Generative Selection

We build productive self-improving scientific intelligence to create beneficial technologies at the edge of human knowledge, starting with biological sciences

[Generative Selection @genselectai]

The first batch of AI-engineered proteins is designed and ordered, the first cells are growing! It is day 4 of our wetlab!

[Generative Selection @genselectai]

We’ve started experimenting with a more human-friendly delivery of our updates (insta & TikTok)! Gen Z & Gen alpha are now covered too!

https://t.co/HS0GjdKf7J

[Generative Selection @genselectai]

Inspired by the last week’s debates about the novelty of AI discoveries, we unleashed our internal AI for science engine to evaluate novelty in a transparent way. Enjoy our prior-art-based eval of a bunch of recent AI claims, be our guest! https://t.co/O8UM3G8LRH

[Generative Selection @genselectai]

We’ve got selected for 5050 by @fiftyyears this year — one of the best deep biotech program for founders out there!

[Generative Selection @genselectai]

Science is fundamentally a compression algorithm for reality that mitigates frontier models’ context limitations. It distills vast bodies of observation into compact causal models required to explain and predict the world. Smarter context engineering typical for science is key

[Generative Selection @genselectai]

With the current scaling laws of hardware and models, the top-tier LLMs would match a human scientist in 3-4 years, though still falling short of the superintelligent level.

[Generative Selection @genselectai]

The state-of-the-art models can fit up to 30-60 scientific papers in their context window, and effectively reason about them. It is still two orders of magnitude less than what an average scientist would learn throughout their career, and casually rely on in their daily work.

[Generative Selection @genselectai]

We believe the most productive setting for building scientific superintelligence is a product-focused, revenue-generating company.

For more detail, check out our white paper (https://t.co/FGJAI9AFzJ)