Bottom line: evaluation with decision rights is governance. Without them it's transparency. Worth being clear which one is on the table. (Views my own.)
Everyone's covering the Zuckerberg vs. Amodei split as speed vs. safety. That's not the fault line. Both want independent evaluators in frontier labs. The disagreement is whether the evaluator holds a brake or a dashboard light.
The compromise is a risk-tiered assurance regime, lab by lab. No pact, no antitrust problem. Zuckerberg keeps his pace. The pacing camp gets a gate that actually closes.
Today, the FDA approved Revolution Medicines’ daraxonrasib (Rasonque) for certain adults with previously treated metastatic pancreatic ductal adenocarcinoma.
In Phase 3, median overall survival was 13.2 vs 6.7 months with chemotherapy, a meaningful advance in a hard-to-treat disease #PancreaticCancer #Oncology #PrecisionMedicine
Bill Gates proposes "Human Reserved" jobs: roles we keep for people the way we keep land for nature reserves. I think it's a good idea.
I'd add one layer. Reserving the role isn't enough. We also have to reserve the way people become capable of filling it.
The human in the chair in 2040 only matters if she brings judgment the machine didn't build for her.
Reserve the judgment along with the job.
(Views my own.)
Congress is asking Open AI for:
model versions
prompts
scoring incentives
logs
monitoring status
reasoning traces
risk classifications
prior boundary violations
human stop points
That’s a real shift.
Agent governance is moving from policy declarations → operational evidence.
Translation: If your AI can act, it needs a flight recorder.
OpenAI just started rolling out free frontier AI access to researchers ,scaling from 10,000 now to 100,000 by 2027. 🎓
Same month, Anthropic gave every verified US K-12 teacher free Claude through 2027.
Same month, China pledged 5,000 AI training seminars for developing nations over the next 5 years.
Three different bets on who owns the next generation of AI users. Whoever wins the research and education community may end up owning enterprise credibility down the line. TBD🤔
#OpenAI #Anthropic #AI #Research
Prepared by Deep Research
Microsoft just launched “Project Perception” a AI cybersecurity system that uses AI to find and fix vulnerabilities. 🛡️🤖 The pitch: “AI that protects you from AI.” Launches in direct response to the Hugging Face hack. The irony: AI created the problem AND the solution. #Cybersecurity #Microsoft #AI
@samhogan Speed-date the new model? Test the chemistry on your real workload. Huh? Rehearse the breakup. Keep the rollback plan. 😂🔥...All jokes aside...brilliant!