Claude Code Leak Reveals Always-On ‘Kairos’ Agent - theinformation.com
After Anthropic released Claude Code's 2.1.88 update, users quickly discovered that it contained a package with a source map file containing its TypeScript codebase, with one person on X calling attention to the leak and posting a file containing the code. The leaked data reportedly contains more th
Here's what that Claude Code source leak reveals about Anthropic's plans
A persistent agent, stealth "Undercover" mode, and... a virtual assistant named Buddy?
Ask HN: How do you get LLMs to stop spewing corpo speak?
OpenAI president on AGI: • "I'd say I'm basically like 70, 80% there. So I think we're quite close." • "I think it's extremely clear that we are going to have AGI within the next couple years."
Greg Brockman was asked if he agrees with NVIDIA's CEO that AGI is already here.
The Bonsai 1-bit models are very good
Hey everyone, Tim from [AnythingLLM](https://github.com/Mintplex-Labs/anything-llm/issues) and yesterday I saw the [PrismML Bonsai](https://prismml.com/news/bonsai-8b) post so i had to give it a real shot because 14x smaller models (in size and memory) would actually be a huge game changer for Loca
Anthropic took down thousands of GitHub repos trying to yank its leaked source code — a move the company says was an accident
Anthropic executives said it was an accident and retracted the bulk of the takedown notices.
Gemini 4 is coming ??
Mac support for external Nvidia GPU available now through TinyGPU
Running SmolLM2‑360M on a Samsung Galaxy Watch 4 (380MB RAM) – 74% RAM reduction in llama.cpp
I’ve got SmolLM2‑360M running on a Samsung Galaxy Watch 4 Classic (about 380MB free RAM) by tweaking llama.cpp and the underlying ggml memory model. By default, the model was being loaded twice in RAM: once via the APK’s mmap page cache and again via ggml’s tensor allocations, peaking at 524MB for a
attn-rot (TurboQuant-like KV cache trick) lands in llama.cpp
80% of the benefit of TQ with almost no downsides. Q8 is now ≈ F16
OpenAI “internal model” solved 3 more Erdős problems
Link to tweets: https://x.com/mehtaab\_sawhney/status/2039161544144310453?s=20 https://x.com/kevinweil/status/2039200605672284572?s=20 Link to paper: https://arxiv.org/pdf/2603.29961
Meta Adaptive Ranking Model: Bending the Inference Scaling Curve to Serve Models
Lemonade by AMD: a fast and open source local LLM server using GPU and NPU
Opus 4.6 vs. GPT 5.4 – an opinionated take for my diary
New laws to make it easier to cancel subscriptions and get refunds
Can we block fresh accounts from posting?
Flood of useless vibe coded projects is getting out of hand...
Men are ditching TV for YouTube as AI usage and social media fatigue grow
Qwen3.6-Plus
Blog post: [https://qwen.ai/blog?id=qwen3.6](https://qwen.ai/blog?id=qwen3.6) From Chujie Zheng on 𝕏: [https://x.com/ChujieZheng/status/2039560126047359394](https://x.com/ChujieZheng/status/2039560126047359394)
A Founder Tried to Pitch – and Got a Restraining Order
Omnivoice - 600+ Language Open-Source TTS with Voice Cloning and Design
OmniVoice is a state-of-the-art zero-shot multilingual TTS model supporting more than 600 languages. Built on a novel diffusion language model architecture, it generates high-quality speech with superior inference speed, supporting voice cloning and voice design. **Key Features** \- **600+ Languag