I was waiting and hoping the ML community on Twitter would move over to Mastodon so I wouldn't have to create an account here. But, well… here we are! :)

1/ Can Large Language Models (LLMs) truly reason? Or are they just sophisticated pattern matchers? In our latest preprint, we explore this key question through a large-scale study of both open-source like Llama, Phi, Gemma, and Mistral and leading closed models, including the

📢Internships at Apple ML Research🍏 We’re looking for a PhD research intern with interests in uncertainty quantification, LLMs, probabilistic ML and/or decision making under uncertainty! See thread for more details 👇 [1/3]

** Intern position on LLM reasoning ** Maxwell Horton, Iman Mirzadeh, Keivan Alizadeh and I are co-hosting an intern position at #Apple to work on understanding and improving reasoning capabilities of LLMs. The ideal candidate: - Has prior publications on LLM reasoning - Is

1/ LLM inference is very expensive; and LLMs don't necessarily use their full capacity to respond to a specific prompt. That's why many researchers have been investigating adaptive computation methods such as early exiting, layer/expert pruning, speculative decoding, mixture of

1/🔔Excited to share my internship work, SALSA: Soup-based Alignment Learning for Stronger Adaptation, (NeurIPS workshop paper)! 🎉 Proximal Policy Optimization (PPO) often limits exploration by keeping models tethered to a single reference model. SALSA, however, breaks free

We have open-sourced GSM-Symbolic templates and generated data! 🎉 - Github: github.com/apple/ml-gsm-s… - Hugging Face: huggingface.co/datasets/apple… I will be also attending #NeurIPS2024. If you are also attending and would like to discuss research ideas on reasoning, let's connect :)

🍏🍏🍏 Come work with us at Apple Machine Learning Research! 🍏🍏🍏 Our team focuses on curiosity-based, open research. We work on several topics, including LLMs, optimization, optimal transport, uncertainty quantification, and generative modeling. Infos 👇

Amazing analysis! This has been THE question I was thinking about every single day in the past month. Although, I think if the model knows the algorithm (multiplication), we can only measure the accuracy of execution by the model and not necessarily their search/reasoning power.

Exactly! I wish that at least academic people understood this. "All" models we have today are trained using cross-entropy to fit a distribution => By design, It is "impossible" for them to generate anything outside of that distribution.

It was a pleasure joining Machine Learning Street Talk during the NeurIPS conference in December. While it might seem that a lot has changed over the past 3 months (e.g., with new models like o3/R1), I still believe the current models are not capable of reasoning :) youtube.com/watch?v=yQPdue…

I will be attending #ICLR this week to present our GSM-Symbolic paper, and we also have a full-time opening on our team! Let me know if you're interested in discussing reasoning and/or joining us!

🧵 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,

Healthy and unhealthy strategies for coping with the Apple paper: - attack Apple for publishing it (which does nothing to address the underlying problems they pointed out) or - figure out its implications and develop a robust alternative (the healthier option)

The biggest weakness was a lack of creativity and deep understanding. This is perhaps most aptly captured by a quote from one of the mathematicians:

Come work with us! The Machine Learning Research (MLR) team at Apple is seeking a passionate AI researcher to work on Efficient ML algorithms: jobs.apple.com/en-us/details/…

Is your AI keeping Up with the world? Announcing #NeurIPS2025 CCFM Workshop: Continual and Compatible Foundation Model Updates When/Where: Dec. 6-7 San Diego Submission deadline: Aug. 22, 2025. (opening soon!) sites.google.com/view/ccfm-neur… #FoundationModels #ContinualLearning

📢Submissions are now open for #NeurIPS2025 CCFM workshop. Submission deadline: August 22, 2025, AoE. Website: sites.google.com/view/ccfm-neur… Call for papers: sites.google.com/view/ccfm-neur… Submission Link: openreview.net/group?id=NeurI…

Join our innovative team at #Apple as a Research Scientist/Engineer specializing in LLM #Reasoning, #Planning, and General #Intelligence. We are seeking an ideal candidate who: - Is available to start by the end of this year - Holds a PhD or will graduate by year-end - Has 3-5