ML@CMU

@mlcmublog

Official twitter account for the ML@CMU blog @mldcmu @SCSatCMU

Pittsburgh, PA

Joined February 2020

20Following

2KFollowers

ML@CMU@mlcmublog · Jul 8

blog.ml.cmu.edu/2025/07/08/car… Check out our latest post on CMU @ ICML 2025!

mlcmublog's tweet card. CMU researchers are presenting 127 papers at the Forty-Second International Conference on Machine Learning (ICML 2025), held from July 13th-19th at the Vancouver Convention Center. Here is a quick...

3.0K

ML@CMU@mlcmublog · Jun 1

blog.ml.cmu.edu/2025/06/01/rlh… In this in-depth coding tutorial, @GaoZhaolin and @g_k_swamy walk through the steps to train an LLM via RL from Human Feedback!

mlcmublog's tweet card. Reinforcement Learning from Human Feedback (RLHF) is a popular technique used to align AI systems with human preferences by training them using feedback from people, rather than relying solely on...

4.0K

ML@CMU@mlcmublog · May 22

blog.ml.cmu.edu/2025/05/22/unl… Are your LLMs truly forgetting unwanted data? In this new blog post authored by @shengyuan_26734, Yiwei Fu, @zstevenwu, and @gingsmith, we discuss how benign relearning can jog unlearned LLM's memory to recover knowledge that is supposed to be forgotten.

mlcmublog's tweet card. Machine unlearning is a promising approach to mitigate undesirable memorization of training data in ML models. In this post, we will discuss our work (which appeared at ICLR 2025) demonstrating that...

871

ML@CMU@mlcmublog · Apr 23

blog.ml.cmu.edu/2025/04/23/car… Check out our latest blog post on CMU @ ICLR 2025!

mlcmublog's tweet card. CMU researchers are presenting 143 papers at the Thirteenth International Conference on Learning Representations (ICLR 2025), held from April 24 - 28 at the Singapore EXPO. Here is a quick overview...

356

ML@CMU@mlcmublog · Apr 21

blog.ml.cmu.edu/2025/04/21/all… Check out our new blog post on ALLIE, a new chess AI that actually plays like a human! Unlike Stockfish or AlphaZero that focus on winning at all costs, ALLIE uses a transformer model trained on human chess games to make moves, ponder and resign like…

mlcmublog's tweet card. Play against Allie on lichess! Introduction In 1948, Alan Turning designed what might be the first chess playing AI, a paper program that Turing himself acted as the computer for. Since then, chess...

340

ML@CMU@mlcmublog · Apr 18

blog.ml.cmu.edu/2025/04/18/llm… 📈⚠️ Is your LLM unlearning benchmark measuring what you think it is? In a new blog post authored by @prthaker_, @shengyuan_26734, @neilkale, @yash_maurya01, @zstevenwu, and @gingsmith, we discuss why empirical benchmarks are necessary but not…

mlcmublog's tweet card. TL;DR: "Machine unlearning" aims to remove data from models without retraining the model completely. Unfortunately, state-of-the-art benchmarks for evaluating unlearning in LLMs are flawed, especia...

2.0K

ML@CMU@mlcmublog · Apr 9

blog.ml.cmu.edu/2025/04/09/cop… How do real-world developer preferences compare to existing evaluations? A CMU and UC Berkeley team led by @iamwaynechi and @valeriechen_ created @CopilotArena to collect user preferences on in-the-wild workflows. This blogpost overviews the design and…

mlcmublog's tweet image. blog.ml.cmu.edu/2025/04/09/cop…

How do real-world developer preferences compare to existing evaluations? A CMU and UC Berkeley team led by @iamwaynechi and @valeriechen_ created @CopilotArena to collect user preferences on in-the-wild workflows. This blogpost overviews the design and…

4.0K

ML@CMU@mlcmublog · Jan 9

blog.ml.cmu.edu/2025/01/08/opt… How can we train LLMs to solve complex challenges beyond just data scaling? In a new blogpost, @setlur_amrith, @QuYuxiao Matthew Yang, @LunjunZhang , @gingsmith and @aviral_kumar2 demonstrate that Meta RL can help LLMs better optimize test time compute

mlcmublog's tweet card. Figure 1: Training models to optimize test-time compute and learn "how to discover" correct responses, as opposed to the traditional learning paradigm of learning "what answer" to output. The major...

18.0K

ML@CMU@mlcmublog · Jan 2

blog.ml.cmu.edu/2025/01/02/ind… Why is our brain 🧠 modular with specialized areas? Recent research by Ruiyi Zhang @Xaqlab shows that artificial agents 🤖 with modular architectures—mirroring brain-like specialization—achieve better learning and generalization in naturalistic navigation…

mlcmublog's tweet card. TL;DR: The brain may have evolved a modular architecture for daily tasks, with circuits featuring functionally specialized modules that match the task structure. We hypothesize that this architecture...

815

ML@CMU@mlcmublog · Dec 12

blog.ml.cmu.edu/2024/12/12/hum… Have you had difficulty using a new machine for DIY or latte-making? Have you forgotten to add spice during cooking? @hciphdstudent @hiromu1996 @mollyn_paan, Jill Fain Lehman, and @mynkgoel are leveraging multimodal sensing to improve the…

mlcmublog's tweet card. TL;DR: At SmashLab, we're creating an intelligent assistant that uses the sensors in a smartwatch to support physical tasks such as cooking and DIY. This blog post explores how we use less intrusive...

2.0K

ML@CMU@mlcmublog · Dec 6

blog.ml.cmu.edu/2024/12/06/scr… A critical question arises when using large language models: should we fine-tune them or rely on prompting with in-context examples? Recent work led by @JunhongShen1 and collaborators demonstrates that we can develop state-of-the-art web agents by…

mlcmublog's tweet image. blog.ml.cmu.edu/2024/12/06/scr…

A critical question arises when using large language models: should we fine-tune them or rely on prompting with in-context examples? Recent work led by @JunhongShen1 and collaborators demonstrates that we can develop state-of-the-art web agents by…

3.0K

ML@CMU@mlcmublog · Dec 2

blog.ml.cmu.edu/2024/12/02/car… Check out our latest blog post on CMU @ NeurIPS 2024!

311

ML@CMU@mlcmublog · Nov 7

blog.ml.cmu.edu/2024/11/07/ide… Demining 70+ war-affected countries could take 1,100 years at the current pace. This AI-powered tool, developed in close collaboration with the UN in work led by Mateo Dulce, halves false alarms and speeds up clearance. Now tested in Afghanistan &…

mlcmublog's tweet image. blog.ml.cmu.edu/2024/11/07/ide…

Demining 70+ war-affected countries could take 1,100 years at the current pace. This AI-powered tool, developed in close collaboration with the UN in work led by Mateo Dulce, halves false alarms and speeds up clearance. Now tested in Afghanistan &amp;…

479

ML@CMU@mlcmublog · Oct 29

blog.ml.cmu.edu/2024/10/29/jai… AI-powered robots are alarmingly easy to jailbreak to perform dangerous tasks, including delivering bombs, surveilling humans, and ignoring traffic laws. What does the future hold for AI-powered robots? Learn more in our latest blog post, based on work…

mlcmublog's tweet card. Summary. Recent research has shown that large language models (LLMs) such as ChatGPT are susceptible to jailbreaking attacks, wherein malicious users fool an LLM into generating toxic content (e.g.,...

2.0K