MIT CSAIL researchers deployed the Claude AI model, nicknamed Eko, on a simulated island with no external rewards or human demonstrations, letting it set its own goals and record its activities in text files as long-term memory. After roughly 30 hours of unguided play, the agent outperformed fresh copies on a subsequent task of stacking objects on the island's peak, and deleting portions of its memory logs reduced its performance.
technologyreview.com
· 2026-10-19
A team led by physicist Gabriel Mindlin at the University of Buenos Aires generated a computer-recreated version of a rufous-collared sparrow song last documented in a Buenos Aires park in the 1960s. After playing it back to local birds, a sparrow began imitating parts of the song within about three months. The work is part of broader research into how declining bird populations lose access to the songs needed for mating and territory defense.
e360.yale.edu
· 2026-10-14
Microsoft researchers published a study in Nature describing a machine learning system that corrects bias in existing physics-based and AI weather models, improving forecast accuracy two to six weeks ahead. The tool, now available as open-source software, can run on standard CPUs rather than requiring supercomputers or powerful GPUs.
theverge.com
· 2026-10-14
Researchers describe a probabilistic forecasting method for predicting weather two to six weeks ahead, built on ERA5 reanalysis data from 1979-2026 and evaluated using ECMWF's AI Weather Quest competition framework. The approach forecasts temperature, precipitation and sea-level pressure as probability distributions across climatological quantile bins, scored using the Ranked Probability Skill Score.
nature.com
· 2026-10-14
Researchers used two-photon imaging and holographic optogenetics to stimulate small groups of neurons in the motor cortex of mice trained on a lever-pressing task. Briefly activating about 20 neurons whose firing normally precedes voluntary movement produced both movements resembling the learned action and brain activity patterns resembling the natural learned pattern. The effect depended on the mouse being task-engaged and on stimulating the correct, early-firing neurons rather than ones that fire after movement onset.
nature.com
· 2026-10-14
A piece describing an approach called AlphaNIST argues that robots should be trained using binary success/failure rewards and broad experience rather than being given explicit, pre-programmed instructions for every movement. The reasoning given is that two-finger grippers and other robot hardware often can't replicate human-like dexterity, so scripting exact motions is impractical. Instead, the approach lets robots discover their own solutions through trial across diverse scenarios.
alpha-nist.github.io
· 2026-10-13
A technical explainer breaks down NVIDIA's DLSS upscaling technology, describing how GPUs render games at a lower internal resolution before upscaling to a monitor's native output using machine learning models trained on motion, depth and brightness data. It outlines the three main presets—Quality, Balanced and Performance—each rendering from a progressively lower starting resolution, trading fidelity for higher frame rates.
engadget.com
· 2026-10-13
A new paper describes Minions, a distributed system that aggregates spare GPU capacity from macOS devices running Apple Silicon chips to run large-scale machine learning inference workloads. The authors say the system handles task scheduling, fault tolerance, and resource utilization across these devices without requiring teams to buy additional GPU hardware.
machinelearning.apple.com
· 2026-10-13
SimpleClosure, a service that helps founders shut down startups, has expanded into selling their code and internal data—emails, chats, documents—to AI labs and other buyers. Its Asset Hub platform has completed over 150 such deals and expects to return more than $10 million to startups by end of 2026, with 47% of its closures now including data sales. Competitor Sunset runs a similar business, and firms like Surge AI are buying this 'operational exhaust' to build training environments for AI agents.
technologyreview.com
· 2026-10-13
Meta has released two new AI models, Muse Wonder and Muse Wonder Full, trained at its gigawatt-scale Prometheus data center. The company says the models build on earlier work shown with Muse Spark and post state-of-the-art results on benchmarks covering agents, memory, and coding, including domain-specific tests in law, finance and health such as Harvey's Legal Agent Benchmark.
research.meta.ai
· 2026-10-12
METR tested frontier AI models on a research task requiring them to invent a new post-training method that would outperform a strong GRPO baseline across question-answering and coding benchmarks. The human algorithmic innovation in the comparison outperformed the solutions generated by the AI models tested, according to METR's report.
epoch.ai
· 2026-10-09
Noitom Robotics has published HiPHI, a whole-body human motion dataset totaling 617.5 hours captured with optical motion capture accurate to sub-millimeter precision. Of that, 245.7 hours cover human-object interaction with synchronized object trajectories and meshes, and the data is organized using FrameNet, a linguistic framework for categorizing human actions. A white paper from IEEE Spectrum and Wiley describes the dataset and shows reinforcement learning policies trained on it transferring from simulation to a physical humanoid robot.
content.knowledgehub.wiley.com
· 2026-10-07
GoKawiil