The week in AI at a glance
-
Reasoning: what top AI models can and can’t do: A mathematician weighs in on AI models’ ability to solve FrontierMath problems.
-
If AI is linear algebra you’re a bunch of atoms: Biologist Michael Levin talks about what he sees as a mistake that people make when thinking about AI.
-
ChatGPT amnesia effect + automation bias = disaster: People will forget that ChatGPT makes mistakes and will uncritically take his word as truth.
-
The race to AGI is more heterogeneous than you think: Not everyone is pursuing OpenAI’s approach, particularly Ilya Sutskever and Francois Chollet.
-
AI has memorized so much—why hasn’t it discovered anything? Dwarkesh Patel, Scott Alexander, and Gwern weigh in on this question.
-
If you care about AI’s effect on writing, watch this: David Perell and Tyler Cowen talk about AI and the future of writing.
-
Manus is NOT a second DeepSeek moment: When something is hyped by influencers first instead of researchers—beware.
The week in The Algorithmic Bridge
-
(PAID) Weekly Top Picks #99: The xAI-DeepSeek spectrum / Grok against Elon and Trump / Tyler Cowen on slow AI take-off / GPT-5 in May / Satya on AGI and GDP / AI skeptic syndrome / Claude Sonnet 3.7
-
(FREE) xAI and DeepSeek Have Shattered the AI Industry’s Golden Tenet: They revealed that scale and efficiency are inversely correlated and to a large degree, interchangeable.
-
(FREE) GPT-4.5 Feels Like a Letdown But It’s OpenAI’s Biggest Bet Yet: In one sentence: OpenAI did not train GPT-4.5 to raise the ceiling but to raise the floor.
-
(FREE) Weekly Top Picks #100: Special “Ask Me Anything” issue.
-
(PAID) If AI Is So Great, Someone Should Tell GDP: Why is AI’s impact not reflected in the main growth and development metrics, like GDP, TFP, and HDI?
-
(FREE) The Human Toll of Waiting for AI to Take Over: An argument in favor of automating terrible jobs without forgetting about the people who dwell there.