Sabitlenmiş Tweet
jack
4.8K posts
jack
@JackNotOld
In the land of the blind the one-eyed man is king
Katılım Şubat 2020
886 Takip Edilen1.4K Takipçiler
😅
Kimi.ai@Kimi_Moonshot
Congrats to the @cursor_ai team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support. Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.
ART
jack retweetledi
we didn’t think it was possible
after multiple years of R&D, the technology is finally here
Notion@NotionHQ
Heading 4 is finally here 😤 The years of “just bold the text and pretend” are over. Rolling out now.
English
I won't stop until all PDFs in the world are converted to MDs
Oliver Prompts@oliviscusAI
someone just open-sourced a tool that converts pdfs to markdown at 100 pages per second. 100% free. runs entirely on cpu. no expensive gpus needed.
English
spent a week doing this exact thing on SGLang with Qwen3.5-27B. 6 patches deep. the mamba_full_memory_ratio defaulting to 0.9 when only 25% of layers are recurrent was a particularly fun one.
hallerite@hallerite
story time > be me > told to get Qwen3.5 MoE LoRA inference working on vLLM > cool model > 397B params > 512 experts > gated deltanet, linear attention > wild stuff
English
jack retweetledi
The correct answer was "over $4.5B"
Susan Zhang@suchenzang
hypothetically, if yann lecun fundraised for a new AGI lab for himself... how much would it be worth?
English
I built a tiny macOS Markdown reader.
Double-click a .md file -> it opens as a clean, print-ready document.
Native app (<1MB), PDF export, live reload, secure rendering.
Show HN:
news.ycombinator.com/item?id=473150…
GIF
English
jack retweetledi
PI's robot can now make a grilled cheese without burning it. It has thus passed the Alfred Test, a higher bar than the Turing Test, because I still cannot do that reliably.
Physical Intelligence@physical_int
We’ve developed a memory system for our models that provides both short-term visual memory and long-term semantic memory. Our approach allows us to train robots to perform long and complex tasks, like cleaning up a kitchen or preparing a grilled cheese sandwich from scratch 👇
English
@taahajkhan @OtsoVeistera @gabriel1 @thetokenco Interesting. Do you see this being useful outside very large RAG contexts? My assumption is this is mainly valuable for enterprise workflows with massive documents (legal, compliance, research), rather than typical prompts.
English
@JackNotOld @OtsoVeistera @gabriel1 @thetokenco it varies by llm and compression level, e.g. at aggression 0.9, you can see faster responses for gpt-5.2 and opus-4.6 at around 15-25k input tokens
detailed benchmarks here: thetokencompany.com/benchmarks/lat…
English
You're wasting half your context window.
We’re launching @thetokenco (YC W26) today.
We compress LLM inputs before they reach the model.
Fewer tokens, lower cost, faster inference.
Models also perform better. In customer case studies we’ve seen a +5% lift in user purchases due to higher preference for outputs from compressed prompts.
The API is live. Link in the comments
English
@liaaip @PriyanshuP1405 Or because he had to go back to school after his internship ended
English
@JackNotOld @PriyanshuP1405 because lehman brothers collapsed and 2008 financial crisis accured :)
English
jack retweetledi
Anthropic said no to the Pentagon.
Now Sam Altman is backing them:
"For all the differences I have with Anthropic, I mostly trust them as a company and I think they really do care about safety."
OpenAI and Anthropic both drawing the same line.
This is a big deal.
English
In the near future we will look back autoregressive models and laugh, like we do at the pre-iPhone blackberry
Stefano Ermon@StefanoErmon
Mercury 2 is live 🚀🚀 The world’s first reasoning diffusion LLM, delivering 5x faster performance than leading speed-optimized LLMs. Watching the team turn years of research into a real product never gets old, and I’m incredibly proud of what we’ve built. We’re just getting started on what diffusion can do for language.
English
jack retweetledi
jack retweetledi
We extract nearly all (95.8%) of Harry Potter and the Sorcerer's Stone from Claude Sonnet 🤷🏻♂️
Anthropic@AnthropicAI
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
Unsurprisingly, the second order implication here is moats in AI are shifting from proprietary model weights into control of one’s ecosystem.
Anthropic@AnthropicAI
We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.
English
jack retweetledi
We estimate that Claude Opus 4.6 has a 50%-time-horizon of around 14.5 hours (95% CI of 6 hrs to 98 hrs) on software tasks. While this is the highest point estimate we’ve reported, this measurement is extremely noisy because our current task suite is nearly saturated.
English