Tag
OBLIQ-Bench introduces a suite of five oblique search tasks that expose a gap between retrieval and verification: reasoning LLMs easily recognize relevant documents once surfaced, but even state-of-the-art retrievers fail to surface them, highlighting overlooked bottlenecks in modern retrieval systems.
MIT researchers compiled the 2025 AI Agent Index, documenting 30 deployed AI agents from major labs, yet only 4 had public documentation explaining what the agent does, its limitations, and failure modes, revealing major transparency gaps.
MIT Technology Review's newsletter covers a new low-cost lithium extraction method using weak acid, a concerning Ebola outbreak in the DRC, and an opinion piece on Pope Leo XIV's AI encyclical urging ethical action.
MIT Technology Review analysis argues that recent tech layoffs are driven by macroeconomic factors, not AI, contradicting widespread fears of AI-driven unemployment.
A thread highlights two separate insights: a Google researcher found that adding 'you are an MIT mathematician' to a prompt fixes math errors in LLMs, and Alex Albert explains how Anthropic trains Claude's personality. Both resources are free and offer deep dives into how LLMs actually work.
MIT associate professor Kevin Esvelt and team are using genetic engineering on mice to reduce Lyme disease transmission on Nantucket. The 60 Minutes segment details the severity of Lyme disease and the innovative approach to combat it.
MIT has made the entire 'Deep Learning' textbook by Goodfellow, Bengio, and Courville freely available online without paywall or signup, providing over 800 pages of foundational knowledge.
A tweet shares a lecture at MIT by David Shirokoff covering the fundamentals of Markov Chains, including transition probabilities, Markov matrices, eigenvalues, and long-term steady state.
MIT Associate Professor Connor Coley discusses his work developing AI models to understand chemical principles and accelerate drug discovery by predicting reaction pathways and analyzing vast numbers of potential compounds.
Justin Solomon, associate professor at MIT, has been appointed associate dean of engineering education at the School of Engineering, effective July 1. He will focus on advancing innovation in engineering education, integrating AI into curricula, and building industry collaborations.
The author investigates the conflicting origins of the word 'zork' as a jargon term for unfinished programs at MIT, uncovering discrepancies in Wikipedia's sources and correcting their own earlier claims.
MIT researchers propose Pedagogical RL, a new reinforcement learning method that uses a teacher model with privileged information and a spike-aware learnability reward to significantly improve sample efficiency and convergence speed over existing methods like GRPO and OPSD.
MIT Technology Review hosts an insider panel discussion on key technology signals and their impact.
MIT introduces Pedagogical RL, a method that trains a teacher to produce trajectories that are learnable for a student by penalizing surprising steps, improving RL training efficiency.
Guest lecture at MIT 6.566 on AI agent security covering system-level threats, prompt injection, tool-use vulnerabilities, and demonstrations with LLMs like GPT-5.4 and Qwen 3.5.
MIT students built a device that gives AI a physical body using a camera and electric pulses to move fingers, enabling tasks like playing piano and drawing, powered by Claude, built in 48 hours.
A user shares how they used MIT's free AI education materials and fed them into Claude to create a rebuilt research system.
A tweet shares the story of an MIT professor who gave his last lecture before dying, distilling his life's knowledge into one hour.
Snapchat paid $150 million for Looksery, a deep learning computer vision startup. A free MIT lecture teaches building neural networks from scratch.
Peter Hotez criticizes the erosion of support for U.S. research universities, citing MIT's 20% reduction in graduate students as a threat to the next generation of scientists.