Intology's Locus auto-designs molecular models, beating GPT-6 and Claude agents
rohanpaul_ai · x · 2026-11-04
Intology's automated research system Locus does a machine-learning researcher's job: proposes experiments, runs many batches in parallel on GPUs, keeps what works and iterates—now applied to drug and antibody design.
In tests, Locus beat GPT-6 Astra and Claude Opus 5.5 agents on molecular prediction. Settings included training molecular property prediction models on a proprietary task, fine-tuning Boltz-2 for small-molecule potency, and building antibody design workflows around BoltzGen, Boltz-2, ESMC, and ESMFold2—often out-of-distribution tasks where Locus's models outperform expert SOTA solutions.
More from AGI Musings
- AI makes chaos-scale hacking cheap, warns dev of lost stabilizing competence — harris_edouard · 2026-11-04
- AI removes the competence barrier that kept national-scale cyberchaos in check — harris_edouard · 2026-11-04
- Hinton: humans aren't reasoning machines, we're 'great big analogy machines' — FlorianGallwitz · 2026-11-04
- The irony: fearing AI power concentration, yet letting governments gatekeep AI — VraserX · 2026-11-04
- 'Schrödinger AI' meme mocks anti-AI crowd holding contradictory beliefs — FlorianGallwitz · 2026-11-04
- Why algorithmic personalization is flattening culture into sameness: 'Temperature Zero' explained — acroll · 2026-11-04