Weekly Top Picks #96
OpenAI's Operator and Deep Research / o3-mini for free / Fully automated companies / Tell your AI to wait / A shield for universal jailbreaks / Dylan and Nathan on Lex Fridman
The week in AI at a glance
OpenAI’s first agents: Operator and Deep Research: OpenAI’s new AI agents, Operator (web actions) and Deep Research (in-depth reports), mark a shift toward specialized AI, with Deep Research proving more immediately useful.
The DeepSeek effect: OpenAI o3-mini is available for free: o3-mini is capable and cheap; it’s also OpenAI’s first reasoning model to be in ChatGPT’s free tier.
What fully automated firms will look like: “Everyone is sleeping on the collective advantages AIs will have, which have nothing to do with raw IQ.”
Tell your AI to wait and see how it becomes smarter: You can improve reasoning performance just by telling the model to continue thinking.
Constitutional classifiers to defend against universal jailbreaks: Anthropic has a new method to stop any attempt at making AI do what it “shouldn’t” do.
Dylan Patel and Nathan Lambert on Lex Fridman: Two big names in the AI industry chat about DeepSeek, China, and more for 5 hours.
The week in The Algorithmic Bridge
(PAID) Weekly Top Picks #95: DeepSeek-R1 = o1 / OpenAI and FrontierMath / Writers coming to grips with AI / Sam Altman's hype-dehype tactic / Video deepfakes / Chollet x (DL + GOFAI)
(PAID) DeepSeek Is Chinese But Its AI Models Are From Another Planet: DeepSeek’s sudden rise with its open-source reasoning model R1 challenges US AI dominance, proving that China can build world-class models quickly and cheaply, narrowing the gap with OpenAI and raising pressing questions about AI's geopolitical future, the sustainability of closed vs. open-source approaches, and whether Western labs can maintain their lead.
(FREE) 16 Musings on AI’s Impact on the Labor Market: AI’s impact on the labor market will be disruptive in the short term but ultimately transformative, with job losses giving way to new roles, increased value for human-made work, and a shift from fearing AI to integrating it—though timing, inequality, and individual adaptability will determine who benefits most.
(PAID) 7 Implications of DeepSeek’s Victory Over American AI Companies: 1) The greatest geopolitical conflict of our era. 2) DeepSeek models rival the best in the West. 3) Blowing it open through radical transparency. 4) How do you outplay a culture you don’t get? 5) Export controls failed to stop DeepSeek’s growth. 6) Lower costs to succeed in the US market. 7) DeepSeek is simply an exceptional AI lab.
(PAID) Debunking 10 Popular Myths About DeepSeek: DeepSeek-R1 is more lobotomized than US models. 2) With just $5 million, DeepSeek can achieve what OpenAI needs billions to do. 3) DeepSeek’s cheap achievements mean the scaling laws don't work. 4) Nvidia’s $600 billion crash happened because the stock was overpriced. 5) Compared to U.S. AI labs, DeepSeek’s team is tiny. 6) The app takes all of your private data. 7) DeepSeek models are open source. 8) DeepSeek is Liang Wenfeng’s side project. 9) The hardware export controls don't work. 10) DeepSeek is China’s “ChatGPT moment.”