- Sources: primary, discussion
- Summary: Anthropic states that a parallel agent run, roughly 950 agents and 210 million tokens over 21 hours, gathered over 200,000 reverse transcriptases, picked out 3,500 candidate systems, and narrowed those to the 20 most compelling, surfacing a reverse-transcriptase system adjacent to a CRISPR-like repeat array. Anthropic states that the underlying reverse transcriptase had been identified in previous studies and that Claude was first to notice the system's defining features, and it reports running its own lab experiments. The system's biological function is stated as unknown and the work is a preprint.
- Comments: HN commenters questioned the lack of peer review and independent wet-lab validation behind the announcement, and noted the search was scoped to a tractable target rather than an open problem.
- Why it matters: The published run figures, roughly 950 agents and 210 million tokens over 21 hours to go from over 200,000 surveyed reverse transcriptases to 20 candidates, are a concrete cost and yield datapoint for large parallel agent work.
- Follow-up: Watch for independent wet-lab validation and any peer-reviewed characterisation of the system's function.
send feedback on this story