Akash Ramdas

[Akash Ramdas @Akash__Ramdas]

@shbhmndshwr7 We're optimistic :D!

[Akash Ramdas @Akash__Ramdas]

Working on this I was impressed by how capable this models were at this task. It was far more than I expected when we started out. We're excited to share a benchmark of this today. Check out https://t.co/qaiZvhKPTd for more details!

[Advaith Sridhar @advaith_sridhar]

In Material Discovery Bench, we give frontier models an open-ended material discovery problem - finding new dielectrics for chips. Models are equipped with everything a competent PhD student would have access to - atomistic simulation software, a coding environment, property prediction models (MLIPs) and web search. Each model run is long-horizon (100M tokens!), and models are allowed to make unlimited submissions over the run.

Sol, Fable, Opus and Kimi-3 are surprisingly capable at this task, generating hundreds of unique materials that are stable and show good thermal, dielectric and mechanical properties.

[Akash Ramdas @Akash__Ramdas]

@hangyosaibaa @advaith_sridhar @discoveredmat Thank you! :D

[Akash Ramdas @Akash__Ramdas]

@saumyagandhi007 @advaith_sridhar @discoveredmat There is some overlap, and it's been quite interesting to see. Notably in one of the runs the models tried to use the Debye temperature as surrogate, which isn't too far off from something I may have tried!

[Akash Ramdas @Akash__Ramdas]

Maybe the best approach to discovering materials is to take short breaks and not burn out 🤷���♂️ (of tokens)

[Advaith Sridhar @advaith_sridhar]

@discoveredmat OpenAI models are far less likely to reward hack like the above, but seem to unravel over longer task horizons. Here’s a snippet of GPT-5.6-terra confusing crystal relaxation with taking a break (relaxing).

[Akash Ramdas @Akash__Ramdas]

Check out our work on finding new materials with @advaith_sridhar at @discoveredmat!

[Advaith Sridhar @advaith_sridhar]

Today, we're introducing @discoveredmat . We build AI scientists that discover new materials for semiconductor chips.

We’re starting by releasing our work - hundreds of new materials discovered using frontier AI models, along with our benchmark for tracking progress in this domain, Material Discovery Bench.

To pursue our mission, we have raised $9 million from @LightspeedIndia (lead investor) , @ycombinator , @peakxvpartners , @paulg , @gokulr , @trq212 and many others.

🧵

[Akash Ramdas @Akash__Ramdas]

Excited for this new journey (and X account) of materials discovery @matforge_ with @advaith_sridhar!

[Y Combinator @ycombinator]

Matforge (@matforge_) is building AI scientists to discover new materials for the semiconductor industry.

The semiconductor industry needs better materials to bring back Moore’s Law, but finding novel materials today takes 10+ years of lab work. Matforge wants to accelerate this process.

Congrats on the launch, @advaith_sridhar & @Akash__Ramdas!

https://t.co/TBOPo66zhm