mattias_akke

co-founder & CEO Halmos Labs (YC F26) | AI for science and open ended research

[mattias_akke @MattiasAkke]

@begimher @cristian_le0 @Armorer_Labs strong team

[mattias_akke @MattiasAkke]

3 open questions for auto-research agents in biology:
- identifying, quantifying and overcoming biases in agents
- deriving methods for verifying scientific reasoning and conclusions
- self-improvement on swarms where signal is scattered across subagents.

[mattias_akke @MattiasAkke]

~15% of CDMO leaders rank fear of AI slop in pharma above tariffs, layoffs, and talent shortage in life sciences

[mattias_akke @MattiasAkke]

@halmoslabs Come for the refreshments.

[mattias_akke @MattiasAkke]

@drecmb What if anyone could generate any molecule?

[mattias_akke @MattiasAkke]

@DavidStalmarck I’ll make one

[mattias_akke @MattiasAkke]

Frontier models are not getting better at balanced exploration and optimization. They remain firm in prioritizing value seeking strategies. A good scientist need to balance these

[mattias_akke @MattiasAkke]

SF is strange. You get: Waymo, robot cleaner. You loose: stable data connection, synced traffic lights

[mattias_akke @MattiasAkke]

Established biotech has been colliding with new upcoming ai-native startups the past month. Is it a sign of fear as more people can innovate in the space, or are we going to see a large number of the newcomers fail to capitalize on ”slop hypotheses”?

[mattias_akke @MattiasAkke]

@DavidStalmarck @halmoslabs @linusvekensteen Great having you

[mattias_akke @MattiasAkke]

@DavidStalmarck @halmoslabs @linusvekensteen Really interesting product. Could see it used in the lab in the future

[mattias_akke @MattiasAkke]

@SoojungYang2 🙏

[mattias_akke @MattiasAkke]

@Time_Spring_Hoo @garrytan A great hypothesis only needs verifying once.

[mattias_akke @MattiasAkke]

@tangming2005 @garrytan We’d be happy to jump on a call and explain how we do it! Send me dm

[mattias_akke @MattiasAkke]

@jatingargiitk @garrytan Verifiabillity, interpretabillity, and an ability to go unbounded while working with proprietary data

[mattias_akke @MattiasAkke]

@OperatingByJohn @garrytan Exactly! That’s why experiment design and multiverse analysis is critical

[mattias_akke @MattiasAkke]

You ask hard question. Hard question doesn’t have answer. We derive the answer anyway

[mattias_akke @MattiasAkke]

@DeryaTR_ @eboyden3 About time

[Garry Tan @garrytan]

AGI Science Loops are coming, and there will be Muse/Instinct for those AGI Science Loops. Halmos is building that.

[mattias_akke @MattiasAkke]

@OPuzanov @garrytan They fail in poor scientific taste, failure of LLMs to make induction and wasteful spending from poor validation. This is why we exist