🤖 Harvard study: AI ER diagnosis beats two doctors
Audio in Mandarin Chinese · English transcript below
⚡ Harvard AI ER outruns docs in a flash, MIT spills scaling bottlenecks, Musk & Altman IPO catfight, NASA crams 96GB GPU, copyright war tears subway posters, face-recognition turns you into a ‘thief’, prompt hacks drop, DIY software goes mass-market—AI is becoming a whole new science!
Today's Top 3 Headlines
- AI Industry News
🤖 Harvard: AI ER triage 67% accurate, beats docs by 10–17 pts
Science: OpenAI o1 scored 67% on 76 ER triage cases, beating two attendings’ 55% & 50%. Shows LLMs can aid decisions amid sparse data, yet lacks accountability & prospective validation.
Source ↗ - Technology
🤖 Dev prompt trinity: neg constraints + JSON + multi-hypothesis sampling
JustSayAI’s negative-constraint, JSON-structured, multi-hypothesis sampling lifts GPT-4 factual accuracy from 72% to 91%. Developers now get reproducible, parseable, high-trust answers with fewer tokens, cutting production LLM risk and debug cost.
Source ↗ - AI Industry News
🤖 “It’s OK” pup hijacked in subway ad; artist slams “AI theft”
KC Green told TechCrunch that Artisan’s subway ads turned his “This is fine” dog into “my pipeline’s on fire” without permission. Creators now face AI ads weaponizing public memes, sending legal takedown time costs soaring.
Source ↗
+9 more headlines
- 🤖 Facewatch UK stores: 50k AI alerts/mo, 99.98% "accuracy" still nabs wrong shoppers
- 🤖 MIT: LLMs stabilize at scale—vector superposition packs 10k concepts
- 🤖 75k AI songs/day, 34% of Deezer's new tracks now non-human
- 🚀 QNAP NAS: Zen 2 EPYC+96GB RTX PRO 6000 Blackwell runs DeepSeek
- 🤖 DeepMind hires Cambridge philosopher Henry Shevlin for AI ethics
- 🤗 ChatGPT’s real menace is temptation, not ruin—fiduciary duty is the fix
- 🤖 Columnist builds personal dashboard with AI in 3 days
- 🤖 Musk vs Altman: OpenAI IPO route & AI monetization feud heats up
- 🤖 DeepMind CEO: AI self-mechanism study to spawn new science