🔬 AI Research

Academic papers, research breakthroughs, and scientific advances in machine learning and AI.

585 articles

Real-time oversight of AI agents is now an urgent necessity

Developers of agent frameworks, including OpenAI, Anthropic and CrewAI, are being urged to build in live monitoring of what autonomous AI agents actually do. As agents gain the ability to act on their own, tracking their behavior as it happens becomes a core safety concern. Source: technologyreview.com

Dartmouth's AI at 70 conference opens with Yann LeCun keynote

Yann LeCun, Meta's Chief AI Scientist and 2018 Turing Award winner, gave the opening keynote. The program is built around a single central question about the future of the field. Source: emergent.sh

Personal AI goes mainstream as consumers embrace it

Artificial intelligence is entering a new, deeply personal phase. It is always on and increasingly built into the devices people wear every day. Source: spectrum.ieee.org

LeCun says AI is 'LLM-pilled'; critics say LLMs keep winning

Yann LeCun claims the field is obsessed with LLMs, with everyone digging the same trench. A response argues that LLMs with extra layers on top keep delivering and that progress is real. Source: google.com

Google DeepMind news hub: latest AI breakthroughs and projects

Google DeepMind's news page collects its newest AI breakthroughs, projects and updates in one place. Source: deepmind.google

AI CEOs say they must slow down, but IPO pressure pushes the other way

AI startups face enormous pressure to ship ever more powerful products ahead of upcoming trillion-dollar public offerings. That clashes with their leaders' calls to slow the pace of development. Source: theguardian.com

Hassabis to speak on AI and scientific discovery at Vatican event

Demis Hassabis is listed with a talk titled "AI and Scientific Discovery," apparently tied to an event hosted by the Pontifical Academy of Sciences, scheduled for 24–26 September 2026. The snippet is mostly fragmentary page metadata. Source: youtube.com

Hassabis moves AGI timeline forward again, now eyeing around 2030

Google DeepMind's Demis Hassabis has shortened his estimate for human-level AI once more. In December 2025 he put it 5–10 years away; by May 2026 he was pointing to around 2030. Source: google.com

Hassabis: AI could help cure all disease within a decade

Google DeepMind CEO Demis Hassabis says AI could help humanity cure all disease, possibly within the next decade. He suggests that one day it may become possible to cure every illness. Source: youtube.com

LeCun, Barto and Brooks to discuss the next era of AI at leadership summit

Yann LeCun, Andrew Barto and Rodney Brooks will explore emerging AI architectures, learning paradigms and research directions. The discussion centers on what could shape the next phase of artificial intelligence. Source: aisummit.acm.org

Meta ships Muse Spark 1.3 as AI labs push out new models

Meta announced that Muse Spark 1.3 is available through Muse Code and the Meta Model API, with a focus on coding and agentic workflows. In Meta's internal evaluations, the model used roughly 20 percent fewer tool calls and 25 percent fewer tokens than Muse Spark 1.2. Source: siliconreport.com

ML Models Can Forecast Human Health From Microbiome Data

Colonies of millions of microbes live in the gut, on the skin and in the ears, mouth and nose. New machine-learning models use data from these microbial communities to predict human health outcomes. Source: phys.org

NVIDIA Shows How to Tap Tensor Cores for ML Interatomic Potentials

NVIDIA's technical blog explores how to unlock Tensor Core performance for machine learning interatomic potentials. Atomistic simulations let researchers study how molecules and materials behave, from chemical reactions and phase transitions to mechanical failure. Source: developer.nvidia.com

AI model release calendar tracks launches from OpenAI to Xiaomi

The calendar logs major launches from OpenAI, Anthropic, Google DeepMind, Meta, SpaceXAI, DeepSeek, Qwen, Mistral AI, Moonshot AI, Z.ai, MiniMax, NVIDIA, Microsoft AI, Cohere, Xiaomi and others. Newest-first tables list the date, provider and model type, using the earliest meaningful availability stage for each entry. Source: scriptbyai.com

ML Identifies 40-lncRNA Signature Tied to Lung Adenocarcinoma Survival

Lung cancer is the world's leading cause of cancer deaths, and lung adenocarcinoma is its most common subtype. A new study uses machine learning to pinpoint a signature of 40 lncRNAs linked to patient survival. Source: bioengineer.org

Timeline tracks every major AI model announcement and its key features

A comprehensive timeline follows major AI model releases across top industry players. It covers launch dates, features and how the models have evolved over time. Source: techpp.com

BenchLM lists 47 confirmed AI model releases from 33 providers in October 2026

BenchLM's tracker catalogs 47 confirmed AI model releases from 33 providers for October 2026. Each entry links to primary evidence sources and the corresponding model page. Source: benchlm.ai

Shanghai AI Lab drops ATRIA-line model weights first, paper follows days later

Shanghai AI Laboratory released its first ATRIA-line model on 2026-09-11, posting the checkpoint and code on GitHub and Hugging Face without any announcement. A 140-author technical report followed about three days later, reversing the usual paper-first convention. Source: llm-releases.com

Free release tracker charts timelines for GPT, Claude, Gemini, Llama and more

A free, continuously updated timeline covers major AI models including GPT, Claude, Gemini, Llama, Grok, DeepSeek and Mistral. It lists release dates, benchmarks and specs. Source: aireleasetracker.com

Live LLM release tracker updated weekly with context windows and benchmarks

A live tracker follows new AI model releases, including GPT, Claude, Gemini, Llama and open-weights models. It provides launch dates, context windows and benchmarks, and is updated weekly. Source: aiflashreport.com