Sinuhet

@thekitze I thought you can have the Revolut app in jusr one mobile device...

trying to make a large bank transfer from revolut (i've enabled face verification for bigger amounts, not face id but a photo scan) android front cam: sorry too blurry.. too blurry.. find better lightning... glitches tries to scan.. sorry too blurry *grabs iphone*: 2 sec scan, sent smh my head

@theo A genuine question - would you do the same if Anthropic decides to ban you & your company to their models for life? Many people are scared to criticise these big companies as they want to have options to use/come back to their services.

Always amazed by the people who dm me in support of the anti-Anthropic stuff. Feels like half the industry is scared to poke the bear. Thankful I am in a position to fight back on the bullshit. Will do my best. Sorry in advance if we get Claude banned on T3 Chat 🙃

@theo You do it right tbh. To publicly display if and how some of these companies have ridiculous TOS while changing them randomly and chaotically overnight is the way to go. Sadly this is common behaviour now - lately @windsurf introducing new terms (daily & weekly limits) overnight

@Rainmaker1973 The study is in vitro! There are so many successful studies in vitro they never worked in humans...

Scientists just made 50-year-old skin cells behave like they’re 20 again. Researchers at the Babraham Institute in Cambridge have developed a groundbreaking method to reverse the biological aging of human skin cells by approximately 30 years, all while keeping them as fully functional adult skin cells. The team used a carefully controlled, short-term version of the Nobel Prize-winning Yamanaka reprogramming technique. By exposing adult skin fibroblasts to a specific set of reprogramming factors (the Yamanaka factors) for just 13 days—and then stopping the process early—they successfully “reset” many molecular markers of aging without pushing the cells all the way back to a stem cell state. After this brief treatment, the rejuvenated cells displayed a dramatically younger profile: their epigenetic clock (a measure of chemical tags on DNA that tracks biological age) and their gene expression patterns (the transcriptome) closely resembled those of cells from much younger individuals. Even more impressively, the cells behaved younger too. The treated fibroblasts produced significantly higher levels of collagen—the protein essential for skin firmness, elasticity, and wound healing—and they migrated faster to close an artificial “wound” in laboratory dishes compared to untreated older cells. The researchers also observed reversal of age-related changes in genes associated with diseases such as Alzheimer’s and cataracts, suggesting the technique could have broader therapeutic implications. While this work is still in its early stages and the precise mechanisms remain under investigation, the findings open exciting possibilities: one day, scientists may be able to selectively rejuvenate aging cells in the body to enhance tissue repair, improve healing, and potentially slow or mitigate some effects of age-related diseases—without the risks associated with fully reprogramming cells into stem cells. [Gill, D., Parry, A., Santos, F., Okkenhaug, H., Seale, M., Dobbs, L. J., Reik, W., & Ocampo, A. (2022). Multi-omic rejuvenation of human cells by maturation phase transient reprogramming. eLife, 11, e71624. DOI: 10.7554/eLife.71624]

@robinebers I am afraid I will have to agree. No reason to replace a monthly limit with daily/weekly limits. I am/was a happy user as compare to other IDEs I did not have to babysit the agentic loop and could go away for 30 - 60 min. Any idea who offers monthly limits with decent harness?

Windsurf is finally dead the last reason people actually used it was price this will be the last straw, and the rest will switch off too really sad to see a once great competitor become so irrelevant in such a short time

@windsurf Hi @theo. Once your shoulder is ok could you pls. look into AI coding subs landscape? As you can see Windsurf sadly decided to copy Cursor and start applying daily & weekly limits and I am afraid people like me who code intensively for a week/month have no place to go anymore.

We're simplifying Windsurf pricing across Free, Pro, and Teams alongside launching a new Max plan for our power users. The new plans replace credits with industry-standard daily and weekly quotas. For the majority of users, this will be enough to fully cover all agent usage. If you’re a paying subscriber, your price isn't changing, and we're including a free extra week to try the new system before you commit.

@jeffwsurf A fair approach would be to give more than one day heads up. My 500 credits over the remaining 3 weeks of my monthly plan will be converted tomorrow and I will not see proper outstanding usage and the agent will stop in the middle as it is happening with Antigravity

As agents become longer running and output tokens become more expensive than before, we had to make changes on how we calculate multipliers. Rather than being forced to increase model multipliers to 20x+ and draining accounts quickly, we are aligning to the industry standard quota system, which refreshes daily and weekly so you can keep using Windsurf over time. Most users won’t reach their quota

@jeffwsurf If you want to remain competitive then swapping to "industry standards" might not be the right move as many users - like me - who coded intensively for a week per month (and used the 500 credits of pro plan) will start looking elsewhere else.

@windsurf It is a bad move. Windsurf was one of the last platforms with monthly limits. For people having a second job like me and who code intensively just a week per month, I will not be able to use my plan in full. I guess I have to start looking at an automatic agent harness...

@Aaronontheweb @TextForgeApp Thx for the article. As you are clear the app has to pass CASA2 audit, do you have similarly minimal requirements for your PRD? Aka at what point you decide "it's a coding time"? Do you have a checklist for a complete PRD?

New Blog Post: "Software 2.0: Planning and Verifying a Greenfield Project" Been asked about how I use Claude / Codex for actually delivering completed software. And in this case, I mean a commercial web application that passed a CASA2 audit, not a toy CLI. This is how 1/3

@MMT_Official_ What charting library do you use?

TradingView is dead. Proof us wrong.

@waynesutton @convex @karpathy Do you recommend using a schema less Convex during development and create schema once all features of your project are implemented?

To build your agentic org, start by running npm create @convex and call it a day. Reactive database Functions as code. Agents as org structure Everything in sync ... @karpathy

@jerryjliu0 Can you share the prompt & extraction instructions? I did some extraction experiments and could never reach such high accuracy. So I guess it was mainly my wrong prompt approach...

I parsed OpenAI’s tax filings for fun 📋 I used the “Extract” capability in LlamaCloud to automatically extract out all relevant fields into structured JSON outputs. You can see their mission statement “ensure that artificial general intelligence benefits all of humanity” with the corresponding bounding boxes. (Inspired by @simonw’s post on this recently: simonwillison.net/2026/Feb/13/op…) The extraction is powered by our core OCR engine, LlamaParse. LlamaParse is able to reconstruct this complex form PDF into markdown tables that captures each cell with ~100% accuracy. Check it out: cloud.llamaindex.ai

@karpathy I like DeepWiki. I just would wish it supports private repos and gets automatically updated as your codebase changes. Btw last time I tried it I discovered all DeepWiki MCP calls were publicly available links... I hope this has changed already.

On DeepWiki and increasing malleability of software. This starts as partially a post on appreciation to DeepWiki, which I routinely find very useful and I think more people would find useful to know about. I went through a few iterations of use: Their first feature was that it auto-builds wiki pages for github repos (e.g. nanochat here) with quick Q&A: deepwiki.com/karpathy/nanoc… Just swap "github" to "deepwiki" in the URL for any repo and you can instantly Q&A against it. For example, yesterday I was curious about "how does torchao implement fp8 training?". I find that in *many* cases, library docs can be spotty and outdated and bad, but directly asking questions to the code via DeepWiki works very well. The code is the source of truth and LLMs are increasingly able to understand it. But then I realized that in many cases it's even a lot more powerful not being the direct (human) consumer of this information/functionality, but giving your agent access to DeepWiki via MCP. So e.g. yesterday I faced some annoyances with using torchao library for fp8 training and I had the suspicion that the whole thing really shouldn't be that complicated (wait shouldn't this be a Function like Linear except with a few extra casts and 3 calls to torch._scaled_mm?) so I tried: "Use DeepWiki MCP and Github CLI to look at how torchao implements fp8 training. Is it possible to 'rip out' the functionality? Implement nanochat/fp8.py that has identical API but is fully self-contained" Claude went off for 5 minutes and came back with 150 lines of clean code that worked out of the box, with tests proving equivalent results, which allowed me to delete torchao as repo dependency, and for some reason I still don't fully understand (I think it has to do with internals of torch compile) - this simple version runs 3% faster. The agent also found a lot of tiny implementation details that actually do matter, that I may have naively missed otherwise and that would have been very hard for maintainers to keep docs about. Tricks around numerics, dtypes, autocast, meta device, torch compile interactions so I learned a lot from the process too. So this is now the default fp8 training implementation for nanochat github.com/karpathy/nanoc… Anyway TLDR I find this combo of DeepWiki MCP + GitHub CLI is quite powerful to "rip out" any specific functionality from any github repo and target it for the very specific use case that you have in mind, and it actually kind of works now in some cases. Maybe you don't download, configure and take dependency on a giant monolithic library, maybe you point your agent at it and rip out the exact part you need. Maybe this informs how we write software more generally to actively encourage this workflow - e.g. building more "bacterial code", code that is less tangled, more self-contained, more dependency-free, more stateless, much easier to rip out from the repo (x.com/karpathy/statu…) There's obvious downsides and risks to this, but it is fundamentally a new option that was not possible or economical before (it would have cost too much time) but now with agents, it is. Software might become a lot more fluid and malleable. "Libraries are over, LLMs are the new compiler" :). And does your project really need its 100MB of dependencies?

@pvncher @RepoPrompt Thx Eric. The new onboarding is in the app or on your RP-web/YT? Can a user see/edit the RP Agent instructions?

Just released @RepoPrompt 2.0! - New built in Agent mode, making full use of the RP MCP tools - First class support for Codex, leveraging it's app server. - Good support for Claude Code + Gemini CLI as well! - Brand new onboarding making the app easier than ever to use.

@cboyack Basically saying "Trust me, history will repeat itself and we will be all fine, new industries will emerge". If your article should provide some hope, you should at least speculate what new industries will emerge. Even @elonmusk has no clue and is predicting universal income.

@OfficialLoganK The limits for PRO in Antigravity are so low that I had to cancel my subscription. Consider making only a monthly limit like Windsurf has. Are you finally planning to show paid subscribers the number of tokens available & used?

Antigravity + Google AI Studio Stay tuned for next week : )

@OfficialLoganK Just fix the agent (a lot of babysitting + interrupting errors) in @antigravity and Google's Studio (single file app by default, terrible tool calling, often claiming changes applied but the opposite is true). Being able to import repo in the studio would be +

@IGClientHelp where is the pdf list of ISA and SIPP eligible shares? 25 min to get connected to South African staff who did not know where is it. Ig chat had it wrong tool and still waiting to get connected to human being after (35 min!). Your page is worse than vibe coded...

@dabit3 If everything is so easy why do so many things take so long to implement in @windsurf for example? It took more than 6 months to fix that while swapping cascade it chose the last used model per Windsurf and not the particular cascade. Codemaps still no delete button and so on...

Local agents are great for solo work, but remote / cloud agents scale to entire teams and across all devices. Anyone can tag an agent in a chat or ticket and ship a change without touching a terminal, or even knowing how to operate one. Also, one interface across every repo so you don't need a separate terminal tab per project, or a separate local checkout, or even a separate environment. Conversations are tranparent and flow across multiple interfaces - Slack, Linear, Github, the @DevinAI web app... Accessible from any device or phone, so you don't need your laptop, make updates via your mobile device or even voice. Just tag the agent in a group chat to fix some outdated docs or pastes in a feature request from @x (see screenshot) or calls out a bug, and a PR shows up few minutes later. This is the future of coding agents and is also how we use @DevinAI to build @DevinAI - a more in-depth writeup coming Monday!