Ali Malik

Theory + AI + Education. Research @ OpenAI

[Ali Malik @AliRMalik_]

I'm excited to announce our work on MentalHealthBench, a new open benchmark developed with mental health experts to evaluate how Al models respond in realistic mental health conversations. 🧵

https://t.co/kz6iiETlwi

[Ali Malik @AliRMalik_]

Really grateful to have worked on this with some amazing collaborators at @OpenAI, especially @declangrabbmd and @thekaransinghal.

Special thanks to all the mental health clinicians who dedicated their time and expertise to making this a reality.

[Ali Malik @AliRMalik_]

I'm excited to announce our work on MentalHealthBench, a new open benchmark developed with mental health experts to evaluate how Al models respond in realistic mental health conversations. 🧵

https://t.co/kz6iiETlwi

[Ali Malik @AliRMalik_]

Our results show the steady progress of Al models in helping people navigate mental health situations.

We're releasing it so researchers can evaluate their own models, build on the work, and help drive progress towards Al that supports human flourishing.

[Ali Malik @AliRMalik_]

Really grateful to have worked on this with some amazing collaborators at @OpenAI, especially @declangrabbmd and @thekaransinghal.

Special thanks to all the mental health clinicians who dedicated their time and expertise to making this a reality.

[Ali Malik @AliRMalik_]

Our results show the steady progress of Al models in helping people navigate mental health situations.

We're releasing it so researchers can evaluate their own models, build on the work, and help drive progress towards Al that supports human flourishing.

[OpenAI @OpenAI]

As AI takes on longer, higher-stakes tasks, we want models to carry beneficial and safe behavior into new domains beyond their training—and maintain it under pressure.

That’s the idea behind our new research on training models to be broadly and persistently beneficial. https://t.co/6Yw45s1RRq

[Ali Malik @AliRMalik_]

This was a great collaboration with @mayhewsw @chrispiech @klintonbicknell

We show how a combination of fine-tuning, PPO, and best-of-k sampling can help open source models match the performance of proprietary LLMs, at just a fraction of the cost.

Some examples from our paper:

[Ali Malik @AliRMalik_]

How can you control the language level at which LLMs speak? For non-native speakers and language learners, this is an extremely useful capability.

Find out more in our latest findings paper at ACL 2024!
https://t.co/CNRLl5R4GH

Work done at #Duolingo AI

[Ali Malik @AliRMalik_]

Had a great time at #sigcse2024 presenting our work on having students teach as a novel form of assessment that is fulfilling, deepens understanding, and also helps others!

Joint work with my amazing collaborators @juliettewoodrow and @chrispiech.

https://t.co/jE129ELG0P

[Chris Piech @chrispiech]

It is a hard time in America. We need to stem police violence and also deliver equitable access to high quality education + healthcare. Keeps me going.

I get to give back to the wonderful community of folks working for access to eyecare
https://t.co/7WhYpdMqtD