Video wird geladen...
Video konnte nicht geladen werden
Mustafa Suleyman on Microsoft's new multi-agent "diagnostic orchestrator": The aim is to simulate AI as a medical diagnostician -- it asks patients questions, reviews tests, and keeps improving its diagnosis we trained it using hundreds of case histories from the New England Journal of Medicine, used as weekly doctor... show more
41,349 Aufrufe • vor 1 Jahr •via X (Twitter)
11 Kommentare

video source complete detail about Agent

Scan any documents, convert images into text, PDF files, etc. 👍

Multi ai agent is the real future of agent

"it's like an ultimate crossword for doctors" - what does that sentence mean?

Alpha tech, WAGMI

Fascinating! But I wonder about the ethical implications of AI diagnostics. Will patients trust machines with their health concerns?

How does it perform on healthbench? The lack of such results raises questions as to whether this is better than o3

To scale Perplexity, CEO Aravind Srinivas says large models now label, evaluate, and train smaller ones. The system handles millions of queries daily. Human evaluation is no longer the bottleneck. “An AI has to do this job.”

For whom it may concern: Scientists found a molecule that can cure baldness by waking up dormant hair folicles

🚨 BREAKING: While doing some deep dives on Figure, I discovered that the venture firm leading Figure’s Series C funding has quietly posted Figure’s Version 3 model directly on their website before Figure has officially announced or revealed it. Big leak? 👀

Grok 4’s leaked benchmark just dropped a bombshell. The chart shows xAI’s model hitting a jaw-dropping 45% on HLE (Humanities Last Exam) — an exam so brutal it was designed to keep LLMs humble: •2,500 expert-written questions across 100+ disciplines •14 % multimodal (text + images) •Anti-memorization traps and a hidden test set to block “cheat-training” For context, most frontier models stumble far below this score. If the leak holds up, Grok 4 just cleared one of the toughest gauntlets in AI benchmarking.
