The study found that training only a specific intermediate layer during RL training can achieve performance similar to full-parameter training.
Google announced GenKit, which helps easily integrate AI features into existing applications.
The CLI and MCP server 'naeryeo', which allows users to look up public transit routes and fares, has been released.
Hexana MCP 0.3 has been released, allowing AI coding assistants to directly inspect the internals of WASM and native binaries.
CursorBench 3.1, a new benchmark for measuring AI code editor performance, has been released.
Methodological flaws have been pointed out in a Nature Medicine paper claiming that frontier LLMs outperform specialized medical AI.
SenseNova-U1-8B-MoT-Infographic-V2, an open-source model specialized for infographic generation and image editing, has been released.
NVIDIA is introducing a revenue-sharing and credit-support model to help AI startups and enterprises build infrastructure.
A new benchmark has been released to evaluate AI agents' practical capabilities at a senior engineer level.
Gnosys unveiled a classifier optimization technique that outperforms existing optimization methods even in environments lacking labels.
This work proposes an Anti-Causal Domain Generalization approach that learns robust predictive models for new environments by leveraging unlabeled data.
Proposes a risk control framework that optimizes token usage during LLM inference while keeping the error rate below a certain level.
By recycling information from discarded tokens through RCD technology, the accuracy and inference speed of diffusion language models have been dramatically improved.
This work proposes a new policy learning method that maximizes the efficiency of token selection (unmasking) in Diffusion LLMs through reinforcement learning.
Vercel AI Gateway has launched a routing rules feature that lets teams control models without modifying application code.
Apple researchers have proposed the Ctrl-R framework to help LLMs effectively learn the diverse reasoning patterns needed for complex problem solving.
MemoryLLM proposes separating the FFN from Self-Attention in Transformers and using it as a context-free, token-level neural retrieval memory.
RL-finetuned VLMs show improved visual reasoning performance, but exhibit vulnerability to textual perturbations.
This proposes Amortized MIPS, which dramatically reduces search computation cost by directly predicting MIPS solutions via neural networks.
VideoFlexTok dramatically improves training efficiency and generation length by representing video in stages, from abstract information to fine details.
This study examines the synergy effects of multi-agent systems that collaborate autonomously without a fixed workflow.
OpenAI has proposed offering the U.S. government a 5% equity stake to ease political pressure and share the benefits of AI.
An analysis of 21,000 companies in the US found that companies that increased AI investment saw growth in both overall headcount and entry-level hiring.
High-quality human annotations were leveraged to train LLMs to perform information filtering and judgment at the level of financial experts.
AI Briefing