The decomposition of the auto-researcher loop is the cleanest read on where LLMs sit in the value chain of AI research itself. Implementing experiments and tuning hyperparameters have already commodified; choosing the next question and escaping dead-ends are still where the human researcher carries the day. The KataGo reference makes the gap material: David Wu and Jane Street took roughly an order of magnitude of compute out of training a Go bot in 2020, and a single builder can now rebuild that stack with rented compute and LLM coding. The same asymmetry between commodified execution and sticky taste is what enterprise AI buyers are running into in production rollouts.
MR.Dwarkesh,The Acclaimed Global AI Technology Leader & Expert,Renowned Author, has done a whole series of excellent video interviews with top AI researchers and engineers, and the instant one—Eric Jang – Building AlphaGo from scratch—brings two Greatest AI minds—and,the podcost is a great distillation of their conversations that helped me to understand modern AI models, how they are built, where they are headed, and why they are important.
Perfect video, you can feel Eric's depth of understanding throughout. Incredibly fortunate to have gotten his time on the pod, if only he could be a recurring guest!
The decomposition of the auto-researcher loop is the cleanest read on where LLMs sit in the value chain of AI research itself. Implementing experiments and tuning hyperparameters have already commodified; choosing the next question and escaping dead-ends are still where the human researcher carries the day. The KataGo reference makes the gap material: David Wu and Jane Street took roughly an order of magnitude of compute out of training a Go bot in 2020, and a single builder can now rebuild that stack with rented compute and LLM coding. The same asymmetry between commodified execution and sticky taste is what enterprise AI buyers are running into in production rollouts.
MR.Dwarkesh,The Acclaimed Global AI Technology Leader & Expert,Renowned Author, has done a whole series of excellent video interviews with top AI researchers and engineers, and the instant one—Eric Jang – Building AlphaGo from scratch—brings two Greatest AI minds—and,the podcost is a great distillation of their conversations that helped me to understand modern AI models, how they are built, where they are headed, and why they are important.
Perfect video, you can feel Eric's depth of understanding throughout. Incredibly fortunate to have gotten his time on the pod, if only he could be a recurring guest!
Amazing stuff, please continue this blackboard series. Highly premium content available for free.