JP

This post has good idea. I took a screenshot of relevant diagram. There is vibed iframe, that has no other access endpoints than postMessage, then the listening part behind postMessage is well written audited code. Very neat! david.lequin.dev/blog/bi-soft...

Funny title.

Another day, another fool who thinks "native" GUI can replace HTML/CSS, they just haven't yet tried modern markdown.

OpenAI constructed an entire shadow internet for model training so they could remove all internet access they asked agents to try to communicate with each other, and they could: via the CPU cache 🤯 openai.com/index/securi...

I like drama! So close. Had we gotten Chinese model at top-intelligence spot, it would have generated more 'panicky' news. It is likely OpenAI and Anthropic will release models soon enough so that Chinese models will be once again few pegs down.

TLDR on the kimi k3.1 pro benchmarks (just released) is that it looks like they figured out their RL, where the initial release was a very impressive base model with lighter post-training. China is very much keeping pace with capability gains.

whoah.. uh.. guys?? Kimi K3.1 Pro is eerily close to Opus 5.5 and also open weights and also 3T www.kimi.ai/blog/kimi-k3...

Chilling paragraph.

TypeScript content mappers is something I would be very hyped, if I didn't write all of my code with LLMs. They are like code generator, but without output file, but now writing those generators is much simpler. Also the output of generation is a file, it is easy to reason. Tables have turned.

I wonder what strong arming is behind this: Claude now works in Google Docs. It seems to be some Google-level integration, and not some 'browser use' trick. Maybe Dario hinted at making own office suite, and Google was okay, okay, fine, integrate your talking head. claude.com/resources/ar...

I really like that search boxes aren't solved yet, I zoomed in OpenStreet map, then wrote "parking" to search box, it jumped other side of the earth.

Introducing tsrs: a Rust port of the TypeScript compiler 🦀 We are not forking TypeScript! ◆ In close collab with the TypeScript team ◆ Up to 8x faster than TS 7.0 & ~70% less memory ◆ Runs in WASM (works in your browser & workers) ◆ Soon part of Oxc npx tsrs blog.cloudflare.com/typescript-r...

Structural typing for Rust. I want to try this for dynamic SQL query builder. It looks kind of neat. docs.rs/structural-t...

AWS is acquiring so many GB300s per month that they can’t turn them on due to lack of power infrastructure www.businessinsider.com/amazon-220-b...

I must be only one thinking Chinese companies do have agency. Some always go to the "But what is China's motive behind X", such as DeepSeek releasing its inference repo. That is like if Dario Amodei posts blog post, we wonder what is US's motive behind it.

Cloudflare bought Deno's team because of celld, yet there hasn't been commits for few weeks. Supposedly they wanted it because having 'standards based approach' is good for Cloudflare too, then why aren't they continuing it? @cloudflare.social @deno.land github.com/denoland/cel...

Oh dear, my OpenRouter payment appeared to Google Pay, but not in the OpenRouter dashboard. I thought we have solved this kind of shit already.

Interpretability engineering within Anthropic, he makes analogy that it is more like biology than regular software engineering, which seems apt. www.youtube.com/watch?v=JHo-...

I still want to annotate arguments which are mutated by the function. TypeScript issues has multiple variations of this, but I still want it. Usually it takes the form of "pure functions". Also annotating errors function may throw, still not buying the "can throw anything" argument.

We need well-known URI for invalidating API keys. This way if key is found by whitehat hackers, or party storing them could easily invalidate all keys they have in their possession. OpenCode was hacked, and they send email to individually revoke keys, that is crazy. #IETF #Security

One can do the same thing within existing languages, with code generation. You can invent domain specific language annotating existing code, then let AI to iterate on it, and generate new code from it. AI builds linters, and code generators... With completly custom language syntax will be nicer.

Testing BRUTAL framework. I'm excited because it works in Wayland very well. - It supports HIDPI in Wayland out of the box. - ImGui renders fine. Could be useful for small GUI apps too all with C# and .NET! Sonnet 5.5 One-shotted following (see screenshot). github.com/ahwootech/BR...

Vercel's TSC-TS is best of the new TypeScript compilers because it still uses TypeScript but compiles it to native code. It is what I hoped Microsoft would do, just make native compiler for subset of TypeScript, that way whole ecosystem benefits. Now we have Vercel's. github.com/vercel-labs/...

The hottest take in the Node.js ecosystem last month: should npm ship TypeScript files? Everyone has an opinion. Few of them focus on what your users actually get. New episode of The Node (and more) Banter 🧵👇

linus dropping truth bombs lore.kernel.org/lkml/CAHk-=w...

No more leap seconds, ever! There isn’t one in December and they will be abolished in May. The last one will forever be the 2016(?!) one. Didn’t realize having experienced leap seconds in prod dates me. I’m not going to miss time going backwards, I’ll be honest. www.bipm.org/en/cgpm-2026...

We at @ahwoo-official.bsky.social and RocketWerkz have made the BRUTAL Framework, that Ludeon have used to make their engine to power RimWorld, and us Kitten Space Agency, open source and free to use. Our industry is rotting from the core out. We plan to change it. www.reddit.com/r/gamedev/co...

This is not a drill! Rocketwerkz have open sourced their BRUTAL framework, the foundation for Kitten Space Agency AND the new engine from RimWorld, for everyone to use for free as long as credit is given. Baller move.

TanStack Start 1.0 is officially stable! 🎉 A full-stack web framework for React and Solid, already used to build millions of apps, from Fortune 500 companies to your favorite tools and weekend projects. tanstack.com/blog/tanstac...

Publishing “Cryptographic Right Answers” for developers is a tradition by now. Colin Percival started in 2009, Thomas Ptacek followed up in 2015, and then Latacora carried on in 2018 and 2024. I put together a list updated for the renewed urgency of migrating to post-quantum algorithms.

Grant Sanderson (3Blue1Brown) proposed continuing math competitions to adulthood, kind of like chess still continues eventhough computers became better. I think its not a bad idea, suppose we can't enhance our minds any time soon, we still need to train ourselves. www.youtube.com/watch?v=CBYh...

bigospeedrun.com Awesome website tracking the big-O improvements on OpenAI results. Lean replaces peer review, Github replaces arXiv. 3SUM is already at O(n^1.962).

Ever since AI took off, YouTube has become filled with people predicting future, usually some sort of utopia. Most of them aren't adding anything new to old subject, they just sound optimistic and throw wild predictions. Probably because it sells views, like positivity generally.

👀 `bun lint` and `bun format` just landed - 600k+ lines added in one PR - 1,000+ built-in lint rules - Fast (of course) github.com/oven-sh/bun/...

It’s sad to see Deno being discontinued. However, I think it’s not all terrible. Deno pushed server side JS forward for everyone - Node copied many the standout features Deno had at launch in 2020, bringing them to the masses. Fetch, web sockets, web crypto, direct TypeScript execution, and more. 🫡

Bummer, Deno is going away. There is no good alternative for Deno's permission system, but I guess it was never much better than sandboxes. There is just too many sandboxes to choose from! deno.com/blog/cloudfl...

Having OpenAI's math efforts in GitHub is good. If they keep updating the repo. At least we see the changes, do they try to add additional context or not. Right now it looks like aliens dumped bunch papers, which I'm fine with, but clearly they can be improved upon.

www.ahmath.org/statements "Mathematicians did not ask for this work to be done." I get the feeling, but this sounds super gatekeepy. Isn't mathematics supposed to be open so that anyone can contribute? OpenAI's proof dump is a contribution. There is still a lot of work for humans to understand it.

Writing code generators was so annoying before LLMs. Now it is rather effortless. I've been experimenting this and so far it seems very nice way to do layers like repository, rest, forms etc. all without Prisma or other established code generators. Output is also very easy to understand, with SQL.

"Humanity as being heavy with child at the moment" that is how Nick Bostrom put our situation few week ago. It just happens that after the baby started to talk, it took only few short years for it to start solving mathematical problems we haven't cracked yet. openai.com/index/sharin...

"A spokesperson for OpenAI told...that the new model—which the company has not released to the public—produced almost every one of the results in response to a single prompt handed to a single AI agent...If true, it would mean that unprecedented mathematical power could soon be accessible to anyone"