a look at health benchmarks in the gpt-6 astra and fable 5.1 model cards, incremental gains, and questions about the judges.

Research CV of Gal Sapir, Staff Research Scientist at Pheno.AI, working on health AI, foundation models, clinical evaluation, and research systems.

on agents making it easier to keep working after my judgment starts to fade, and a simple rule for stopping.

Worker Automation, RL as a Service, Anthropic's next big bet, GDPval and Utility Evals, Computer Use Agents, LLMs in Biology, Mid-Training, Lab Procurement Patterns, Platform Politics and Access

Why selling data to train models is not as easy a decision as you think for bio companies

Vibecoding and the possible collapse of a shared language.

A unified benchmark for evaluating hierarchical reasoning capabilities of LLMs over general health time series.

benchmarks as a way to turn messy domain data into a measurable, optimizable substrate for models.

a loose reading list around agents, memory, benchmarks, and the hard parts of making models useful outside language and code.

We introduce PEEK, a system that caches reusable orientation knowledge about a recurring external context as a small, prompt-resident context map.

The Darwin Gödel Machine: AI that improves itself by rewriting its own code

Claude Code plugin: deep project interview command that produces actionable specs before implementation - galsapir/skills

the curation regression, the openness trade-off, and what a substrate worth evaluating against would actually need: on Medmarks.