Veille scientifique · Page 8

How UK AISI and EvalEval Are Making Benchmark Results Reproducible

Evaluation & safety · primary_technical_publication

Direct Optimization of Generators for Search in Automated Theorem Proving

Formal verification · preprint

Fine-tuned Large Language Models (LLMs) significantly advance Automated Theorem Proving (ATP), but are often deployed as guiding policies within tree search rather than for single-attempt

Transformers now runs llama.cpp quants

AI models · primary_technical_publication

Advisory Group on Mathematics and Artificial Intelligence

Mathematical reasoning · provider_announcement

OpenAI is working with an independent Advisory Group on Mathematics and Artificial Intelligence to guide the review and communication of emerging AI results.

Partnering with Accenture on embedded evaluation

Evaluation & safety · provider_announcement

We’re partnering with Accenture on independent evaluation of frontier AI—part of our recent commitment to embed evaluators at Anthropic. Both we and Accenture expect to

Your Agent Aced the Task. Will It Do It Again?

Agent systems · primary_technical_publication

Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

AI models · provider_announcement

Improving Mathematical Reasoning Capabilities in Large Language Models via Reasoning Process Error Classification

Mathematical reasoning · preprint

The reasoning ability of large language models (LLMs) is a critical factor for practical LLM-based applications. To investigate the current reasoning capability of LLMs, we

Transformed in Translation: Two-Stage Structural Uncertainty in LLM-Based Scientific Autoformalization

Formal verification · preprint

Scientific autoformalization turns verbal accounts into executable mathematics, but executable code does not settle which model has been constructed. We examine two sources of structural

Beyond Solver Verdicts: Generative Reward Models for Autoformalization

Formal verification · preprint

Neurosymbolic systems rely on mathematical solvers to guarantee reasoning correctness, yet solvers are fundamentally blind to whether a formal translation maintains strict reference-equivalence to a

Magenta: Closing the Loop Between Mathematical Reasoning and Lean Verification

Formal verification · preprint

Most of mathematical knowledge has been communicated through so-called informal use of mathematics and natural language. With large language models (LLMs) being highly adept in

AxQM: A Textbook-Scale Benchmark for Formal Proof Synthesis in a Library of Finite-Dimensional Quantum Mechanics

Formal verification · preprint

Formalizing mathematics in a proof assistant, where a machine checks every definition, statement and proof, has set a new standard of rigor. Large language models

NeoMME: an efficient Multimodal-native and Multilingual Encoder

Compute & efficiency · primary_technical_publication

Introducing WeatherNext 3, our most advanced and accurate global weather AI model

AI models · provider_announcement

Give Your Coding Agents a Memory You Own

Agent systems · primary_technical_publication

Training a coding model to paint watercolours with TRL and OpenEnv

AI models · primary_technical_publication

Introducing Gemini 3.8 Flash and 3.8 Flash Cyber

AI models · provider_announcement

BenchMIRT: What are LLM benchmarks actually measuring?

Evaluation & safety · primary_technical_publication

Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI

Compute & efficiency · primary_technical_publication

Introducing agentic video understanding with Gemini

Agent systems · provider_announcement

SHADOWBENCH: Toward Reliable Automatic Evaluation of Semantic Alignment in Autoformalization

Formal verification · preprint

Autoformalization translates informal mathematical theorems into code for proof assistants such as Lean. A central challenge is that current evaluation metrics can accept type-correct but

Learning to Use Tools: Reinforcement Learning for Tool-Integrated Mathematical Reasoning

Mathematical reasoning · preprint

Current large language models (LLMs) increasingly benefit from external tool integration, especially for tasks requiring reliable computation and verification. Motivated by this, we study calculator

Piloting the world's first double-blind AI evaluations

Evaluation & safety · provider_announcement

Piloting the world's first double-blind AI evaluations

Gemini Omni 1.1 Flash lets you build with more control

AI models · provider_announcement