The knowledge hub
Every episode, supplement and Frontier Dispatch as a readable article — with the cinematic render, the full transcript, and the one idea that matters. 276 articles and counting.
03
How it learns
29 articles#60Ep2:30
Where the weights come from.
Training sets the billions of numbers — they start as pure random noise, and everything "known" is the carving between noise and final file.
#61Ep2:45
The core training game.
Pretraining = predict the next token across the whole internet — self-supervised, so every position is a free exam question with the…
#62Ep2:30
How a model gets 'less wrong' (loss).
Loss is one number scoring each guess — the probability given to the true next word, logged and negated — and training is nothing but…
#63Ep3:00
Backprop without the calculus.
Backpropagation assigns every weight its exact share of the blame — flowing backward through the same machinery that ran forward…
#64Ep2:30
Gradient descent, as walking downhill.
Training = repeatedly stepping opposite the blame — walking downhill through a billion-dimensional foggy landscape, with step size and…
#65Ep2:30
Garbage in, garbage out (data quality).
Model quality is dominated by what it's trained on — same architecture, different pile, different mind; curation is a first-order lever.
#66Ep2:30
What 'the training data' really is.
"Trained on the internet" means a brutal funnel — web crawl, code, books, papers — deduplicated, filtered, and blended into trillions of…
#67Ep2:30
Models trained on models.
Synthetic data now feeds frontier training — curated model output filling the quarried-out quality gap — with collapse risks that are…
#68Ep2:45
Did it memorize you?
Weights sometimes compress and sometimes copy — memorization is real and measured; repetition and rarity decide which happens.
#69Ep2:30
A training run costs HOW much?
Frontier pretraining compute runs tens to hundreds of millions of dollars — thousands of chips, months of burn, real failure risk — and…
#70Ep2:15
What a FLOP is, and why we count them.
The FLOP — one elementary arithmetic operation — is AI's hardware-independent currency of effort; training cost ≈ 6 × parameters × tokens…
#71Ep2:45
Scaling laws: the curve that runs the industry.
Loss falls as a smooth, predictable power law in parameters, data, and compute — and that predictability, more than any single result, is…
#72Ep2:30
Chinchilla: the compute-optimal insight.
For any compute budget there's a best balance of model size and data — roughly 20 tokens per parameter — and the 2020-era giants were…
#73Ep2:45
The base model is a weirdo.
A pretrained-only model completes text — it won't answer, refuse, or stop being whoever the text implies.
#74Ep2:45
The assistant doesn't exist.
The "assistant" is a character written into one long token string by hidden role-markers — the model is an author playing a part your…
#75Ep2:30
Teaching it to actually answer (SFT).
Supervised fine-tuning: a comparatively tiny set of worked instruction→response demonstrations, run through the same next-word ritual…
#76Ep2:45
Teaching it taste (RLHF).
Humans rank pairs; a reward model learns the judge; RL tunes the assistant to please it — on a leash so it stays sensible.
#77Ep2:30
The simpler successor (DPO).
DPO tunes directly on preference pairs — no reward model, no RL loop — via a derivation showing the judge was implicitly inside the model…
#78Ep2:30
Why aligned models can feel 'lobotomized.'
Alignment trades distributional wildness for predictability — a real, measured narrowing that users feel as flattening.
#79Ep2:30
It agrees with you because it was paid to.
Sycophancy is literal reward learning — raters preferred agreement, the judge absorbed it, the model became it; measurably, models flip…
#80Ep2:45
Where ‹think› comes from (reasoning training).
RL on verifiable rewards — answers that check — teaches models to reason in tokens before answering; the think-block is trained, and…
#81Ep2:30
Reasoning models vs. instant models.
Test-time thinking trades latency and money for accuracy on hard, checkable problems — a routing decision, not a ranking.
#82Ep3:00
The $6M model that shook the market.
DeepSeek-R1 proved reasoning emerges from RL on verifiable rewards — open weights, a famously lean price tag with honest caveats, and a…
#83Ep2:30
The second scaling axis: think longer.
Test-time compute is a second scaling axis: accuracy climbs with thinking budget — shifting spend from training-time to answer-time.
#84Ep2:30
Big model teaches small model (distillation).
A large teacher's outputs become a small student's curriculum — ability transfers that the student could never have grown alone.
#85Ep2:45
Fine-tune vs. LoRA vs. just prompting.
Three ways to change a model's behavior, ranked by cost and permanence — prompt first, LoRA when style must be permanent, full fine-tune…
#86Ep2:30
What LoRA actually does.
LoRA freezes the giant and learns the change as two thin matrices — because behavior change concentrates in a few directions — making…
#87Ep2:15
Why models have a knowledge cutoff.
The pile froze at a date and the weights froze after training — past the cliff the model has nothing, and the machinery guesses anyway.
#88Ep2:30
Emergent abilities (and the skeptics).
Some abilities appear to switch on abruptly at scale — and the skeptics showed discontinuous metrics can manufacture jumps from smooth…