QUBRIC shows that rewriting training prompts alongside rubrics unlocks RL training for tasks where no ground-truth answer exists, gaining +5.5 points on ArenaHard
A new RL framework replaces scalar rewards with reward distributions, making behavioral diversity a natural byproduct of uncertainty rather than a fragile tuning knob.
NetKV adds a network-cost oracle to LLM inference schedulers, cutting mean TTFT by up to 21% on 64-GPU clusters with zero changes to hardware or inference engines
VEPO fixes a blind spot in multimodal RL training by combining token entropy with visual sensitivity, boosting visual reasoning by up to 3.15 points over entropy-only baselines
A new study retrained all 31 encoder subsets from scratch, revealing that standard encoder evaluation methods are fundamentally misleading -- and that two encoders can match five.
A new pre-registered study protocol examines whether configuring Claude Code and OpenAI Codex can actually control which libraries they silently pull into your codebase.
FFR extends Hinton's Forward-Forward algorithm to regression tasks, cutting training memory to just 8% of backpropagation's at depth 32 while recovering 98.6% of BP accuracy
A new multi-agent LLM pipeline turns raw wearable data into narrative health updates for remote family caregivers, improving trust and satisfaction over raw data summaries
scTranslation is the first systematic benchmark for single-cell multi-omics translation, revealing how feature selection, data quality, and few-shot settings dramatically affect model performance
IBM Research's PROVE framework trains small LLMs to reliably chain multi-step tool calls using live MCP servers and a judge-free reward system, beating larger models on key benchmarks.
A training-free multi-agent framework combines knowledge graphs and LLMs to summarize multiple documents, beating supervised models without any fine-tuning
PyraMathBench exposes a critical gap between LLM reasoning and numerical computation, then fixes it with a tool-augmented training system that boosts Qwen-2.5 by 5 points
Researchers at U of Toronto built an AI worm that exploited 73.8% of a test network autonomously, using stolen GPU compute and open-weight LLMs — no cloud APIs required.
EvoDS is an open-source data science agent that learns new skills on the fly and manages its own context window, beating existing agents by 28.9% across four benchmarks
E2LLM uses genetic algorithms and dynamic programming to run full LLMs across heterogeneous edge devices, cutting wait times by over 50% versus Splitwise
LiveBand generates real-time music accompaniments on consumer hardware by replacing teacher forcing with adversarial sequence-level training, eliminating error drift entirely