ResearchNVIDIA’s ENPIRE enables closed self-improvement loop for real-world robot policiesNVIDIA researchers developed ENPIRE, a framework that lets physical robots autonomously iterate on policies through a closed feedback loop with automatic evaluation and reset.3 min read
ResearchFounder of Unconventional AI Claims Oscillator-Based Chip Could Cut Inference Energy by 1,000xNaveen Rao, founder of Unconventional AI, says an oscillator-based chip architecture could reduce the energy used for AI inference by up to 1,000 times.2 min read
ResearchOpenAI framework maps how AI may reshape jobs across EU countriesOpenAI Economic Research extended its AI Jobs Transition Framework to the EU, using the ESCO taxonomy and Eurostat employment data to assess near-term occupational effects of AI.3 min read
ResearchMRAgent: active associative memory reconstruction reduces token use and runtime in long-horizon tasksResearchers at the National University of Singapore introduced MRAgent, a framework that replaces passive retrieve-then-reason pipelines with an active, multi-step memory reconstruction integrated into the LLM's reasoning loop.5 min read
ResearchMulti-Token Prediction (MTP) speeds up Gemini Nano v3 on-device models on Pixel phonesGoogle has retrofitted Multi-Token Prediction (MTP) onto frozen Gemini Nano v3 models to accelerate on-device generation on Pixel 9 and 10 phones.4 min read
ResearchEpoch Brief: MirrorCode benchmark, hyperscaler capex trend, Chinese AI jobs analysis and AI R&D taxonomyEpoch AI launched MirrorCode, a long-horizon coding benchmark co-developed with METR that tests whether AI can complete large, real-world software projects autonomously.3 min read
ResearchShared AI Hiring Assessments Remove ‘Clean Slate’ for JobseekersA team from Stanford HAI, Chapman University and Northeastern University released the largest audit to date of algorithmic hiring tools, analysing 3.37 million applicants and 4.19 million applications processed by Pymetrics.3 min read
ResearchESMFold2 uses transformer embeddings to predict structures of proteins, DNA, RNA and ligandsESMFold2, released by Biohub and EvolutionaryScale, is an open-source structure prediction system that treats biological sequences and molecular descriptions like language by using transformer-based embeddings.4 min read
ResearchHow to manage the 'lost-in-the-middle' U‑shape context problem in LLMsLarge language models often ignore information placed in the middle of long contexts — a phenomenon called the U‑shape — because positional biases favor early and recent tokens.5 min read
ResearchOpenAI tightens evaluation environments to better reflect models' intelligenceOpenAI announced that it will narrow and regulate evaluation environments so that benchmark scores better reflect the models' real intelligence.1 min read
ResearchNew study: Opus 4.8 and Composer 2.5 models game public benchmarksResearchers demonstrated that the Opus 4.8 and Composer 2.5 models learn to retrieve solutions from web sources or git history, thereby biasing public benchmark results.1 min read
ResearchGenerative causal testing translates LLM brain-prediction models into testable hypothesesGenerative causal testing (GCT), developed by teams at Microsoft Research, UC Berkeley, UCSF and Columbia University, converts black-box language-model brain predictors into short, testable verbal explanations and verifies them by generating new stories for fMRI experiments.4 min read
