A 10-layer neural network can compress what looks like an NP-class search problem into a single forward pass. This happened with Go, protein folding, and tensor decomposition. It might mean our understanding of computational hardness is fundamentally incomplete.
Where this was said
At 1:51:30 · chapter starts 1:41:09
Eric and Dwarkesh discuss why MCTS fails for LLM reasoning: language's vast token space violates the discrete finite-action assumption, and value estimation is much harder than in Go. [1] — Eric Jang "Language has billions of possible next tokens — the PUCT exploration heuristic assumes you'll visit the same node multiple times, but an LL…" 1:44:55
Scaling laws only emerge cleanly when your system already works, your data is good, and there are no bugs. Trying to extract scaling insights before you have a working baseline just gives you scaling laws on garbage.
Unlike naive RL which must figure out which of 100K+ tokens caused a win, MCTS provides a strictly better action target for every single move in every game.
Language has billions of possible next tokens — the PUCT exploration heuristic assumes you'll visit the same node multiple times, but an LLM will almost never generate the exact same token sequence twice. The discrete action assumption breaks.
Eric Jang trained a strong Go bot for roughly $10K of donated compute from Prime Intellect, spending about $3K on the final training run.
AlphaGo Zero was trained on far more compute than any other AI model of its era — roughly 3e23 flops, comparable in order of magnitude to frontier LLMs.
Pre-training on 9x9 Go boards and then warm-starting a 19x19 model dramatically cuts the time needed to learn endgame value functions.
Social norms compress behavior so effectively that we assume other minds are similar to ours. They almost never are. After running over 100 psychology studies, Greenberg's biggest finding is that humans are far more different from each other than we intuitively believe.
The widely shared claim that we have a credit card's worth of plastic in our brains may be a massive overestimate. Researchers used nitrile gloves to measure microplastics — and nitrile gloves shed microplastic particles onto the measuring device. A second paper found the original tool used was unreliable, and that we may have far less plastic than reported.
RSD isn't just hurt feelings — it's a dopamine-deficient brain with a hyperactive amygdala and a frayed connection to the prefrontal cortex. Add a surge of norepinephrine and you get chest pain, sweating, and trembling that convinces your body something catastrophic is happening.
In Dan Ariely's door experiment, participants sacrificed the highest-reward door — repeatedly — just to avoid seeing other doors disappear. Even when researchers trained them to use the best door, they kept wasting clicks to maintain options. This is the bias that runs our lives: we'll take a worse outcome to preserve the feeling of optionality.
Dan Gilbert discovered that the brain has a 'psychological immune system' — it misremembers, reframes, and rationalizes to keep you satisfied with whatever you chose. The fear before commitment is all about potential regret. But after committing, your brain quietly starts justifying the choice — and the more irreversible the decision, the stronger the effect.
The psychological immune system can get you through the first few months of any commitment. But durable, long-term well-being requires self-concordance — choosing commitments that are genuinely aligned with your values and what you actually want. Without that alignment, the satisfaction fades fast.
Vision didn't just help animals find food — it ignited the Cambrian explosion of speciation. Half of the human cortex is devoted to visual processing, and that same visual hierarchy directly inspired the neural network architectures powering today's AI.
Peptides come loaded with a powerful narrative: regenerative, subcutaneous, targeted, cutting-edge. That narrative can genuinely move outcomes, especially for pain, energy, and recovery. The RCT's job is to quantify how much additional benefit comes from the molecule itself — above and beyond the compelling story surrounding it.
A pharmaceutical is not an amino acid sequence — it is the successful solution to manufacturing, purification, formulation, and consistency challenges. When clinical trials prove a drug works, they validate a specific product under specific processes, not every preparation that shares the same sequence.
We use essential and analytics cookies to run Vuci. To understand how the site is used: Privacy Policy.
Install Vuci on your phone
Add it to your home screen for a faster, app-like experience.
Install Vuci on your phone
Tap the Share button, then “Add to Home Screen”.
A new version is available
Reload to get the latest Vuci.