🔥Telegram İfşa

How is AI simultaneously super intelligent and producing such terrible reviews 🤔

What if learning and unlearning happen simultaneously, with unlearning requests between updates? Check out our work on online learning and unlearning-algorithms with certified unlearning, low computational overhead and almost no additional regret. arxiv.org/abs/2505.08557

Yes NeurIPS Conference, this is confusing. Information available at abstract deadline was that abstract is editable but can't be placeholders. My understanding was that we can't materially change it but edits are possible. Is that not true anymore ? Then why is it still editable?

Ad for postdoc position with me in University of Copenhagen with a relatively tight deadline. employment.ku.dk/all-vacancies/… Note: The Assistant Professor position is essentially a senior postdoc position. We will discuss which position is more suitable for you after shortlisting.

We’re happy to announce the Call for Competitions for SaTML Conference! The competition track has been a highlight of SaTML, featuring exciting topics and strong participation. If you’d like to host a competition for SaTML 2026, visit: 👉 satml.org/call-for-compe… ⏰ Deadline: Aug 6

🚨 Got a great idea for an AI + Security competition? SaTML Conference is now accepting proposals for its Competition Track! Showcase your challenge and engage the community. 👉 satml.org/call-for-compe… 🗓️ Deadline: Aug 6

GPT5 doesn’t seem to be anything remarkably better than o3 or o4-mini-high. Is that other people’s experience as well ?

A lot of the current LLM research looks like experiments social scientists would have done with humans if they didn't have an ethics board to report to. 🤔 P.S. Not complaining, they're pretty cool!

An Iterative Algorithm for Differentially Private $k$-PCA with Adaptive Noise ift.tt/AjsLEDv

Can't even put emdash anywhere anymore as people will think its written by ChatGPT...

📢Adam Smith, Gautam Kamath, and I are putting together a list of job market candidates in Foundations of Responsible Computing! Last year's list was a great success so we're keeping it going! If you want to be included, or nominate someone, see link in the replies!

If you work on theory in the space of responsible computing and are looking for a job, fill out this form before September 28! Responsible computing includes private data analysis, fairness, robustness, and mathematical approaches bridging computer science, law, and ethics, etc.

#NeurIPS2025 Spotlight 🌟 In fundamental ℓₚ Stochastic Convex Optimization setup, we characterize the error region s.t. for every algorithm, the minimum # training points required for learning are equal (~log) to # training points that become traceable (membership inference)🧵

The Learning Theory Alliance (Learning Theory Alliance) will have a social at EurIPS Conference this year, which will be run by the amazing Carolin Heinzler Gergely Neu Amartya Sanyal. Stay tuned for more details!

This is amazing!

🤖 What happens when ML models learn from their own past predictions? They can get trapped in an "echo chamber" of their own errors. To learn more, see our new #SODA2026 paper: arxiv.org/abs/2509.25135 Joint work with Daniil Dmitriev , Harald, and Carolin.