Gemini 4 Argon: The Model and Safety Cards...

Gemini 4 Argon is much more capable, including in cyber and biological domains, and may approach thresholds for serious risks. Google has added safeguards and monitoring, but the author questions whether tests can reliably assess risks when the model often recognizes evaluations. Overall, the model shows better alignment and task performance, while its heightened evaluation awareness has unclear effects on safety.

ORIGINAL LINK: https://thezvi.substack.com/p/gemini-4-argon-the-model-and-safety

Anthropic ends Coding Interviews: "Hunger beats tenure" — Latent Space...

AI labs released major updates to models for world-building, coding, and on-device use. AI systems also claimed progress on hard math problems, but experts called for proofs that others can check. The article highlights rapid AI advances and the need for reliable evidence.

ORIGINAL LINK: https://latent.space/p/ainews-anthropic-ends-coding-interviews-hunger-beats-tenure/

Import AI 478: Math for alignment; ProgramBench; and Mirendil shows what RSI means...

AI agents are speeding up math, coding, and scientific research. One researcher used AI to build a coding model in just a few months. These advances could make AI development faster, but also make systems harder to secure.

ORIGINAL LINK: https://importai.substack.com/p/import-ai-478-math-for-alignment

Anthropic Files Its S-1...

The full text is finally here, so we are giving it the full RTFB of model card treatment.

ORIGINAL LINK: https://thezvi.substack.com/p/anthropic-files-its-s-1

How Anthropic ships the #1 LLM in the World safely: Separating the Brain from the Hands...

Katelyn Lesse and Angela Jiang join Latent Space to explain the engineering behind the Claude Platform: launching Opus 5.5, separating the model’s “brain” from the agent’s “hands,” scaling APIs, and building reliable managed agents. A new introduction ...

ORIGINAL LINK: https://latent.space/p/how_anthropic_ships_the_1_llm_in_the_world_safely_separating_the_brain_from_the_hands/

GPT-6.1 Terra...

GPT-6.1 Terra is a fast, mid-priced model with strong visual reasoning, computer use, and communication skills. It is cheaper than Sol but costs more than Luna. It may suit developers who need capable, affordable help with repetitive computer tasks.

ORIGINAL LINK: https://simonwillison.net/2026/Oct/26/gpt-61-terra/

The Sequence Knowledge - Issue #959: Learning RSI: How Models Learn to Direct Their Own Updates...

SEAL and the architecture of persistent adaptation

ORIGINAL LINK: https://thesequence.substack.com/p/the-sequence-knowledge-issue-959