MIT researchers have developed Tilted REWeighting (TREW), a method that fixes privacy leaks in AI class unlearning by adjusting the model's distribution to mimic a scratch-trained model. On CIFAR-10, …
Rachel Sava, a PhD candidate in the Harvard-MIT Program in Health Sciences and Technology, won the fourth annual Envisioning the Future of Computing Prize with her essay warning that neural implants, …
A new study finds that transformer language models implement a unified mechanism for handling multiple mental spaces—counterfactual, belief, fictional, temporal, and depictive—using a shared router ov…
More than 200 economists and AI researchers, including 16 Nobel laureates, signed an open letter organized by Stanford University's Digital Economy Lab demanding immediate action from policymakers and…
Reed Jobs has turned Yosemite from a 2023 oncology fund into a 17-person venture firm using AI, grantmaking and company formation to accelerate cancer science from university labs to clinical developm…
MIT researchers have developed a new auditing method called Gaussian probing that can detect whether an AI model has been fine-tuned to generate child sexual abuse material (CSAM) with 100% accuracy w…
MIT engineers have developed an ultrasonic wristband that tracks hand movements in real-time by capturing ultrasound images of wrist muscles and tendons, which an AI algorithm translates into hand and…
MIT researchers propose B-spline Policy (BSP), which parameterizes robot actions as continuous curves to speed up manipulation policy inference. The method reduces compute overhead by predicting B-spl…
MIT researchers, led by graduate student Vinith Suriyakumar and associate professors Ashia Wilson and Marzyeh Ghassemi, in collaboration with child safety nonprofit Thorn, have developed a new auditin…
World models, a new category of AI that simulates the physical world, are gaining momentum as an alternative to large language models (LLMs), with experts from MIT, Runway, and World Labs highlighting…
A developer built Claude Supervisor, an open-source tool that automatically pauses and resumes Claude Code when usage limits are hit, allowing long-running agentic tasks to complete unattended. The to…
Cognitive neuroscientists at MIT's McGovern Institute for Brain Research have shown that people can perform logical reasoning tasks even with severely impaired language abilities, challenging the long…
A lawyer-turned-developer open-sourced OpenSymphony, an AI agent framework that enforces personality, governance, and self-evolution through YAML-based souls, a multi-layer governance system, and a to…
A new study published in PNAS found that AI language models are far more susceptible to subtle nudges in choice architecture than humans, overreacting to default options and highlighted choices. Resea…
The AI industry's claims about world models being revolutionary are met with skepticism due to a lack of empirical evidence and real-world applications. Despite billions in investment, these models re…
Using AI as a career coach can undermine networking, increase techno-stress, and reduce critical thinking, according to experts and recent studies. While AI offers some benefits, overreliance may hurt…
A developer built StandupBot, a CLI tool that reads recent git commits and generates standup updates with sections for Yesterday, Today, and Blockers. The tool supports local execution via Ollama for …
Octochains, a zero-dependency Python framework for parallel, isolated multi-agent reasoning and consensus, has been released. It executes domain specialists in parallel isolated threads to prevent cog…
DeepSeek releases V4 Pro, a 1.6-trillion-parameter Mixture-of-Experts model with 49 billion active parameters per token, achieving a top open-weight score of 80.6% on SWE-bench Verified. The model req…
A new analysis reveals that 56% of CEOs report zero return on AI investments, while enterprise AI projects face widespread cancellations and cost overruns. Major companies including Accenture, Meta, A…