AI Discovers Antibiotic for Drug-Resistant Bacteria
A deep learning model screened millions of compounds and found one that defeats MRSA and other superbugs in lab tests.
In-depth coverage, analysis, and updates on Research in AI and tech. 22 articles on AI Pulse.
A deep learning model screened millions of compounds and found one that defeats MRSA and other superbugs in lab tests.
Google's flagship model processes 3-hour videos and answers questions about specific moments. It's like having a research assistant who actually watched everything.
Users are generating 50-page literature reviews with proper citations. Academics are both impressed and terrified.
GPT-5 and Claude are generating training data that makes them better. The loop is closing.
New interpretability tools show how Claude and GPT-5 'think.' The process looks nothing like human reasoning.
Tests reveal Claude solving novel problems in ways that don't match its training data. Is this emergence or pattern matching we don't understand?
The compound works against MRSA and other superbugs. It was found by screening 12 million compounds in 3 days.
A drug candidate discovered entirely by AI reverses motor symptoms in lab animals. Human trials could start this year.
AlphaFold 3 can now predict the structure of any protein, including ones that don't exist yet. Biology will never be the same.
Google DeepMind's AlphaProof cracks the Keller Conjecture variant. Mathematicians are both amazed and unsettled.
Traditional drug discovery takes 10 years. AI found a novel antibiotic compound in an afternoon. Clinical trials start next month.
DeepMind's SIMA agent completed Minecraft in 12 minutes using strategies no human has ever discovered.
The model can analyze scientific papers, charts, and lab imagery together.
Multi-agent systems are solving problems single AIs can't touch. Here's why everyone from startups to OpenAI is going all-in on swarms.
Eye on AI explores the architecture that might replace autoregressive transformers. Diffusion models already dominate images. Language could be next.
MIT Technology Review named mechanistic interpretability a 2026 breakthrough. Anthropic's 'microscope' and OpenAI's chain-of-thought monitoring are revealing how AI actually works.
GPT-5 Pro scores 18.3% on the new benchmark. The previous version? 70.2%. Francois Chollet's test exposes what AI still can't do — and it's not what you'd expect.
From security cameras to surgical rooms, real-time video comprehension is unlocking applications that were science fiction a year ago.
Five years ago, GANs were the future of generative AI. Now they're a footnote. What happened?
A controversial new theory locates awareness in the claustrum—and early tests are surprisingly supportive
Forget chatbots. The next frontier is AI that can simulate entire 3D environments from a single image.
AI is quietly revolutionizing climate science. Here are five developments that actually matter.