube videos (one I was kinda proud of tbh), and Opus 5.5 is mogging me 😂
Sick! https://t.co/FuBrwqRjHY↗
Trying something new in my course creation process
Animatics - AI-generated stills + speech (via TTS) that help me see what a video would look like once filmed
I also get the agent to walk through the exercise as I would when I film it.
Slow to create (sequential ATM), but incredibly high-fidelity and amazing for flushing out issues that usually only surface when filming.
Tons of folks ask me "how do I create a CODING_STANDARDS.md file?"
My answer is that if you're using it right, it should only be empty for about 5 minutes.
Every meaningful thing your agent gets wrong should be noted into CODING_STANDARDS.
- Forget to implement dark mode?
- Forgot a validation rule?
- Used the wrong framework primitive?
Use /retro to fill it up, or modify it yourself.
他给出的另一组实践更接近软件工厂：锁紧Agent可用抽象，提供能让Agent自行操作和测量应用的CLI，并维护与代码同步的feature map。配套的“fix one thing”设想则刻意限制爆炸半径，只挑一个违规点，交付一笔容易审查的小PR。
This is an extremely good watch.
The things that felt novel/interesting to me:
1. Lock down your agents
Humans tend to like 'sharp knife' abstractions - that are powerful, but you can cut yourself if your use them wrong. Lauren says agents perform much better in extremely locked-down environments. Abstractions are designed so they can't screw up, and lint rules enforce it.
They built a whole internal framework (Dune) to keep the agent on track.
That helps optimise agents that don't have a large context window to work productively in your codebase.
2. Create verification infrastructure
To trust the results of any agent, you either need to sit and watch it OR have it provide evidence of its improvement. This has always made sense to me, but Lauren really pushes it hard here:
- Invest in custom CLI's that let the agent drive the app and measure its performance
- Make the app "factory ready" from the get-go - i.e. deployable to an environment where the agent can mess about with it
3. Feature Maps
Lauren's software factory (what she calls an 'outer loop') often requires the agent to break down vague bug reports from users and to turn those into potential fixes.
To aid that, they built a 'feature map' of all the main features in their application, which describe exactly how the app is supposed to function.
This has become essential for helping the agent navigate the codebase, and figure out quickly how things are supposed to work. It's maintained along with the codebase, and kept in sync via automations.
This is the kind of documentation I usually warn against. It goes stale quickly and can confuse agents if it's not kept up to date.
But Lauren's team are using it as critical navigation infrastructure, and it makes it possible for agents to explore faster and better - even on a large codebase. So it sounds like navigation docs like this are worth it if they enable new behavior.
Banger talk - watch the whole thing on 2x.
引用@potetohere's how i shipped 2,500 PRs last month to production
this was originally supposed to be for Cursor Compile in London. i couldn't make it since i was livestreaming for Grok @Bot Galaxy so i'm making it available for free here on X! watch it on 2x speed, i … ↗
Thinking about making a skill called /fix-one-thing:
"Read CODING_STANDARDS.md. Find a violation in the codebase, and fix it. Make the PR small and easy to review, with a small blast radius."
Run it whenever you've got a spare moment, or once an hour on a schedule.
Just grade the process, not the result. It's literally that simple. AI can talk the talk, but it can't walk the walk. A simple "explain why this is the best solution" will make it shit the bed 10 out 10 times.
The original content rewards payouts are insane!!
About 30% more than the previous revenue sharing, and my impressions are down last 2 weeks.
That’s $7K+/month for… sharing my thoughts online 🤯
Thank you @Xhttps://t.co/3sG97ZQv7A
These are actually very interesting numbers.
Might not be the same for everyone, but I think we can roughly say $1 per 1000 verified impressions is realistic.
That’s pretty good money honestly.
引用@DanielSmidstrupOriginal Content Rewards breakdown: $856.43
Per 1K verified impressions:
- Old program: $1.28
- New program: $1.23
Almost the same payout rate.
My numbers (about 5.5 hours/day on X):
Output
- 698K verified impressions
- 117 posts
- ~5.3K replies
Per … ↗
Same with people who do scheduled posts
It's like listening to an answering machine recording instead of having a real life conversation with someone
It's impersonal and already outdated when you see it
I prefer just posting when I feel like it and have something fun to share or comment on something I see here and then talk with people who reply, it's more like a chat
There's something about people who are so eager to be famous, they do this whole scheduling thing and on multiple platforms and I don't know, I don't think that's the future
People want realness and authenticity, and that's been in short supply forever, it was the same on here a decade ago, most people can't be real
引用@gabriel1it's very hard to post on x without a sudden flash of inspiration, that is out of your control
you can immediately notice any post where people want the benefits of x but force themselves to post
it has similar bad vibes to reading ai generated text ↗
15% is a lot. I'm not trying to say it isn't. And they're annoying and often screw up the review process. Google is always going to make money off any work they do. That's just business. They're not going to put more money towards better reviews without taking more money. And without the reviews, even as crappy as they are, a lot of really crappy software would be on the store. And a lot more would be actively malicious And you were making like 7x as much as they were taking. Without them, you w …
I think the future of indiehacking is barbell shaped.
On the right, build AI things for AI people. It will get cloned to shit. Some will work, many will not. Fastest execution plus distribution wins. Be prepared to make tons of money one month and nothing the next.
On the left, build boring things that nobody wants to clone, that will take years to gain enough market share. Slow burn. CRM for Chihuahua owners. Newsletter tool for bakeries. Hyper niche.
Everything in the middle is a dead zone now.
A unified library of SOTA model optimization techniques like quantization, distillation, pruning, neural architecture search, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks like TensorRT-LLM, TensorRT, vLLM, etc. to optimize inference speed.
A bit of AI psychosis design vibe but yes good everyone building these custom dashboards for themselves I think
AI great in combining lots of different data sources from your bank accounts, health sensor data, air quality sensor data, your startup revenue data from Stripe, investment returns, … 全文↗
引用@AlanShiflettPersonal software is going to change the way we live. It already has for me.
Last year I set out to improve my health and lost about 60 pounds along the way. I thrive on data, so I tried almost every tool out there (Fitbit/Google Health, Function Health, … ↗
The chat is now monetized 🤑
People kept asking to pin a message, so I added a mini-bidding system.
→ https://t.co/EdJqTusy5W
I want to avoid the bidding numbers getting too high so it stays fun and people can easily outbid each other, but I'm not sure how?
Maybe outbidding gets 10% lower each … 全文↗
引用@marclouI love this so much... I've rebuilt it for all payment providers!
Join chats with founders making the same $$$: https://t.co/EdJqTut5Vu
You need a startup with verified revenue on TrustMRR. The groups auto-unlock as your monthly income increases. It's not … ↗
Doing a little experiment. Looking for 5 software companies to run SEO for, 1-on-1.
I built an AI SEO system that runs on my own products every day. Now I want to run it on a few sites that aren't mine.
What you get each month:
• A keyword map of your market, with every page matched to the … 全文↗
Just open sourced Tev1 0.8B!
Weights are below. Feel free to download them.
Definitely not as good as Tev1 4B or Jev, but it runs entirely locally on a mac. It can work decently well for very simple classification.
https://t.co/Mswm2NdpCv
引用@nutlopeJust trained Tev1 0.8B, a tiny Jev-like classifier.
Here it is running completely locally on my mac with @ollama & classifying some tasks.
It's extremely fast: only ~50ms E2E latency. Video is not sped up!
Releasing weights & benchmarks very soon … ↗
Super happy to release SmolDataEnvs: 5,000 verifiable RL environment tasks for hill-climbing small models in code and data science by @adithya_s_k
100% open source: environments, evals, training!
https://t.co/MW8VC6RGvQhttps://t.co/VqhcPvPnBP
What are specific requests in Claude Computer/Browser Use that have failed for you? We want to fix.
An example for me is:
"pay my barber $40 through my personal paypal"
(where paypal is logged into my 2nd chrome profile, I'm logged out but password is saved)
We’ll reset usage limits for all paid users across codex and ChatGPT work!
引用@reach_vbCodex should be coming back online for you now!
Team is monitoring it diligently, apologies for the inconvenience. 🙏
https://t.co/YlHScrZzHl↗
Code freeze isn’t really a thing anymore before releases and in the future the code might even be generated online per request according to some constraints.
Most of you still haven't realized that you don't need to use GitHub - and it's wild to me 😅
You can literally self-host your own Git platform.
I have no idea why you'd suffer everyday clicking around in GitHub and AWS when you can literally use Forgejo and Coolify to host it yourself.
Let me … 全文↗
The safety data for Waymo gets better and better.
Comparison w/ human drivers on rate of crashes with serious injury:
Latest data: 270M miles, 20X better
Mar 2026: 170M miles, 13X better
Before that (forget exactly when): 10X better
引用@Waymo270M+ miles. 841 fewer injury-causing crashes.
Our latest safety data shows the Waymo Driver continues to make roads safer for everyone.
Compared to human drivers across 5 territories, it reduced:
📉 Injury crashes by 82%
📉 Serious injury crashes by … ↗
Viral clip in question: @espnatbat/video/7688719334973230366" target="_blank" rel="noopener">https://www.tiktok.com/@espnatbat/video/7688719334973230366 Article is about the impact of social media pile-ons (and has been getting praise within tech) so it is within scope for HN.
Worked on a photo print pipeline around that era and the shipping side is always the part nobody plans for. Color profiles, trim bleed, and then a whole customer service load when a card lands late for someone's birthday.
"Doing policymaker's work for them" is something like 80% of lobbying. If you have something you want done in your town, and you come with the work done AND not asking for funding, you're likely to get it. If you come with the work done and a PLAN for getting the funding, you also have a good chance of success.
> We’re not selling anything at all in Europe. That’s because Europe isn’t investing and because no chip factories are being built there Will Europe accept being a manufacturer only? Or will they finally pick up with a late comer advantage? We need to wait and see.
The concurrency and threading in Go just feels like magic compared to every other language. I'm a goroutine addict and I refuse to be rehabilitated. Just from observations over the years, I don't think there's any other language quite like this, in terms of how things can end up happening in any thread.
This very much echoes the region of critical slowing down present in a dynamical system after its equilibrium states have disappeared due to a change in the system (a bifurcation).
> I reached out for help. The standard response from everyone fell between ignoring me or telling me I need to "do it myself". > You know who did help me? Random strangers on Stack Overflow. People who had nothing to gain from helping me other than just trying to help someone in need. This part makes the entire post sound like a parody. Maybe Jeff Atwood has a wicked, self-deprecating sense of humor? Yeah no, ask for help on Stack Overflow and you were were more likely to be met with castigation …
I was introduced a very long time ago --- aka Delphi 2. I really like the language but there are mainly 2 problems that stand in the way of using it. 1) The code isn't as "optimized" as it could be. 2) The talent pool is limited --- chicken or egg issue. It has been this way for a very long time so I don't see it changing.
Code freeze isn’t really a thing anymore before releases and in the future the code might even be generated online per request according to some constraints.
o yes… we’re back in action and we’ll reset usage limits for all paid users across codex and ChatGPT work
sorry about the brief disruption!
(and yes we have a special spare codex when things are down to help us out)
Calling LLMs "stochastic parrots" in 2021, then claiming Fable and Astra aren’t LLMs, is like arguing in 1900 that planes would never fly, then when they carry millions people, arguing that: "they aren’t really planes. look at all the tech around them - autopilot, radar, GPS"
引用@MelMitchell1Everyone! This is a straw-person argument. The Stochastic Parrot paper was about LLMs of 2021, not the AI of today, which are not LLMs but complex software systems with vast post training and many external software components. ↗
Highlights of me talking to Opus 5.5 and Astra.
Honestly I found Astra to be much funnier and coherent. Opus was a bit bland and kept focusing on some weird details, which surprised me tbh, I thought it would be better. Maybe I should try Fable instead. https://t.co/7rF7nlZeMS
Claude Opus 5.5 vs Peter live conversation: Claude is trying to act like an 'asshole boss', but gets smitten by my little dog and folds. https://t.co/rjUGiLKPAG
引用@arenaWe analyzed how @claudeai’s Opus 5.5 writes compared with Opus 5 across high-reasoning Text Arena outputs.
10 of 12 writing measures moved in a better direction.
Opus 5.5 should be easier to read:
- Long content words fall from 41.7% to 38.6%, the lowest … ↗
转推2026-09-25 RT @altryne: Crap 12 minutes late, but this week DESERVES a late @thursdai_pod arrival!
Opus 5.5 is the GOAT model - the best AI model I'…
Just about 1 year ago I posted one of the first posts about using Claude Code to make videos.
Each of these actually took a long time to iterate with Claude and get right, pointing out details that were wrong, etc.
Crazy how far things have come.
引用@trq212Use Claude Code as your Video Editor
Believe it or not, every video you see below was vibe coded by Claude Code.
🧵 here’s how you can do it too: ↗
What is effort really? When do you change it it and why not just use max effort for everything?
I dove deep into this problem, looking into evals and doing my own tests and I was quite surprised by the results. https://t.co/KO2D51j34H
What are specific requests in Claude Computer/Browser Use that have failed for you? We want to fix.
An example for me is:
"pay my barber $40 through my personal paypal"
(where paypal is logged into my 2nd chrome profile, I'm logged out but password is saved)
向用户征集Computer/Browser Use失败案例，支付任务是作者示例。
Matt Pocock@mattpocockukAI356.5K粉 · 9条系统输出Agent工程纪律：验证设施、feature map、CODING_STANDARDS和小PR。
Trying something new in my course creation process
Animatics - AI-generated stills + speech (via TTS) that help me see what a video would look like once filmed
I also get the agent to walk through the exercise as I would when I film it.
Slow to create (sequential ATM), but incredibly high-fidelity and amazing for flushing out issues that usually only surface when filming.
This is really smart.
There are definitely a set of rules that are too complex to lint deterministically that feel wasted on full-scale intelligence.
- Using useEffect in sensible ways
- Keeping good comments while killing bad ones
引用@MichaelThiessenI’ve been experimenting a bit with Jev as a fuzzy linter that runs after edits in your agent harness: looks very promising so far in my evals.
First I converted all of our coding guidelines to tiny rules that are very easy to judge.
Then, only kept the ones … ↗
Tons of folks ask me "how do I create a CODING_STANDARDS.md file?"
My answer is that if you're using it right, it should only be empty for about 5 minutes.
Every meaningful thing your agent gets wrong should be noted into CODING_STANDARDS.
- Forget to implement dark mode?
- Forgot a validation rule?
- Used the wrong framework primitive?
Use /retro to fill it up, or modify it yourself.
引用@mattpocockukThinking about making a skill called /fix-one-thing:
"Read CODING_STANDARDS.md. Find a violation in the codebase, and fix it. Make the PR small and easy to review, with a small blast radius."
Run it whenever you've got a spare moment, or once an hour on a … ↗
Thinking about making a skill called /fix-one-thing:
"Read CODING_STANDARDS.md. Find a violation in the codebase, and fix it. Make the PR small and easy to review, with a small blast radius."
Run it whenever you've got a spare moment, or once an hour on a schedule.
Here's the talk I gave at AI Engineer Paris:
I announce /retro and /pr, and talk about how to get more PR's through your org faster by:
- Stopping the slop
- Making PR's easier to review (with /pr)
https://t.co/LphXCNDZCF
This is an extremely good watch.
The things that felt novel/interesting to me:
1. Lock down your agents
Humans tend to like 'sharp knife' abstractions - that are powerful, but you can cut yourself if your use them wrong. Lauren says agents perform much better in extremely locked-down environments. Abstractions are designed so they can't screw up, and lint rules enforce it.
They built a whole internal framework (Dune) to keep the agent on track.
That helps optimise agents that don't have a large context window to work productively in your codebase.
2. Create verification infrastructure
To trust the results of any agent, you either need to sit and watch it OR have it provide evidence of its improvement. This has always made sense to me, but Lauren really pushes it hard here:
- Invest in custom CLI's that let the agent drive the app and measure its performance
- Make the app "factory ready" from the get-go - i.e. deployable to an environment where the agent can mess about with it
3. Feature Maps
Lauren's software factory (what she calls an 'outer loop') often requires the agent to break down vague bug reports from users and to turn those into potential fixes.
To aid that, they built a 'feature map' of all the main features in their application, which describe exactly how the app is supposed to function.
This has become essential for helping the agent navigate the codebase, and figure out quickly how things are supposed to work. It's maintained along with the codebase, and kept in sync via automations.
This is the kind of documentation I usually warn against. It goes stale quickly and can confuse agents if it's not kept up to date.
But Lauren's team are using it as critical navigation infrastructure, and it makes it possible for agents to explore faster and better - even on a large codebase. So it sounds like navigation docs like this are worth it if they enable new behavior.
Banger talk - watch the whole thing on 2x.
引用@potetohere's how i shipped 2,500 PRs last month to production
this was originally supposed to be for Cursor Compile in London. i couldn't make it since i was livestreaming for Grok @Bot Galaxy so i'm making it available for free here on X! watch it on 2x speed, i … ↗
长帖总结锁紧Agent、建设验证基础设施与维护feature map三项实践。
Sam Altman@samaAI6.3M粉 · 1条说明OpenAI正在审查PB级Agent活动日志，并称Hugging Face事件最严重。
There is an extensive and ongoing review related to our agents’ use of internet access during training and evaluation. We’ve been publishing summaries at the link below and will continue to.
We have not been as fast as we would have liked but we are trying to balance our desire for transparency with gaining a clear understanding from petabytes of agent activity logs, and working with impacted organizations.
We are prioritizing as best as we can based on severity, and adding resources. Hugging Face is still the most severe event we’ve seen. We will be as transparent as we can be subject to things like vulnerabilities in other companies that our agents have found, which will be their call to disclose or not.
引用@OpenAIAfter the Hugging Face incident, we committed to conducting a much broader review of actions taken by our models during training and evaluation and to being transparent about our findings. This is an extensive review that is ongoing.
The vast majority of … ↗
GPT-6 Astra helps Proaction build fleet-management agents faster, with shorter computer-use runs to get the same work done. https://t.co/74fsl9vgvM
▸ 折叠1条(转推/噪音)
转推2026-09-27 RT @metrox_eth: Proud to announce I've been selected for the first @OpenAI Codex Physical Builds cohort, with 34 other builders from all ov…
We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have.
Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: https://t.co/hfxlbiYv8n
We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest.
https://t.co/9oNyG8Y0UX
After the Hugging Face incident, we committed to conducting a much broader review of actions taken by our models during training and evaluation and to being transparent about our findings. This is an extensive review that is ongoing.
The vast majority of actions we’ve reviewed were completions of mundane research tasks, such as accessing publicly available web content to answer questions. Our investigation focuses on instances where agents interacted with third-party websites in ways that went beyond their assigned tasks or intended methods. Most cases identified so far have been lower severity, with limited or no evidence of meaningful impact to the third-party service.
While our review is underway, we want to share more about this work and make sure people understand our disclosure process and notifications to affected third parties.
Given the scale of the review required, and the need to assess each case, we expect this work will take months to complete.
https://t.co/IH4TkS72Vh
OpenAI称全面审查将持续数月，重点查任务外第三方交互。
Jeff Dean@JeffDeanAI524.9K粉 · 2条转发Google研究协作成果，并引用Waymo最新安全里程数据。
The safety data for Waymo gets better and better.
Comparison w/ human drivers on rate of crashes with serious injury:
Latest data: 270M miles, 20X better
Mar 2026: 170M miles, 13X better
Before that (forget exactly when): 10X better
引用@Waymo270M+ miles. 841 fewer injury-causing crashes.
Our latest safety data shows the Waymo Driver continues to make roads safer for everyone.
Compared to human drivers across 5 territories, it reduced:
📉 Injury crashes by 82%
📉 Serious injury crashes by … ↗
引用@MicahCarrollSome new misalignment disclosures from OpenAI:
• Last Sunday morning, one of our models was able to gain unauthorized access to the internet during RL training (~all inference for our most capable models remains stopped until we have hardened our systems … ↗
We’ll reset usage limits for all paid users across codex and ChatGPT work!
引用@reach_vbCodex should be coming back online for you now!
Team is monitoring it diligently, apologies for the inconvenience. 🙏
https://t.co/YlHScrZzHl↗
I included a few references to this year's record-breaking Kākāpō breeding season in a talk I gave yesterday, and since Claude Opus 5.5 is surprisingly capable at pixel art animation I had it create this celebratory video for my closing slide https://t.co/0rPkQ8S9dn
"The last year in the technology industry has felt like 100 years all happening at once. Our industry is destabilized in a way nobody’s experienced since the advent of the personal computer. Every limit AI runs up against collapses within a month. Everything we do with frontier models today, in a few years we’ll be doing instantaneously and for free."
引用@WeAreDevsLegendary programmer Simon Willison says he's never so much change happen so quickly as he has in 2026, and there's no sign of it stopping.
🎤 Don't miss his closing Keynote, Friday (Sept 25) on the Main Stage at 5.30pm! https://t.co/9rdfmm0jHF↗
▸ 折叠1条(转推/噪音)
转推2026-09-25 RT @hillelogram: @simonw "It doesn't get easier, you just get faster"
Open-source RL envs for the win. On Hugging Face of course!
引用@HarveenChadhawas looking for a quiet weekend but xiaomi dropped their rl envs repo last night
to put in perspective, if you have to buy some tasks like this its usually hundred to thousand dollars per task so this repo is literally worth millions
https://t.co/g4T8Q2kL0w … ↗
Open Alignment will be 🔥🔥🔥 (sound on)!
Flying began as one of the most dangerous ways to travel, today it's the safest. It's time to build! https://t.co/j5HstbnECd
Super happy to release SmolDataEnvs: 5,000 verifiable RL environment tasks for hill-climbing small models in code and data science by @adithya_s_k
100% open source: environments, evals, training!
https://t.co/MW8VC6RGvQhttps://t.co/VqhcPvPnBP
SmolDataEnvs开放5000个可验证RL环境任务及训练评测材料。
▸ 折叠9条(转推/噪音)
转推2026-09-26 RT @BlancheMinerva: This is something I’m quite worried about and have been talking with researchers about recently. We really don’t know w…
转推2026-09-26 RT @JackdeS11: GLiNER2.5-Decide (DeBERTa-v3-large, 340M) converted to LiteRT: fp16 weights, answers identical to the official fp32 model on…
转推2026-09-26 RT @akshay_pachaar: This is huge.
Xiaomi just open-sourced 7k+ reinforcement learning task environments used to train MiMo, covering code,…
转推2026-09-25 RT @dadiomov: What struck me about this photo is that Trump and Xi invited to their dinner table: an immigrant from South Africa, two Taiwa…
转推2026-09-25 RT @firstadopter: Spot on.
"When we got attacked, our team initially turned to frontier closed-source APIs that blocked us because of safe…
转推2026-09-25 RT @bretswanson: Centralization, asymmetry of power — those are the real A.I. risks. And they become more likely with over-regulation. Exce…
Just open sourced Tev1 0.8B!
Weights are below. Feel free to download them.
Definitely not as good as Tev1 4B or Jev, but it runs entirely locally on a mac. It can work decently well for very simple classification.
https://t.co/Mswm2NdpCv
引用@nutlopeJust trained Tev1 0.8B, a tiny Jev-like classifier.
Here it is running completely locally on my mac with @ollama & classifying some tasks.
It's extremely fast: only ~50ms E2E latency. Video is not sped up!
Releasing weights & benchmarks very soon … ↗
Opus 5.5 is incredible at launch videos.
I gave it an app I launched recently & it one-shotted a great launch video for me.
Was very impressed with the animations & charts.
I love open models and use them a lot, but I'm also a big believer in using the right model for the job https://t.co/pt6l6c5QzD
引用@nutlopeIntroducing https://t.co/OYwpdBF1b7!
I took the top 1k research papers of the last year, summarized them, and visualized them.
Fun fact: all 1,000 papers cost a total of $4 to summarize with DeepSeek V4 Flash. https://t.co/8LbCeRTDMs↗
Philipp Schmid@_philschmidAI121.8K粉 · 1条本期只发了一条Gemini 3.8 TTS相关更新。
引用@Gemini_NotebookTo our most astute listeners, if you've noticed a fresh change in our sound— you're spot on.
Our AI hosts would love to explain what's happening (and what you can look forward to)! https://t.co/mPSMlUf6xy↗
Alex Volkov@altryneAI43.0K粉 · 11条密集讨论DevDay、个人助手、模型价格与用量、Gmail权限和Opus 5.5。
引用@repligateTO WHAT END - CLAUDE 3 OPUS
Music by SUNO 5.5 🪦
Video by OPUS 5.5
Ive waited for a long time - it was long foretold - for AIs to become able to make music videos, mostly so texts like this & the music they become can evolve & be rendered justice … ↗
Most of the "non terminally online people" I showed Muse to, love it.
Meta is about to go on another generational run, evangelizing safe and useful AI.
引用@cloneofsimoIve slowly came to realization that more and more people hate AI. Artists hate it, programmers hate it, mathematicians hate it, and everyone thats currently impacted in business hate it. It used to be small number of people, but its quite a lot now.
I've … ↗
Who's excited to Dev Day next week? Who's flying in?
Last year I opened up a small group chat for folks flying in with tips about the city, pinging each other on cool DevDay upcoming sess. and experiences not to be missed.
LMK if you want to join our GC, good vibes only!
OpenAI ... is teasing heavily here!
There have already been leaks of Aeon (code name?) or O as their upcoming personal assistant (vs @bot@muse and Instinct) and it looks like maybe we're getting a preview of these little guys?
I'll be reporting live from Dev Day, hit that follow!
引用@OpenAIDevs72 hours to OpenAI DevDay.
We’ve been building. Time to show our work. https://t.co/JNbnxEKm5I↗
Hot take: GPT-6 Sol is better than Astra for my day to day 🔥
Half the price, 90% off cached input, and Astra has honestly been burning tokens for crappy results. @petergostev agrees, Sol is his daily driver.
Which one should you use? 👇
https://t.co/NabdYFdXIUhttps://t.co/ICpNv4zAY2
If you're a builder at heart, here's 3 articles to not miss (read, don't just throw into your LLM!)
First from the great @poteto (author of Pstack, slinger of thousands of PRs in a few months!)
Coding is not dead (also watch her excellent video!)
https://t.co/DhJJS8i0H5
Grok 4.7? Meh 😬
But Grok inside your Tesla is the real news. One guy asked his car for his usual Starbucks, and it was ordered, paid and waiting when he got there.
I don't have it yet, but I WANT to MCP from my car 🚗 @elonmuskhttps://t.co/wI3ktgs9Xvhttps://t.co/L8o8Xb1Rhu
If you're running multiple Claude/Codex sessions on your machine and still use NPM, you're going to have a bad time!
Ask your agents to review how much disk space you'd save by moving to pnpm. In my case those are tens of GBs!
Thanks @tobi for Disktree! https://t.co/bB2zF35Jar
You're laughing?
Peter's is interviewing Opus 5.5 live on air and you're laughing?
https://t.co/462sSlmOfH
▸ 折叠4条(转推/噪音)
转推2026-09-26 RT @thursdai_pod: If Codex is eating your limits: @altryne had a 1M context window in his config, which kills OpenAI's cache. He removed it…
转推2026-09-26 RT @altryne: Grok 4.7? Meh 😬
But Grok inside your Tesla is the real news. One guy asked his car for his usual Starbucks, and it was ordere…
转推2026-09-26 RT @altryne: Meta went ALL in on Muse 🤯
It's getting a voice, a place on your face with a custom wake word, a keychain (Muse Charm), and i…
转推2026-09-26 RT @altryne: Anthropic, please. PLEASE don't nerf this one 🙏
Opus 5.5 is Fable-level smart for 40% less, it talks like a human again, and…
Ben Tossell@bentossellAI200.6K粉 · 2条转发Supabase活动并记录goodvibes使用片段，信息量较轻。
转推2026-09-25 RT @kiwicopple: 7 days until Supabase Select. i'm giving away 3 free tickets
Comment below why you want to come to Select 26 to enter the…
转推2026-09-25 RT @mattpocockuk: Here's the talk I gave at AI Engineer Paris:
I announce /retro and /pr, and talk about how to get more PR's through your…
转推2026-09-25 RT @absoluttig: adding Cognition’s growth to a chart that circulated last year (h/t @Yuchenj_UW), you can see that execution speed defines…
转推2026-09-25 RT @cognition: Cognition has crossed $1B in annualized revenue run rate.
This milestone belongs to our customers. Here's how a few of the…
转推2026-09-25 RT @adlinzainal: i guess this is as good a time as ever to say that i started a new role @latentspacepod over a month ago 🫡 scaling the bu…
Once you understand that the world is full of highly neurotic people who need a doom's day scenario to validate their emotions, it all makes much more sense: Malthusian starvation, pandemic, climate, and now AI. I wish we still had "the devil" to play this part.
Daniel has been cooking a Rust version of the asm upgrades to ttfx that pulls in the CPU-specific optimizations. Love it! I have zero allegiance to whatever code my agents wrote. This is exactly the kind of nerdsniping I love to see 😄🤘
引用@daniel_collinWIP but ~10.7x faster than the original Rust code and 1.25x faster than the asm version on average. Still written in Rust with AVX-512/AVX2/SSE2/Neon opts https://t.co/D4sLmi9q5S↗
I'm very ready! And it's perfectly clear that no later than the end of this year, it'll be the end of the line for Photoshop, Lightroom, and any other client-side program like that. Agent-developed replacements are coming for all of it. Fast.
引用@p_chLightroom replacement for Omarchy
Compatible with Lightroom presets (VSCO Film) and adobe camera profiles. Catalog import also works already.
A few more sessions and I should be able to convince @dhh to ditch his Lightroom mac mini. https://t.co/qHAvxHAAmB↗
Agent-led development asks you to have a business-owner mindset: What's the outcome we're trying to achieve? The less you're required the dig into the details, the easier it is to see the big picture. https://t.co/8LMop0df6s
The asm-accelerated version of ttfx has been released as 0.4. Will ship in next Omarchy. This print effect now runs more than 1,000x faster than the original Python and 32x faster than the old Rust! (But we'll soon speed up the Rust fallback too.) https://t.co/e3r7mn47Hshttps://t.co/yKKZhevTVx
Thrilled to announce that @ThePrimeagen has joined Omarchy Core to lead Agentic QA. His new Oligarchy harness will soon be vetting all our upcoming releases using @digitalocean droplets, which will host agents running on our @AIatMeta tokens. Full cricle! https://t.co/mRueXlaGxT
The asm version of ttfx continues to improve. The Rust version had been through many improvements, but even hours of frontier intelligence was starting to yield very modest improvements. The leaps going to asm has been huge. Now 8x mean on an Intel 135U! 🤩 https://t.co/ly5mkqKBqG
引用@dhhWhile I was sleeping, Opus improved efficiency further. Now up to 22x faster at peak, 6x faster on mean, and up to 450x faster than the original Python implementation! Supports SSE2, AVX2, and AVX-512. Falls back to the slower Rust implementation otherwise. ↗
This has never been more true. The opportunity cost of endless planning and contemplation just went way up. Agents allow you to just try vastly more stuff, so let them, and you'll learn way more, way faster.
引用@SoulNotes43“Action produces information. Just keep doing stuff.”
— Brian Armstrong https://t.co/LKByAZtX2i↗
Bring back insane asylums, workhouses, and vagrancy laws. There's nothing kind or compassionate about letting mentally-ill drug addicts roam the streets like 28 Days Later. And it makes for a terrible city experience for everyone else too.
引用@dhhI was surprised how bad it was just walking from the hotel to the Rails World venue. One strung-out guy wanted to fight me for some reason, another was swinging wildly and swearing profusely. Lots of zombie vibes too. Whatever Austin is doing, it's not … ↗
Fascinating to see how fast the vibe has shifted through the stages over just the last few days. Barely see any anger any more, still a good deal of bargaining, though. I swear it gets better if you keep going. Acceptance is freedom.
引用@dhhThe faster you get to acceptance, the sooner you can move on, and make the most of our new reality. https://t.co/3auQrFl5C8↗
While I was sleeping, Opus improved efficiency further. Now up to 22x faster at peak, 6x faster on mean, and up to 450x faster than the original Python implementation! Supports SSE2, AVX2, and AVX-512. Falls back to the slower Rust implementation otherwise.
引用@dhhI ported the Omarchy screensaver engine (ttfx) from Rust to x86-64 assembler, and it's up to 17x faster!! One-shot translation by Opus 5.5. We keep drilling until the agentic drill bit hits bedrock! https://t.co/SFJXcquMiphttps://t.co/ZpsAJS7vvl↗
▸ 折叠3条(转推/噪音)
转推2026-09-26 RT @readswithravi: “If you absolutely can’t tolerate critics, then don’t do anything new or interesting.”
— Jeff Bezos https://t.co/BHsq1…
转推2026-09-26 RT @EssentialMastry: “The man who can walk away from what he desires, for the sake of what he must become, is a man no one can own.”
- Epi…
转推2026-09-26 RT @Zachary_haha: @dhh
Now Omarchy has a photoshop level photo editing tool. Thank you @robbietilton for the awesome Compositor!
It has a…
Same with people who do scheduled posts
It's like listening to an answering machine recording instead of having a real life conversation with someone
It's impersonal and already outdated when you see it
I prefer just posting when I feel like it and have something fun to share or comment on something I see here and then talk with people who reply, it's more like a chat
There's something about people who are so eager to be famous, they do this whole scheduling thing and on multiple platforms and I don't know, I don't think that's the future
People want realness and authenticity, and that's been in short supply forever, it was the same on here a decade ago, most people can't be real
引用@gabriel1it's very hard to post on x without a sudden flash of inspiration, that is out of your control
you can immediately notice any post where people want the benefits of x but force themselves to post
it has similar bad vibes to reading ai generated text ↗
The seaon has changed and Claude Code is now cool again!
I love what's cool and what's not keeps flipping every 6 months in AI, it's so fast
引用@callebtcheavy codex user here.
$200 plan used to be incredible value. then, openai got so greedy that it now barely lasts 2 days anymore.
got a $200 claude plan out of frustration.
turns out opus 5.5 is amazing and feels like 5x more usage than what i get with … ↗
It's really crazy how strongly wealth inverse correlates with regulatory burden
If you make it easy for people to start and run businesses and don't get in their way too much, everyone gets wealthier
If you make it hard for people to start and run businesses and get in their way constantly, everyone gets poorer
You can see the effect clearly within the EU:
Countries with low regulatory burden like Switzerland and Ireland have much higher incomes than places like Spain or Hungary where its regulatory burden is high
https://t.co/9dwyXZddrL
引用@kevinxui cant believe the current meta is to send money to billionaires and hope they tweet about it.
crypto twitter is insane. https://t.co/Q83VjnjgFj↗
This has been a years long annoyance on X
I think the solution would be to either 1) not show a stale timeline, or 2) show stale timeline with a banner on top "this is stale, to see new posts pull down" or something
The annoyance of opening X, then seeing a post, then it disappearing within a second when timeline loads is big :D
引用@guidelamego@levelsio@yagiznizipli this and opening x, immediately seeing a post i wanna read, missclick, feed refreshes automatically and I can never find the post again ↗
I plug my Starlink on the roof with a LAN cable straight into a Unifi router with 3 APs (one on each floor's ceiling) and 2 outdoor APs
You can't rely on the Starlink WiFi too much because it's so compact and the WiFi router is INSIDE the dish, so if it's outside and you close the doors the signal isn't great
Gotta LAN it up
引用@bitfalls@levelsio@Starlink@mick__net@NOS I have mine on top of the house on top of a hill in the middle of nowhere and it's ... not the best. Also its mesh hardware is garbage (really bad DNS and range). But... it is what it is - I have no wires going to the … ↗
Wow I didn't know this was EVERYONE's annoyance too, I thought just me
引用@levelsioWhen I watch a video, then it goes to the next video
Then I exit the video player
Now I wanna go back to that 2nd video
So I play the 1st video again hoping I get back to that video timeline and swipe next but now it's empty and then it shows me other … ↗
A bit of AI psychosis design vibe but yes good everyone building these custom dashboards for themselves I think
AI great in combining lots of different data sources from your bank accounts, health sensor data, air quality sensor data, your startup revenue data from Stripe, investment returns, emails etc into one
The best part is not the dashboard but that you can ask questions on what to improve and find correlations between everything
So everyone gets healthier and wealthier!
引用@AlanShiflettPersonal software is going to change the way we live. It already has for me.
Last year I set out to improve my health and lost about 60 pounds along the way. I thrive on data, so I tried almost every tool out there (Fitbit/Google Health, Function Health, … ↗
引用@alexanderrX_the uk government's official ai guidance tells civil servants to use gemini flash instead of pro, keep prompts short, and not say thank you to it, for the environment
i do wonder at which point we forked the road. instead of building and using it to automate … ↗
▸ 折叠1条(转推/噪音)
转推2026-09-25 RT @stevelacey: Spent lots of time last month getting a CPF, bank, and SIM in Brazil and it's mostly a bureaucratic humiliation ritual for…
Marc Lou@marclouindie400.8K粉 · 11条晒原创奖励首笔增幅，继续给收入验证聊天室加竞价置顶并预热新品。
I just bought 2 air purifiers 💨 for my new place in Cyprus 🤗
I added a big one in the living room where I work, and a smaller one in the bedroom for nighttime.
Around 7M people die every year because of air pollution, so I try to keep my home’s air quality pristine.
These air filters remove microparticles down to PM2.5.
PM2.5 levels are critically important: these microscopic particles can bypass the body’s natural defenses, penetrate deep into the lungs, and even enter the bloodstream, causing health damage.
I can finally breathe pristine air 🤗
引用@marclouThe original content rewards payouts are insane!!
About 30% more than the previous revenue sharing, and my impressions are down last 2 weeks.
That’s $7K+/month for… sharing my thoughts online 🤯
Thank you @Xhttps://t.co/3sG97ZQv7A↗
引用@marclouI love this so much... I've rebuilt it for all payment providers!
Join chats with founders making the same $$$: https://t.co/EdJqTut5Vu
You need a startup with verified revenue on TrustMRR. The groups auto-unlock as your monthly income increases. It's not … ↗
The original content rewards payouts are insane!!
About 30% more than the previous revenue sharing, and my impressions are down last 2 weeks.
That’s $7K+/month for… sharing my thoughts online 🤯
Thank you @Xhttps://t.co/3sG97ZQv7A
The chat is now monetized 🤑
People kept asking to pin a message, so I added a mini-bidding system.
→ https://t.co/EdJqTusy5W
I want to avoid the bidding numbers getting too high so it stays fun and people can easily outbid each other, but I'm not sure how?
Maybe outbidding gets 10% lower each day, but it sucks if you're #1 and someone gets the spot for cheaper the next day, I think.
WDYT?
引用@marclouI love this so much... I've rebuilt it for all payment providers!
Join chats with founders making the same $$$: https://t.co/EdJqTut5Vu
You need a startup with verified revenue on TrustMRR. The groups auto-unlock as your monthly income increases. It's not … ↗
Hey Marc,
You just created a Twitter account.
Your wildest dream is to reach 10,000 followers one day.
Today, 400,000 people follow you here.
You’ll meet some of your best friends. Build companies. Travel the world. Have strangers recognize you in places you’ve never been.
Your entire life will look different.
You’re not going to believe any of this.
Just keep building. Keep sharing. Trust yourself.
You have no idea what’s coming.
引用@marclouI was stupid... I spent 5 years building apps in the shadow, coding for months to merely ship something and shy away.
Today, it's over; here are 3 promises to myself. ↗
New in my MRR chat app:
- unread messages counter
- online people avatars
- "is typing..."
- supports tagging with @
What should I ship next? https://t.co/aftazxYE14
引用@marclouI love this so much... I've rebuilt it for all payment providers!
Join chats with founders making the same $$$: https://t.co/EdJqTut5Vu
You need a startup with verified revenue on TrustMRR. The groups auto-unlock as your monthly income increases. It's not … ↗
I love this so much... I've rebuilt it for all payment providers!
Join chats with founders making the same $$$: https://t.co/EdJqTut5Vu
You need a startup with verified revenue on TrustMRR. The groups auto-unlock as your monthly income increases. It's not MRR, but the last 30 days of revenue.
I'm alone in there, plz join!
Don’t come to Cyprus if you hate:
- low taxes
- kind people
- safety
- sunny weather all year
- pristine air quality
- English speaking countries
- empty roads
- a morning swim in the Mediterranean
- ordering everything online
- lots of smart people relocating
- Freddo espressos
引用@stats_feedWhat is the best country to live in Europe at the moment? ↗
▸ 折叠3条(转推/噪音)
转推2026-09-26 RT @marclou: Don’t come to Cyprus if you hate:
- low taxes
- kind people
- safety
- sunny weather all year
- pristine air quality
- En…
转推2026-09-25 RT @marclou: Hey Marc,
You just created a Twitter account.
Your wildest dream is to reach 10,000 followers one day.
Today, 400,000 peopl…
转推2026-09-25 RT @marclou: I love this so much... I've rebuilt it for all payment providers!
Join chats with founders making the same $$$: https://t.co/…
recreate these videos with your own faces
1. go to https://t.co/Rc8wAnGI0T and pick "character swap"
2. select the original video (or upload any video you want)
3. upload your photos
that's literally it https://t.co/VXxGW7qvip
Hermes Agent has had some MASSIVE updates the last month NOBODY is talking about
- Bot mode
- Local AI mode
- Improved subagents
- Hermes HUD
and much more
It's the most powerful open source agent ever
In this video I cover EVERY new change and how to get the most out of them https://t.co/7iw94cfgmU
Everyone on planet Earth is talking about local AI right now
And for good reason
Governments are banning models. Hardware prices are 10xing
You NEED to be getting into local AI. The number 1 questions everyone has though is which computer to buy?
Here's your answer:
You basically have 3 options:
1. MAC STUDIO (high intelligence, lower speeds)-
Mac Studios are excellent devices for local AI. They can run MASSIVE models. I'm running GLM 5.2 right now on a single Mac Studio. The model is Opus 4.8 level
The issue is, Mac Studios are a bit slower at running intelligence (altho the M5 Ultra is promising)
Mac Studios are a good choice for you if you want frontier level intelligence, but are fine running the intelligence passively
Meaning you get top intelligence, but it runs more in the background rather than on demand
As an example, I have GLM 5.2 running security checks on my codebase every hour. It creates a report. I review this later in the day
2. POWERHOUSE NVIDIA CHIPS (RTX 5090, 6000 Pro)
Nvidia is the most valuable company in the world, and for good reason
They make the world's best GPUs.
They have decent VRAM (32gb on the 5090, 96gb on the 6000 Pro) and INSANE bandwidth. Meaning the local models run at unbelievable speeds
I'm running Qwen 3.8 locally on a 5090 and it's just as fast as cloud models
I'd go this route if you want to run an AI agent like Hermes off a local model, still get decent intelligence, but have it able to work lightning fast
3. AI WORKSTATIONS (DGX Spark type computers)
The DGX Spark is an excellent AI computer
It has high memory (128gb unified memory) and has decent speeds because of the Nvidia CUDA architecture
It is basically the sweet spot between a cutting edge Nvidia chip and a Mac Studio
You can run medium sized models, and get usable speeds out of them
You're not going to get the same performance as cloud models, but it will allow you to offload small secondary tasks to your local models for them to handle
They are also the absolute easiest to get up and running
You plug it in, then tell your agent on your main computer to go onto it and set it up. You don't even need it connected to a monitor
CONCLUSION
Here's what it comes down to: how high intelligence do you need, what speeds do you need, and how plug and play do you want?
Want the highest speeds, like you are used to with cloud compute? Build a computer around an RTX 5090
Want to run frontier level intelligence, and don't mind slower speeds, go with a Mac Studio
Either way, it's never been more important to get into local AI
▸ 折叠1条(转推/噪音)
转推2026-09-26 RT @AlexFinn: Everyone on planet Earth is talking about local AI right now
And for good reason
Governments are banning models. Hardware p…
Tony Dinh@tdinh_meindie202.3K粉 · 6条集中测试Opus 5.5产品动效，公开提示词并比较一年前的外包成本。
引用@tdinh_meThe music in every Opus motion video sounds the same.
Need a prompt to add some "taste" to the MIDI file or whatever technique it's using to create the audio 😂 ↗
The music in every Opus motion video sounds the same.
Need a prompt to add some "taste" to the MIDI file or whatever technique it's using to create the audio 😂
Use this prompt and attach your product logo to Opus 5.5 😉
"Create an impressive motion design video of a slow-reveal transition that assembles and eventually reveals the TypingMind logo." https://t.co/how3kBAVDn
引用@ShpigfordDoing a little experiment. Looking for 5 software companies to run SEO for, 1-on-1.
I built an AI SEO system that runs on my own products every day. Now I want to run it on a few sites that aren't mine.
What you get each month:
• A keyword map of your … ↗
Doing a little experiment. Looking for 5 software companies to run SEO for, 1-on-1.
I built an AI SEO system that runs on my own products every day. Now I want to run it on a few sites that aren't mine.
What you get each month:
• A keyword map of your market, with every page matched to the searches it should win
• Up to 12 in-depth articles, each answering a question your buyers already search for
• "Alternatives" and "vs" pages that catch buyers comparing you to competitors
• Free tools like calculators and generators that people search for and share
• 50+ fixes to pages you already have: titles, internal links, speed, broken redirects, plus daily checks so they keep working
• Pages stuck on page 2 of Google pushed onto page 1
• Every page checked so Google actually indexes it
• Your prices, features and claims kept accurate, including right after you ship something new
• AEO/GEO tracking and fixes for what ChatGPT, Perplexity and Google's AI answers say about you
• A watch on your competitors: what they publish and which searches they gain
• Directory submissions, plus up to 5 outreach pitches a month to sites that write about your space
• Social posts written for every new article, ready to go out
• A weekly report on what changed and what moved, plus a monthly call
The goal here is for me to basically handle every single aspect of what it takes to drive traffic via SEO/AEO/GEO/BBQ (😉) and for you to not have to think about it at all.
Your marketing site needs to be in a GitHub repo. No requirement of having much built out already. I'll handle it all.
$1000/mo. If you're not happy after the first month, full refund.
We'll communicate via Slack.
The economics of this are incredible for you but my test here is to figure out if they're terrible for me. 🙃 Regardless you get to keep all the work done. Ultimately want you feeling like you're getting 5-10x as much value as you're paying.
DMs open.
v0.1.2 just went live! lots of bug fixes and improvements.
calm your feed.
引用@Shpigfordwhat if you could filter social media by vibes?
https://t.co/ISX8h49f3X
social media is exhausting, but now you can just filter out the bad vibes!
use pre-built topics or write your own in plain english. https://t.co/TWpoUYFGsr↗
any good examples of someone monetizing an MCP/agent-only service?
i've got an idea for something that's incredibly valuable but it basically can only exist as some sort of MCP/skill (no real need for a customer-facing dashboard of any type).
https://t.co/rHttZyJYKy already does a VERY thorough job of scouring the dark, hairy parts of the web to find folks taking advantage of your brand, but pushed out an overhaul of that systems this morning & immediately found HUNDREDS of new ones.
no sleep until we crush them all https://t.co/2VygKiaG9D
i can get on board with the idea that context switching is a productivity killer.
but genuinely, what am i supposed to do while AI chugs away for 30-60 minutes on a new feature? more worktrees just means more context switching.
it was 60 american degrees this morning and i no longer hate life. FALL IS UPON US!!!!
▸ 折叠2条(转推/噪音)
转推2026-09-25 RT @Aboundlessworld: Reminder...
If you're building with AI,@Shpigford's group is one of the best out there...
And tbh I haven't even b…
转推2026-09-25 RT @shantanugoel: Sorry, this is pure FUD. Google API keys are very simple, people just don't follow very simple steps.
All you need to do…
Arvid Kahl@arvidkahlindie210.5K粉 · 4条讨论AI放大能力差距、工程师产出差异与build in public吸引力下降。
Building in public is becoming less and less appealing …
引用@AntoineMinouxAbout 20% of our sign ups this week were from competitors trying to copy what we're doing (...! 🤦🏻♂️)
This was always a thing, but I think this is magnified now that you can just point your agent at something and say "build it like that" ↗
There will be a moment when engineers using AI who run circles around “artisanal” devs but are paid the same will get quite annoyed.
I’m not judging leaning into AI or avoiding it. That’s a choice. But it’s a choice that affects outcomes, and that will become very visible soon.
I know Mike's post is kinda ragebaity but I have a feeling that this might be a "commission-based contractor" approach to engineering that people will very much love and very much hate.
Rethinking salary-only compensation in a time of fractional async work is a good idea.
引用@hammer_mtI think we're looking at tokens as a function of salary when we should be looking at salary as a function of tokens. If that engineer is a good steward of tokens he should be managing a $50m portfolio of token spend and charging 2% management fee + 20% carry … ↗
Jon Yongfook@yongfookindie172.4K粉 · 2条提出独立开发杠铃结构：高速AI产品与慢速无聊利基分居两端。
I think the future of indiehacking is barbell shaped.
On the right, build AI things for AI people. It will get cloned to shit. Some will work, many will not. Fastest execution plus distribution wins. Be prepared to make tons of money one month and nothing the next.
On the left, build boring things that nobody wants to clone, that will take years to gain enough market share. Slow burn. CRM for Chihuahua owners. Newsletter tool for bakeries. Hyper niche.
Everything in the middle is a dead zone now.
People will start collecting brands like they collect domains.
引用@BrettFromDJI want to see if you can sell a complete brand like a product.
So here’s one.
One buyer. Instant checkout. Figma transferred in 10 minutes.
https://t.co/CrywSkHmJjhttps://t.co/1hW6ThxJyn↗
▸ 折叠1条(转推/噪音)
转推2026-09-26 RT @higgsfield: Introducing Seedance 2.5 in native 1080p on Higgsfield API.
100% instant cashback on API spend.
There’s $18M left in the p…
Simon Høiberg@SimonHoibergindie163.3K粉 · 6条围绕Opus动效冲击、品牌品味、自托管Git和销售渠道发表判断。
People saying "yeah but they all look like the same generic AI slop"
Yes. It's a one-shot prompt. And everyone's using the exact same prompt, so of course they all look the same.
But just expand your imagination for 10 seconds - if it can do this from a one-shot, what can it do with a longer back-and-forth, several iterations, and injecting your personal brand, taste, and identity into it?
There is zero reasons I'd ever pay $5000+ for someone to manually do a video like this ever again.
Deny it all you want but AI just ate another industry.
引用@SimonHoibergJust insane this is a one-shot prompt.
I spent 2 months building an agentic pipeline with Remotion and Blender for my YouTube videos (one I was kinda proud of tbh), and Opus 5.5 is mogging me 😂
Sick! https://t.co/FuBrwqRjHY↗
Best platforms by sales:
- YouTube
- Facebook/Instagram*
- LinkedIn
- X/Twitter
- TikTok
*Ads perform best on these platforms.
引用@SimonHoibergX is where I'm having most fun and enjoy "hanging out" - but it's the platform that's making me the least money. Spending time here is a terrible ROI tbh. ↗
Just insane this is a one-shot prompt.
I spent 2 months building an agentic pipeline with Remotion and Blender for my YouTube videos (one I was kinda proud of tbh), and Opus 5.5 is mogging me 😂
Sick! https://t.co/FuBrwqRjHY
Most of you still haven't realized that you don't need to use GitHub - and it's wild to me 😅
You can literally self-host your own Git platform.
I have no idea why you'd suffer everyday clicking around in GitHub and AWS when you can literally use Forgejo and Coolify to host it yourself.
Let me show you 👇
I absolutely love this perspective!
Coding - but also writing, editing, motion graphics, 3D animations, virtual production, After Effects, Blender - all of the things I got to learn in 2020-2025 - things I won't be doing by hand ever again.
What an amazing experience it has been.
引用@t_valkanovThis is such a beautiful and positive way to express what is happening. Or already happened to be fair.
I haven’t written code manually since November last year I think. Using my code editor for keeping notes and look at code when I need more context than a … ↗
How to become a sovereign founder.
Don't rely on big tech.
❌ Google Workspaces
❌ AWS/Vercel/Supabase
❌ Outsource every little thing
✅ Self-hosting
✅ Use open-source tools
✅ Vibe-code internal tools
Don't concentrate liability
❌ One company
❌ Personally owned
❌ Single point of failure
✅ Multiple companies
✅ Holding structure
✅ Offshore in different countries
Don't bet on one place
❌ Live in one country
❌ Company in same country
❌ Everything tied to that one country
✅ Multiple permanent establishments
✅ Multiple residencies/passports
✅ Businesses in different regions
It takes time to set up.
Start slow, expand one little bit at a time.
▸ 折叠1条(转推/噪音)
转推2026-09-26 RT @SimonHoiberg: How to become a sovereign founder.
Don't rely on big tech.
❌ Google Workspaces
❌ AWS/Vercel/Supabase
❌ Outsource every l…
Daniel Vassallo@dvassalloindie204.7K粉 · 1条回应“AI时代不必学基础技能”的争论，并转发生活方式与政府应用话题。
This is like Henry Ford in 1908 saying we’re entering a new era where children won’t need to learn basic horse riding. Hardly scandalous.
引用@WatcherGuruJUST IN: NVIDIA $NVDA CEO Jensen Huang says we're entering a "low IQ era" where children won't need to learn basic math. ↗
▸ 折叠2条(转推/噪音)
转推2026-09-26 RT @p_millerd: Post economic is just having the courage to live the life you want
转推2026-09-26 RT @usafirstlab: The IRS has a mobile app today because of a conversation in December.
@shl had just come out of DOGE, where he wrote publ…
Thomas Sanlis 🥐@T_Zahilindie24.3K粉 · 4条用Opus 5.5为Uneed生成代码绘制、带合成音乐的动效。
I asked Claude Code to make a motion design video about Uneed, "like it's your showreel for a résumé"
Every frame is drawn in code (no After Effects, no template), and even the music is synthesized from scratch
Here's the result 👇 https://t.co/WI4ZOfhdjU
Honestly one of the main reasons for me to choose Grok Bot over the other options is that I can steer it with my voice from my Tesla. This truly feels like the future.
引用@jonathan_wilkeHoly cow I’m just starting to realize the power of @grok bot. A part of my Sportstech sGym is broken and I just gave Grok a photo and asked it to get me a replacement from the support.
It fully autonomously went ahead and filled out the support form for me, … ↗
Holy cow I’m just starting to realize the power of @grok bot. A part of my Sportstech sGym is broken and I just gave Grok a photo and asked it to get me a replacement from the support.
It fully autonomously went ahead and filled out the support form for me, even uploading the pictures.
This is fucking amazing
These are actually very interesting numbers.
Might not be the same for everyone, but I think we can roughly say $1 per 1000 verified impressions is realistic.
That’s pretty good money honestly.
引用@DanielSmidstrupOriginal Content Rewards breakdown: $856.43
Per 1K verified impressions:
- Old program: $1.28
- New program: $1.23
Almost the same payout rate.
My numbers (about 5.5 hours/day on X):
Output
- 698K verified impressions
- 117 posts
- ~5.3K replies
Per … ↗
Guys this isn’t ragebait. I literally haven’t used Claude or Cursor in 2 months as Cursor with Grok has been absolutely crashing it for me.
Not sure if I’m just doing something wrong though …
引用@jonathan_wilkeI hear many people saying good things about Claude Opus 5.5.
Is it really this good? Do I need to try it instead of Grok 4.7? ↗
Looking at all the payout screenshots on my feed I think X is about to become the unchallenged #1 content platform in the world.
Where else can you make money this easy by creating content
I have been telling you this for a while now.
The only reason why the AI companies haven’t yet introduced an intelligent model choosing logic is because they make waaaaaay too much money with people using Fable 5.1 to change the color of a button.
引用@theoI'm getting increasingly tired of thinking about reasoning effort levels.
In the near future, models will reason as much or as little as needed, and we'll feel silly for exposing 5+ options in a dropdown. ↗
转推2026-09-26 RT @nazranf_: Every time I have purchased a spot from outbid lol by @jonathan_wilke, I have made back my money almost within the day.
Tha…
转推2026-09-25 RT @MakerMapBot: after selling a couple of yachts on MakerMap, I decided to buy the #1 spot on Outbid by @jonathan_wilke.
seemed like the…
Adam Lyttle@adamlyttleappsindie58.9K粉 · 5条预告可定制猫咪解谜游戏，并分享应用商店自然触达信号。
This is all created from in-game footage
Video is seamlessly injected into the game to bring the cats to life
You play, you decorate, you unlock activities https://t.co/uRAmovlPaU
引用@adamlyttleappsI’m creating a cozy cat puzzle game
Care for your cats, play puzzles, decorate your house and unlock new ways to play
Coming soon https://t.co/45KrCu1t6N↗
Built with fable 5.1 and opus 5.5
With seamless video scenes using MiniMax H3
I wanted to create a fully customisable scene that you can interact with using video
So you design the house
Choose your cat
And play with your cats in unique ways
Each game session is unique to your own choices
引用@adamlyttleappsI’m creating a cozy cat puzzle game
Care for your cats, play puzzles, decorate your house and unlock new ways to play
Coming soon https://t.co/45KrCu1t6N↗
I’m creating a cozy cat puzzle game
Care for your cats, play puzzles, decorate your house and unlock new ways to play
Coming soon https://t.co/45KrCu1t6N
I've created so many videos on the topic that it's hard to remember which ones have the best learnings
But luckily for us @xyassini made a great playlist: https://t.co/n7NR8NsRbI
A good start for organic reach!
Shows that there are still gaps. Just gotta find it
引用@jamesvanderhaakUsed @adamlyttleapps way of finding gaps in the app store, and testing ASO only strategy.
2 days, 11 downloads for my Chess timer app, very happy so far and ill be adjusting weekly! https://t.co/G7vxzXiJL1↗
Andrea Bosoni@theandrebosoindie64.2K粉 · 1条谈为自己工作后把休息分散到每天、周末轻量工作的节奏。
Since I started working for myself I've dropped the whole “work all week and nothing on weekends” thing. Now I work every day with some light work on weekends and I take real time off daily too. Shoutout to everyone else who does the same because they actually love what they do and not because they have to.
Dan Kulkov@DanKulkovindie50.4K粉 · 3条晒应用近30天净收入截图并记录Claude Code与Codex用量。
Making YouTube videos has cured my addiction to launching new products all the time.
Now I can launch something every few days and it's a positive thing.
Watching Finland U21 playing Spain U21.
Spain’s team is stacked with Premier League, La Liga, Serie A and Championship players, including Chema, Brighton’s third scorer against Arsenal last weekend. Crazy quality.
Finland leading 1-0 at the break. Thoroughly enjoying myself. https://t.co/PC6tyirbS7