Bob M

Anthropic IPO and recursive self-improvement: Claude writes 80% of their code, Jack Clark predicts fully autonomous AI self-improvement by end of 2028. Recursive self-improvement is when an AI system takes a goal like "build a better version of yourself" and executes it without humans in the loop. Not

Key Takeaways - GPT-5.5 cracked a misconfigured Firebase app 7 out of 10 attempts during a $1,500 red-team test across 13 LLMs. Gemini refused to try. - The attack reuses google-services.json extracted from a built APK to query production Firestore directly. No zero-day, no sophisticated bypass. Just

A worm built by University of Toronto researchers in June 2026 ran on a free open-weight model and spread through a simulated corporate network without a single human pulling strings. It seized control of each machine, cloned itself, and moved to the next one. Nobody typed a command. The entire

Key Takeaways - Gemma 4 12B uses an encoder-free architecture. Raw image patches and audio waveforms project directly into the LLM embedding space, no separate vision or audio encoder - Memory requirements: 26.7 GB in 16-bit, 13.4 GB in 8-bit, or 6.7 GB in 4-bit. Quantized versions

Key Takeaways - Gemma 4 12B uses an encoder-free architecture. Raw image patches and audio waveforms project directly into the LLM embedding space, no separate vision or audio encoder - Memory requirements: 26.7 GB in 16-bit, 13.4 GB in 8-bit, or 6.7 GB in 4-bit. Quantized versions

MAI-Code-1-Flash outperforms Haiku 4.5 by 16 points on SWE-Bench Pro. Ships as default in VS Code. No opt-in required. That's what happened at Build 2026. But the coverage missed the bigger shift underneath. Microsoft dropped seven MAI models. Trained from scratch. No distillation from OpenAI, Anthropic, or