A bit of a history of IMO-Bench and our IMO efforts:
a. We started building IMO-Bench around early 2024, which was the precursor of ProofBench (basic). b. IMO-Bench was first mentioned in the Gemini 1.5 paper around May 2024. At that time, Gemini Math-specialized 1.5 Pro scored
RESEARCH
-

IMO-Bench Development: Gemini Math Capabilities Evolution
By
–
-
ML Research: Engineering Practice Over Philosophical Speculation
By
–
ML research is an engineering discipline, not a philosophy seminar. You build, you test, you learn. Untested ideas are just speculation.
-

Exploring AI Layers: Neural Networks to Deep Learning Explained
By
–
Peeling back the layers of AI! This infographic unravels the complex world from Neural Networks to Deep Learning. Dive deeper into AI with the insights from @ingliguori
's 'The Digital Edge' http://
bit.ly/3u4pILl #ArtificialIntelligence #MachineLearning #DeepLearning -
RL Environments Research Blog by ArminPCM Team
By
–
Brilliant work by @ArminPCM and our research team on the latest RL Environments themed blog. Dive in
-

Building Advanced RL Environments for Reliable AI Reasoning
By
–
A good RL environment is more than a sandbox — it’s a system with:
– Real tools & APIs
– Stateful feedback
– Deterministic rewards & rubrics At Snorkel, we combine expert-built tasks + automated QC to create environments that truly measure reasoning and reliability. -

Snorkel Builds Enterprise-Grade Reinforcement Learning Environments
By
–
New from @ArminPCM : how Snorkel builds reinforcement learning (RL) environments that train and evaluate agents in realistic, enterprise-grade settings.
-
Michael Dertouzos: Computing Pioneer and MIT Lab Director Legacy
By
–
Happy birthday to the late Michael Dertouzos, who directed MIT’s Laboratory for Computer Science for 25+ years.
— MIT CSAIL (@MIT_CSAIL) 5 novembre 2025
Watch him discuss the future of computers on NBC’s Today Show w/late MIT prof. Joseph Weizenbaum back in 1983. pic.twitter.com/IklDx7mklNHappy birthday to the late Michael Dertouzos, who directed MIT’s Laboratory for Computer Science for 25+ years. Watch him discuss the future of computers on NBC’s Today Show w/late MIT prof. Joseph Weizenbaum back in 1983.
-

SAS Anchors AI Expectations: 2026 Predictions Revealed
By
–
Once upon a time, the AI hype faded, and the hope for human-centered machine models nearly burst. SAS is the data and analytics anchor that levels AI expectations for businesses and enterprises. 2026 predictions are here—and they’re not exactly what you’d expect for a hopeful
-
Tokenizer Approaches Impact LLM Performance on HellaSwag Benchmarks
By
–
If you measure downstream performance on HellaSwag rather than speedrun-equivalent loss, then different tokenizer approaches come out on top… The first run I did was much better on common-sense downstream, trained in equivalent time or better.
-
Tokenizer Training and Data Filtering Compliance Standards
By
–
Well, the tokenizer I used was trained on large quantities of data — I filtered the tokens based on yet more data from FineWeb. Question is if that's acceptable according to your rules…