Wow, this is massive and unbelievable. 50 subscribers. i’m super thankful to all of you, seriously. it feels like i’m just getting started, and there’s more fun on the way. btw, those video clips you see on my feed every day? yes, they were edited by my junior (i paid him) so

gpt-5.1-codex-max is my favorite right now it's super thorough and a bit robotic in a good way it double-checks everything. opus 4.5 is better for general software dev work, but it feels a bit less strict about details i guess codex is a smaller model in terms of active

one of the first things we can benefit from AI productivity, reducing the standard work week from 40 hours since 40 hours is just a random number. could cut it down to 30 hrs, and anything over 30 should be paid at time-and-a-half first step toward an AI-powered economy

Jensen Huang predicts that within 2–3 years, 90% of the world's knowledge will be generated by AI Right now, knowledge is created, shared, and modified by humans Soon, most of it will be synthetic — created by machines, mixed with human input, some true, some not "that's

about 6 months ago, the entire coding world changed radically with the introduction of agent-first frontier models and agentic workflows in common IDEs especially after the claude code and codex. there's really no going back now it feels like a 10x or more productivity jump

pretty sure Polymarket flipped because another openAI model is dropping this month the openAI folks have gone pretty quiet, which usually means they're deep in launch mode we'll see how strong it is, but at the very least, it should beat opus 4.5 i guess claude / gemini will

one of the first things we can benefit from AI productivity, reducing the standard work week from 40 hours since 40 hours is just a random number. could cut it down to 30 hrs, and anything over 30 should be paid at time-and-a-half first step toward an AI-powered economy

both openAI and anthropic are basically trying to automate the work of AI researchers if models can do what top researchers do, they probably see that as the clearest path to AGI and that is why both are so focused on software engineering and why openAI trains models for math

these models are so good now that what really matters is how much you vibe with them, not small benchmark gaps i like gpt-5 no-nonsense personality, but others might prefer claude, grok, or gemini friendlier vibe so a small bump on an already saturated math benchmark won't

Nick Bostrom says the future must be good for humans, animals, and the digital minds we create Most minds will soon be digital, and we might even upload ourselves to become digital too The challenge is that we already struggle to empathize even with animals that can see and

Jensen Huang predicts that within 2–3 years, 90% of the world's knowledge will be generated by AI Right now, knowledge is created, shared, and modified by humans Soon, most of it will be synthetic — created by machines, mixed with human input, some true, some not "that's

about 6 months ago, the entire coding world changed radically with the introduction of agent-first frontier models and agentic workflows in common IDEs especially after the claude code and codex. there's really no going back now it feels like a 10x or more productivity jump

pretty sure Polymarket flipped because another openAI model is dropping this month the openAI folks have gone pretty quiet, which usually means they're deep in launch mode we'll see how strong it is, but at the very least, it should beat opus 4.5 i guess claude / gemini will

both openAI and anthropic are basically trying to automate the work of AI researchers if models can do what top researchers do, they probably see that as the clearest path to AGI and that is why both are so focused on software engineering and why openAI trains models for math

these models are so good now that what really matters is how much you vibe with them, not small benchmark gaps i like gpt-5 no-nonsense personality, but others might prefer claude, grok, or gemini friendlier vibe so a small bump on an already saturated math benchmark won't

Nick Bostrom says the future must be good for humans, animals, and the digital minds we create Most minds will soon be digital, and we might even upload ourselves to become digital too The challenge is that we already struggle to empathize even with animals that can see and

google is ahead because gemini is built into software that billions of people already use every day gmail, docs, drive, and google search with AI mode plus tools like notebookLM, etc. so yeah, benchmarks don't matter if nobody uses your product that's the real code red for

wow... 4-month-old startup "Axiom" solves 9/12 problems on the 2025 Putnam with fully formal Lean proofs that's roughly 1 out of 4000 — Putnam Fellow level, on a test where the median score is 0 no train-on-test or benchmaxxing either. feels like a real first step toward an

i believe scaling LLMs will eventually get us to something that behaves like AGI but it would be very inefficient. we'll still need breakthroughs to make AGI work as efficiently as the human brain or better that's basically demis view, and yann too, though he focuses on world

gpt-5 is mostly about getting performance and efficiency to scale google already solved scale and efficiency years ago, so they can focus more on pushing raw capabilities gpt-5 feels like a reset for openAI a clean, solid base they can build on from here

Jensen Huang predicts that in 10 years, the energy required for AI will be utterly minuscule Today, energy is the bottleneck In 6-7 years, the current energy bottleneck will be solved by small nuclear reactors, effectively turning tech companies into their own power generators