Jason Yosinski

@kchonyc Stress notwithstanding the talk was great!

so stressful

@ManasJoglekar @GabrielDWu1 @j_asminewang @boazbaraktcs @mia_glaese See also @techreview writeup x.com/techreview/sta…

This was a fun project (that I jumped into halfway through) with @ManasJoglekar, Jeremy Chen, @GabrielDWu1, @j_asminewang, @boazbaraktcs, and @mia_glaese. Boaz wrote a good casual summary: x.com/boazbaraktcs/s…

We just posted a blog + paper on a a simple but effective approach to model honesty called "Confessions" TL; DR: normal RL training rewards for high performance on a task. Confession training is a separate phase that rewards only for honesty. Test look promising! More:

Today, OpenAI is launching a new Alignment Research blog: a space for publishing more of our work on alignment and safety more frequently, and for a technical audience. alignment.openai.com

@oh_that_hat @hugo_larochelle I still remember seeing those plots in group meeting showing way better than chance performance at iter 0 and thinking "yay, another bug to track down".

@jasonyo @hugo_larochelle I peaked at those egg key diagrams!

Super proud to share that after a long sabbatical, I have finally defended my PhD 🥳🥹 Growing up, I never thought I’d pursue a PhD, and I complained a lot during it, but looking back now it was a very special and precious experience. Special thanks to @hugo_larochelle @AaronCourville @HanieSedghi @bneyshabur @PreetumNakkiran @jasonyo and many MILA friends for advising me through it all 🌺

A little while ago, many of you gave generously to support a number of MLC-Nigeria researchers in attending Deep Learning Indaba #DLI2025. Here's the crew 👇that attended; from what we hear it was a bustle of talks, posters, mentorship, and sparks of collaboration!

Today we say goodbye to @DeepIndaba after six inspiring days in Kigali rich with keynotes, tutorials, workshops, mentorship circles, and insightful posters that kept us learning non-stop. Some of us were only able to make it down to #DLI2025 because of your generous support.

Help send a bunch of researchers to DL Indaba this year! For less than one H100 we can send 25 people!

@mrcslws Wow can't believe that tweet was from 2012!

Hat tip to Coding Noir x.com/jasonyo/status…

My new favorite hidden iOS feature: Background Sounds Add it to your Control Center Then listen to Daft Punk + rain

Starting in 1 hour: @thebasepoint presents Anthropic's "Biology of a Large Language Model" work at the DLCT reading group. Paper: transformer-circuits.pub/2025/attributi… Come for the chain of thought, stay for the rabbits and habbits. Zoom info below 👇

@MFarajtabar Nice work! Any of you authors want to present this at the @ml_collective Friday reading group? mlcollective.org/dlct/ @ParshinShojaee @i_mirzadeh

🧵 1/8 The Illusion of Thinking: Are reasoning models like o1/o3, DeepSeek-R1, and Claude 3.7 Sonnet really "thinking"? 🤔 Or are they just throwing more compute towards pattern matching? The new Large Reasoning Models (LRMs) show promising gains on math and coding benchmarks, but we found their fundamental limitations are more severe than expected. In our latest work, we compared each “thinking” LRM with its “non-thinking” LLM twin. Unlike most prior works that only measure the final performance, we analyzed their actual reasoning traces—looking inside their long "thoughts". Our analysis reveals several interesting results ⬇️ 📄 machinelearning.apple.com/research/illus… Work led by @ParshinShojaee and @i_mirzadeh, and with @KeivanAlizadeh2, @mchorton1991, Samy Bengio.

Starting in 30 min!

Next MLC Research Jam is tomorrow; sharing two ideas myself to mix things up :)

Starting in 15 min!

I am sitting here watching my HF smolagent slowly reason about and click on Captcha squares one a time 🙈. Is this general AI?

@scychan_brains Love this simple/sticky mental model!

When we have low confidence (imposter syndrome), we are often afraid to take action in the world. Specifically, I mean actions that have external impact. Because of this fear of taking action, we then get very little feedback (positive or negative), which reduces learning and growth. Crucially, we also get very few successes! This lack of growth and success further decreases our confidence, and the cycle continues… 1/

Some years ago, I got trapped in a Massive Trough of Imposter Syndrome. It took more than a year to dig myself out of it, but the following framework really helped me. It feels a bit vulnerable to share, but I hope it might help a few others too! A short thread 🧵🙂

@NotACyborgYet @AttarZiv @insightpartners @PraveenAkkiraju @JonahWaldman2 Congrats, Tom and team!

Exciting news: we raised a $20M Series A at Glass Imaging to massively improve image quality using AI. @AttarZiv and I are excited to have @insightpartners @PraveenAkkiraju and @JonahWaldman2 join us for the next phase of our growth!

Tomorrow at 10am PT we'll have our next MLC OpenClubHouse, our 25th 🎉! Stop by to hang out, catch up with friends, and chat about ML or anything else. We'll meet in the MLC Discord #openclubhouse channel: discord.gg/6Za9MBr4?event…