Microsoft Research
·
2 weeks ago
Microsoft researchers developed Memora, a memory system for AI agents that separates what is stored (rich content) from how it is retrieved (lightweight abstractions and cue anchors) to balance detail preservation with efficient scaling. Memora achieved state-of-the-art performance on LoCoMo and LongMemEval benchmarks while reducing token consumption by up to 98% compared to full-context inference and halving the memory entries needed versus Mem0. The system enables AI agents to maintain detailed project histories and context over long-horizon tasks without repeatedly re-reading entire conversation histories.
Hugging Face Blog
·
2 weeks ago
Researchers at Google DeepMind introduced DiScoFormer, a transformer model that estimates both the probability density and score (gradient of log-density) of a distribution from a set of data points in a single forward pass without retraining. In 100 dimensions, DiScoFormer reduces score estimation error by 6.5x and density estimation error by more than 37x compared to kernel density estimation while continuing to improve with more samples. The model enables a single pretrained estimator to work across different problems in generative modeling, Bayesian inference, and scientific computing, eliminating the need to retrain for each new distribution.
IBM Research
·
2 weeks ago
IBM is accepting applications for its third-annual Quantum Developer Conference 2026, taking place November 11-13 in Chicago, where attendees will learn about quantum advantage demonstrations and access hands-on workshops. The application deadline is September 21, with first-round acceptances sent August 10 and final acceptances by October 12. Participants will gain practical experience with IBM's latest quantum computing tools and methods to apply to their own projects.
Import AI
·
2 weeks ago
NVIDIA developed ENPIRE, software that enables physical robots to autonomously improve their own policies through iterative experimentation and execution loops similar to AI agents. The system achieved 99% success rates on dexterous manipulation tasks like pin organization and zip-tie cutting, with larger multi-agent setups converging on higher-scoring solutions faster than individual agents. This infrastructure allows robots to self-improve with minimal human intervention for evaluation and reset operations, though scaling challenges emerge as robot fleets grow.
IEEE Spectrum AI
·
2 weeks ago
Researchers discovered that a standard MOSFET transistor, when its bulk terminal is left floating rather than grounded, exhibits neuron-like behavior by producing sudden current spikes followed by relaxation, potentially enabling more energy-efficient AI hardware. The accidental discovery occurred in 2024 when a student forgot to connect the bulk terminal while measuring a memory circuit, revealing current spiking with nonlinear properties similar to biological neurons. If this single-device approach can be reliably manufactured and scaled, it could replace the dozens or hundreds of transistors currently needed to simulate each artificial neuron, reducing the enormous energy consumption of AI data centers.
Exponential View
·
2 weeks ago
AI quarterly revenues have begun exceeding quarterly depreciation expenses but have not yet covered cumulative historic depreciation costs. Chinese AI labs hire talent averaging 1.6 years of experience compared to 5.5 years for equivalent US roles. SK Hynix's market value surpassed Samsung Electronics' for the first time, reflecting shifts in the semiconductor industry.
Allen Institute (AI2)
·
2 weeks ago
Researchers introduced DiScoFormer, a transformer-based model that estimates both the probability density and score (gradient of log-density) of a distribution from data in a single forward pass without retraining. In 100 dimensions, DiScoFormer achieved 6.5x lower score error and 37x lower density error compared to kernel density estimation while continuing to improve with more samples. The model generalizes across different distribution types and could reduce computational costs for generative modeling, Bayesian inference, and scientific computing by serving as a reusable estimator.
OpenAI Blog
·
2 weeks ago
OpenAI released a report analyzing how AI adoption could affect employment across European Union member states, identifying occupations likely to experience automation, job growth, or changes in work processes. The report examines workforce impacts at a granular level across multiple EU countries and job categories, though the specific numbers and findings are not detailed in this summary. Organizations and policymakers can use the analysis to plan workforce transitions and target retraining efforts toward roles most vulnerable to automation or most likely to emerge as AI tools spread.
