Mit dem Guide verknüpftes Quiz · Schwer Ebene
Sparse Autoencoder für Feature-Extraktion-Quiz
Studieren Sie Wörterbücher und Rekonstruktionen spärlicher Merkmale mit dem Golden Gate Bridge-Experiment, übervollständigen Latentdaten, Steuerung und Modellgrenzen.
Frage 1 von 8
Which interpretability difficulty motivates sparse dictionary learning for neural activations?
Sehen Sie sich jede Frage in der Vorschau an8 Fragen
Während Sie am Quiz teilnehmen, werden Antworten und Erklärungen angezeigt.
- 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
- Was war in Anthropics Werk „Scaling Monosemantity“ das berühmte Beispielmerkmal?
- Die Golden Gate Bridge-Funktion
- Ein Feature für die Nummer sieben
- Eine Funktion nur für die Farbe Blau
- Eine Funktion für Vokale
- 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
- Wie können SAEs genutzt werden, um das Verhalten eines Modells zu beeinflussen?
- Durch Ein- oder Ausschalten einer Funktion zur Steuerung der Ausgänge
- Durch Neuschulung des gesamten Modells von Grund auf
- Durch Löschen des Tokenizers
- Durch Konvertieren in ein Bildmodell
- 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