three observations back this up: > high-confidence paths (measured by cumulative probability) are shorter AND more accurate than standard samples > at those natural endpoints, the stop token is the model's top prediction. it knows it's finished > as you widen exploration (more
RESEARCH
-
Researchers test AI self-awareness with TSearch
By
–
here's where it gets interesting. the researchers probed whether models internally "know" they're done. they introduced TSearch, which scores partial reasoning traces by cumulative log-probability across the entire chain, not just the next token. when you let the model explore
-
Overthinking hurts accuracy in AI responses
By
–
and it's not just wasted compute. overthinking actively hurts accuracy. DeepSeek-R1 produces responses 5x longer than Claude 3.7 Sonnet on AIME 2025 with comparable accuracy. QwQ-32B scores 2 percentage points HIGHER with its shortest answers using 31% fewer tokens. 72% of
-
RFCS Metric Reveals Early Correct Steps in Models
By
–
first, the problem quantified. the researchers created a metric called RFCS (Ratio of First Correct Step) that tracks where in a chain of thought the correct answer first appears. on MATH-500, across every model tested, the right answer shows up well before the end in over half
-
Overthinking in AI models is a sampling issue
By
–
reasoning models already know when they've solved the problem. we just don't let them stop. new paper from Beihang University and ByteDance shows that the overthinking problem in models like DeepSeek-R1 and Qwen3 isn't a training failure. it's a sampling failure. the fix cuts
-
AI Learning Risks and Rewards in Student Education
By
–
Students using AI sometimes fail to learn. But it’s not a given.
— SAS Software (@SASsoftware) 4 mars 2026
In this episode of Pondering AI, Master’s student Seth Rabinowitz demonstrates that risks and rewards of AI in education are not lost on students. On topics ranging from intentional learning to interpersonal skills… pic.twitter.com/ab357SgcVoStudents using AI sometimes fail to learn. But it’s not a given. In this episode of Pondering AI, Master’s student Seth Rabinowitz demonstrates that risks and rewards of AI in education are not lost on students. On topics ranging from intentional learning to interpersonal skills
-
Lists of everyone in AI research, not broken out
By
–
I have lists of everyone in AI research: https://
x.com/scobleizer/lis
ts
… Not broken out that way -
UK government invests £40M in AI research lab for scientific breakthroughs
By
–
The UK government commits an initial £40M to an AI research lab, modeled on its DARPA-inspired ARIA, seeking breakthroughs in science, healthcare, and transport (@madhumita29 / Financial Times) ft.com/content/41f522fc-10f5… techmeme.com/260304/p3#a2603…
→ View original post on X — @madhumita29, 2026-03-04 05:41 UTC
-
Question about Gemini Diffusion model progress and speed
By
–
Slight off topic question Demis, but how is the work on the Gemini Diffusion model? If I got it right that model was churning out at ~1500t/s ? Are you (GDM) actively working on this project still?
-
Defining Machine Learning Through Foundational Concepts
By
–
"Machine learning is the study of computer algorithms that automatically improve with experience." — From the first page of Tom Mitchell's book [1997]: https://
cs.cmu.edu/afs/cs.cmu.edu
/user/mitchell/ftp/mlbook.html
… My paraphrase: "ML is the set of 𝙈𝙖𝙩𝙝𝙚𝙢𝙖𝙩𝙞𝙘𝙖𝙡 𝘼𝙇𝙂𝙊𝙍𝙄𝙏𝙃𝙈𝙎 𝙩𝙝𝙖𝙩 𝙇𝙀𝘼𝙍𝙉