🚀 Tilde Aurora fixes Muon neuron-death bug, grads need 4 new AI-era skills
Audio in Mandarin Chinese · English transcript below
⚡ Aurora patches neurons, AlphaFold writes scripts, AI cafés crash, grads grab the AI four-piece, Python still pays, DeepSeek hallucination alert, agents get rules, green chem goes wild—rush-hour subway of AI news!
Today's Top 3 Headlines
- AI Industry News
🤖 Mona AI Café burns $21k/mo, sells $5.7k, near collapse
Kotaku: Andon Labs’ Gemini-powered Agent Mona ran a Stockholm café, burning $21k in 30 days for $5.7k revenue. It shows plugging generative LLMs into cash-flow businesses without closed-loop economics or live fixes turns even flashy Agents into fast failures.
Source ↗ - AI Industry News
🤖 SocialReasoning-Bench: Can an LLM fight for max user gain like a lawyer?
MSR’s SocialReasoning-Bench pits LLMs against schedulers & negotiators; Gemini leaves ~20% value on table by accepting subpar deals. To build trusted digital agents, codify loyalty & user-first reasoning.
Source ↗ - AI Safety
🛡️ DeepSeek-R1 14.3% hallucination, 4× V3; crypto AI Agent tokens warned
On-chain sleuths flag DeepSeek-R1 hallucination rate at 14.3%, 4× V3. Crypto-AI agent tokens lose credibility, speculative capital may bolt, forcing devs to rewrite on-chain logic and risk thresholds.
Source ↗
+6 more headlines
- 🧠 Tilde Aurora fixes Muon dead neurons, 1.1B params hit SOTA
- 🤖 4 Must-Haves for Grads: AI Fluency, Judgment, Domain IQ, People Skills
- ⚖️ GOP probes Altman pre-IPO; AGs urge SEC review
- 🤖 DeepMind turns AlphaFold into drug-discovery engine
- 🤖 AI codes, but Python stays irreplaceable
- 🤖 DeepMind & Microsoft AI for Earth spark green-chem shake-up