λ @lambda_functor

MCP vs CLI for LLMs is the wrong discussion. People are debating at the wrong layer. Both work incredibly well as long as they are run through a repl like environment that can persist state. The funny thing about CLI is that CLI tools are usually accessed through BASH which happens to be a repl with persistent state (the file system), therefore cli works much better than mcp. But this isn't even close to being an intrinsic property of MCP. Just means that we need better harnesses. Right now the best repl for this are: - bash + fs - jupyter kernels - codemode type repls, usually quickjs LLMs understand the concept of forward evolving a system to solve a need very well. This comes from the agentic RL. They understand how to change the state of a codebase or acquire data from databases and then operate on it similarly to humans. But without an execution environment, they actually cannot do this properly. My bet? Sometime soon someone will (or has already?) create a embeddable, sqlite style mini execution environment that parses bash, typescript, or tool calls into a common IL execution plan that's easy to security check before executing. And design this specifically for durable execution environments. Then we will just connect cli, mcp, webmcp, whatever to that and it accepts any of the input modalities as they can all be represented as each other.

How is Jev so much faster than an LLM? It's not about how the model is trained - it's the inference technique. In fact, you can modify an inference engine to provide a performant Jev-like API with any open-weight LLM. Say you are trying to ask N multiple choice questions in parallel. In standard text generation, you decode tokens one at a time to generate a JSON representing those properties, and then parse the JSON. In a Jev-like system, what you can do is: - Label choices for each property 1,2,3,... - Prefill the shared information (output shape and other context) - Fork this inference context for each property, re-using that cached prefill - Add an individual question to each parallel context - Look at the logits (token probabilities) for each one, constrained to the valid choice labels - Map the labels back to the choice values. Include the softmax of the logit itself to get a combined choice+confidence score This means you decode one time in parallel across all questions, instead of one at a time - so the efficiency gains roughly scale with the number of questions per request (until compute is saturated). So what makes Jev's model unique? It's because they train a small model to perform well in this specific inference setting - improving both cost and capability compared to a non-specialized model. Ultimately though, if you have a use case Jev would be good for, you could do it with open-weight models too! (Some speculation made here about how Jev's engine might work, but this is one way you could get similar performance)

Diogo Almeida @CompleteSkeptic

6 days ago

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x

4K 8K 74K 38.3M 58K

I built a voice controlled computer-use for my mac using @typesafeai's Jev and it's INSANE how fast it is! I can dictate "open the notes app and create..." and the app opens before I even finish my sentence.

busy cooking

TypeSafeのJevで、ライブコメントを「質問・感想・要望・その他」に振り分けるデモを作ってみた、動かしてみるとイメージしやすい。 コメントの意味をAIが判定して、その結果で行き先が変わる。アプリやワークフローの小さな判断を、AIに任せる使い方とかに向いてそう。

More proof that they've turned an entire profession into an eval metric for marketing purposes

Rohan Paul @rohanpaul_ai

5 days ago

Sam Altman just said an internal post-Astra model can solve things the world's best mathematicians cannot. in his conversation with Mark Benioff (co-founder and CEO of Salesforce) "GPT 5.5 was maybe as good as like an average math professor. 5.6 was as good as like a maybe top

100 120 1K 344K 378

Procedurally generated FlashAttention with CuTe tilings, working on runnable CUDA code

I've been trying to think of what to say (if anything) about the recent discourse around Navier-Stokes, AI math, etc. I then realized that one of my heroes, William Thurston, already said everything that needs to be said—sixteen years ago: mathoverflow.net/a/44213

a few people asked me how i timed the raccoon animation on my portfolio! i basically asked claude to generate me a red line to visualized its path, and used @joshpuckett's dialkit to tinker around (adjusting the slope, peaks, etc.)

Anatomy of the Fresh TUI elements on screen. I rewrote the renderer as a data-first retained mode pipeline.

Was really interesting to hear John, Beren, and Charlie speculate about why Sonnet 5 and Opus 5 feel like worse models than GLM 5.3 (despite the fact that Anthropic can do raw logit distillation from Fable, and can also train Sonnet/Opus on the environments from which Fable was trained). Led to some interesting thoughts about value of distillation, what it takes to do distillation effectively, and what kinds of model behaviors are hard to extract from distillation.

I've asked Astra to read the DeepSeek v4.1 Flash paper and compare it to the original Transformer architecture in 3D - you can zoom in an inspect each element side by side. Things have changed quite a bit. Try yourself: …architecture.petergostev.chatgpt.site

This release is a major step toward making bb’s core customizable! On Omarchy, I made a bb plugin for Btrfs snapshots instead of git worktrees. Thread setup for the bb repo with packages installed went from ~15s -> ~1s. I haven’t seen another ADE that lets you customize things to this extent.

bb @get_bb_app

a week ago

bb 0.43.0 is out. New plugin APIs for custom workspaces, cloud machines, and browser control, with experimental Modal sandbox and Browser Automation plugins. Plus drag-to-nest threads, synced sidebar organization, and message editing out of experiments. getbb.app/changelog#0-43…

0 5 90 19K 34

又优化了一下 @lobehub ，清除页面文件之后完全冷启动，已经接近 Electron 的天花板了。 credit: @unixzii

CodeDB v0.2.5855: here's what we've been working on. Sponsored, zero-retention embeddings. A local OpenPuffer search index. And a fresh comparison with Graphify. Watch the walkthrough, then see the results below.

OPENAI NO YOU CAN'T LET GOOGLE WIN LIKE THIS

Tibo @thsottiaux

a week ago

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 and was a lot of fun, but usage has been declining and we have significantly better models now. Time to make room for the future.

2K 401 15K 3.1M 569

This is Knap. It's a new language I created that turns data into Markdown. The syntax should feel familiar and comes with wonderfully pleasant features to modify and format plain text. Knap is open source. Over a million people already use Knap directly or indirectly because it started as the templating language I made for Obsidian Web Clipper. Now any tool can use it. You can add Knap to your app or use it via CLI. Create a batch of Markdown files from JSON or CSV, or pipe Defuddle directly into Knap to generate Markdown from HTML/URLs. The Knap site is color-coded to help beginners understand the relationship between variables (blue), filters (orange), and logic (green). I've tried to document it as comprehensively as I can. Try the Knap Playground so you can see how fun the syntax is to use! Knap is pronounced /knæp/ (with a hard k). It's named after knapping (with a silent k): the process of shaping stones to form arrowheads, scrapers, and other tools. Enjoy!

The resolution of Navier-Stokes is an amazing accomplishment. Some comments … First, a brief mathematical postmortem. Millennium problems are important, but when you confront an unsolved problem, it is always unclear how far the solution is out of reach. The initial opinion of at least some experts (not myself – I have no expertise on this problem) seems to be that, in retrospect, the conceptual program underlying this solution happened to already be in place*, and the contribution of AI was a very intricate technical execution. Technical execution is one of the axes on which we've already seen AI become superhuman, though never before with such a dramatic demonstration. This is not to diminish the solution or OpenAI’s achievement. But there are many problems which are out of reach of current models, no matter how much inference-time compute one invests. I’m still convinced that human mathematicians are laying indispensable groundwork for future progress. *Terry Tao describes one possible program at mathstodon.xyz/@tao/117207851…. As I understand, the one Tao describes is NOT the Cordoba--Martınez-Zoroa program that OpenAI (and Alpoge/Buckmaster) followed. However, the successful approach was not as esoteric as Buckmaster’s statement might make it seem – there were certainly other people (not me) also prompting that approach; it just isn’t the kind of thing that one goes around saying publicly. As far as I know, Navier–Stokes is the only one among all Millenium problems with a well-defined program towards a solution. Concretely, for the Birch and Swinnerton-Dyer Conjecture (over number fields) I am not aware of even a vague proposal of what a proof might look like, so I doubt that current AI models will get anywhere close to a resolution. (My opinion is that this conjecture cannot be resolved in the negative; perhaps the most refined formulation is false, but then the real problem would be to correct the formulation and then prove it.) I'd be curious to see what progress was made on it!

24 Fields Medalists speak out against "A Severe Misalignment of AI in Mathematics" mathandai.org

The de-identification excuse is like: "no, I'm not front running my client if my system is designed to plagiarize and never attribute source"

Valerio Capraro @ValerioCapraro

2 weeks ago

Andreas’ mathstodon post:

keyboard_arrow_left Previous keyboard_arrow_right Next

22 242 2K 129K 422