Kvíz spojený s průvodcem · Těžko úroveň
Řídké automatické kodéry pro kvíz extrakce funkcí
Studujte řídké slovníky funkcí a rekonstrukci, s experimentem Golden Gate Bridge, přeplňte latentní, řídící a modelové hranice.
Otázka 1 z 8
Which interpretability difficulty motivates sparse dictionary learning for neural activations?
Náhled každé otázky8 otázky
Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.
- Which interpretability difficulty motivates sparse dictionary learning for neural activations?
- Every model activation is already a complete dictionary definition
- Several learned signals can overlap in an activation space
- The tokenizer must be permanently removed
- All neural weights are stored as image pixels
- What does an SAE sparsity constraint encourage for a given input?
- Every feature to activate equally on every input
- A mandatory removal of its decoder
- A guarantee that every feature has exactly one perfect human label
- Only a limited subset of learned feature activations to be nonzero
- Jaký byl slavný příklad funkce v práci Anthropic „Scaling Monosemanticity“?
- Funkce Golden Gate Bridge
- Funkce pro číslo sedm
- Funkce pouze pro modrou barvu
- Funkce pro samohlásky
- What does an overcomplete SAE dictionary mean?
- It must activate every feature for every token
- It contains no reconstruction mapping
- It has more learned feature directions than input activation dimensions
- It always requires less storage than the input vector
- What does a monosemantic feature label aim to describe in this research context?
- A guarantee that the feature never responds to related examples
- A feature associated with one interpretable concept
- A language model that has only one parameter
- A token represented by exactly one raw byte
- Jak lze SAE použít k ovlivnění chování modelu?
- Zapnutím nebo vypnutím funkce řídit výstupy
- Přeškolením celého modelu od nuly
- Smazáním tokenizeru
- Převedením na obrazový model
- What does an SAE’s reconstruction residual measure for an activation vector?
- The difference between two unrelated document titles
- The total number of words in the model vocabulary
- Only the model’s hardware temperature
- The difference between the original activation and its decoded reconstruction
- When an SAE is fitted to collected activations of a frozen language model, which weights need to be trained for that fitting?
- Every original language-model weight must be updated
- Only the tokenizer’s character list
- The SAE’s own encoder and decoder weights
- No parameters of the SAE can change