Claude discovers a novel enzyme system with CRISPR-like repeats
Anthropic’s claim that its Claude AI helped identify a previously unnoticed enzyme system with CRISPR‑like features has prompted both excitement and skepticism. Commenters see it as a promising example of AI-driven pattern finding in huge biological datasets, but stress that the result is early, modest, and heavily dependent on human wet‑lab work, making the “Claude discovered…” framing feel more like marketing ahead of an IPO than a scientific breakthrough. The exchange broadens into worries about AI labs doing in‑house biology without strong oversight, the risk of dual‑use bio capabilities, and whether frontier models will evolve into closed “everything companies” that capture both scientific credit and economic value.
Scope & significance of the discovery
- Many see the result as technically interesting but modest: an AI-guided bioinformatics search found a previously undescribed RT‑associated repeat system, with limited wet‑lab validation (short RNAs expressed; function unknown).
- Several biologists in the thread frame this as “early lab-meeting result,” not a CRISPR‑level breakthrough. True impact depends on future functional work and applications.
- Others counter that automating this kind of discovery on large genomic databases is itself a big step, even if this particular system ends up being biologically unremarkable.
Marketing, framing, and attribution
- Strong criticism of the headline and copy for anthropomorphizing “Claude,” downplaying human scientists, and leaning on “CRISPR-like” framing to trigger hype.
- Some argue the honest description is: agents ran large-scale computational searches; humans designed the harness, chose the problem, and did the lab confirmation.
- Concerns that AI labs are turning incremental findings into PR assets, using preprints and blog posts as stock/IPO marketing rather than engaging normally with peer review.
AI capabilities, RSI, and singularity debates
- Split between commenters who see this as another data point in accelerating AI scientific capability and those who say it’s just large-scale pattern matching plus tools.
- Long back‑and‑forth on whether current “AI research interns” meaningfully change scaling laws or are constrained by data, compute, infrastructure, and human bottlenecks.
- Some argue we are inching toward recursive self‑improvement; others say we’re still firmly in “humans orchestrating tools” territory.
Biology as a risky domain
- Multiple commenters flag bio as far more dangerous than math or code, raising worries about AI‑designed pathogens and automated wet labs.
- Others note this work used public sequence databases and low‑risk wet‑lab validation; danger is seen as more about future autonomous lab integration.
- Tension highlighted: Anthropic warns the public off bio use while internally pushing into AI‑driven drug discovery, which some see as hypocritical or power‑consolidating.
Business model, moat, and “everything companies”
- Widespread view that pure “model API” businesses lack a moat; labs are pivoting into verticals like biology, law, and security to own higher‑value IP.
- Some speculate labs will keep their most capable models internal, using them to generate proprietary scientific results and profit‑sharing deals, not just sell tokens.
- Others worry this creates an “AI aristocracy” controlling both compute and the frontier of scientific discovery.
Impacts on scientists and research culture
- Mixed reactions: excitement about productivity gains vs. concern that postdocs and domain experts will be further squeezed, underpaid, or displaced.
- Some stress that even if AI proposes candidates, human expertise remains essential to design experiments, interpret noisy biology, and ensure safety.