Current Limits on AI Automating Its Own R&D
LLMs can already automate routine tasks like monitoring training runs and fixing simple errors overnight, but full AI R&D automation is not here....
Created by Mari
AI research papers and clear explainers across healthcare, robotics, finance
Explore the latest content tracked by AI Research & Applied Insights
LLMs can already automate routine tasks like monitoring training runs and fixing simple errors overnight, but full AI R&D automation is not here....
OpenAI dismissed three safety researchers for mishandling research information, with one case tied to accessing an executive’s email that the company...
AI is emerging as a participant in mathematical breakthroughs, yet gains hinge on human input and uneven access to computing power.
A measured path forward combines realism about dangers with practical incentives, avoiding both panic and inaction.
China's new guidelines target AI theory breakthroughs alongside expanded computing power, algorithms, and data supplies, while explicitly aiming to curb tech bubbles and contain AI risks.
OpenAI's dump of over 700 math papers on GitHub has wiped out entire research programs for early-career mathematicians, NYU professor Tristan...
AI model training now spans multiple datacenters as single sites hit power and space limits.
A new approach trains text-to-image models directly in pixel space using local attention windows, achieving 32x32 token reduction without a VAE. This...
Neural networks produce human-like outputs through pure mathematical functions, yet their internal vectors can mimic symbolic structures.
A new non-invasive method evaluates ergonomic risks in manufacturing tasks by combining 3D pose estimation with Physics-Informed Neural Networks. This could streamline workplace safety analysis.
AI labs should stop marketing personal assistants for vacations, reservations, and party planning since these aren't pressing issues. Instead, focus on the cost of living crisis, where agents could deliver real value to Americans.
Bechtel is using four-legged robots, autonomous equipment, and automated welding to speed up projects, with design changes reaching the field in hours...
DaltonTx's engine shows how integrated AI can fuse experimental results, project data, and scientific judgment to decide what to test next in antibody...
Global venture funding reached $679 billion in the first three quarters of 2026, the highest total ever for that period.
Gecko Robotics and Anduril are deploying AI-powered inspection robots at a new software-defined shipyard to cut forged-component inspection times by...
Two NEI-funded projects reveal a clear progression in ophthalmology.
Computer vision is reducing human rater inconsistency in neurotech by automating motor assessments and improving targeting precision. Machine...
ChatGPT still dominates, yet durable players like Suno and ElevenLabs show staying power while most revenue flows to prosumer tools.