Module FT00 — The Steering Stack

The Steering Stack

Fine-tuning steers behavior; it does not teach knowledge. The five-layer mental model that anchors the entire course — and the one judgment that prevents the cardinal error of treating fine-tuning as knowledge injection.

60
minutes
8
artifacts
4
sub-sections
Every technique in this course — SFT, DPO, GRPO, abliteration, quantization — is a steering technique. It redirects an already-capable base model. Once that lands, four mysteries become obvious: why QLoRA works at 1.5% of parameters, why uncensoring degrades math, why an uncensored model is only safe inside a harness, and why your data matters more than your algorithm.
Key Claims
Load-Bearing Claims

Fine-tuning steers behavior; it does not teach knowledge. The model already knows how to do X — it saw X during pretraining. Fine-tuning redirects the model's probability mass so it does X reliably, in your format, under your conditions.

You can swap any layer above the base without touching the one below. That is why LoRA adapters are swappable, why abliteration works without retraining, why you quantize after training, why the harness is model-agnostic. The stack is modular by design.

If the base model, with a perfect prompt, could already produce the target behavior, that's steering. If it could not, no amount of fine-tuning will reliably get you there — you need a different base, continued pretraining, or (usually) RAG.

The model steers; the harness bounds. This is the complement, not the contradiction, of Course 1's thesis. The two courses describe the same system from opposite ends.

After This Module
01
State the course thesis — fine-tuning steers behavior, it does not teach knowledge — and defend it with the LoRA-vs-full-FT evidence (intrinsic dimension, structural non-equivalence).
02
Draw the five-layer Steering Stack (Base, Adapter, Steer, Export, Boundary) and explain why each layer is independently swappable.
03
Distinguish steering (SFT, DPO, GRPO, abliteration) from knowledge injection (continued pretraining) and place each technique on the correct side of the line.
04
Apply the three-outcome test (prompt the base; observe reliability, refusal, or foreignness) to predict which layer a given goal belongs at.
Artifacts