What is the largest annual revenue a single company will get via mechanistic interpretability of deep learning models in 2028?
InflationOne-Off2a
Manifold MarketsSans KYCDonnées de résolution vérifiéesBien calibréPrévision communautaire actuelle
x >= $1B 51.9%En tête parmi 6 options
x >= $1B 51.9%Prévisionnistes
18
Type de question
multiple choice
Méthodologie
Play-money forecasting platform
Type de source
Prévision
Variation 7j
0.0 pts
Données du marché
Mis à jour il y a 3 heures
Obsolète18 déc. 22, 21:51 UTC31 déc. 28, 23:59 UTC
Tendances
Règles
Resolves according to my best estimate of revenue in inflation-adjusted 2022 USD on January 1st, 2030.
- I will include and try my best to estimate returns for difficult and illegible cases.
- Mechanistic interpretability is here meant to mean 'any work that is about mechanistically understanding in fine detail what the computations in a neural network are doing'.
- One test of this is if after the work, someone is able to erase and rewrite large parts of the model while maintaining performance, although that is a high bar that doesn't include all work I would consider mechanistic interpretability.
- I will determine what work is/isn't mechanistic interpretability subjectively.
- Examples of possible sources of revenue: - interpretability consultants producing evidence that an AI system is not discriminating based on race for regulators of e.g. insurance. - interpretability is used to specify what humans want from an AI system in a way that improves revenue of an AI lab. - interpretability is used to understand what an AI system has learned about some problem/system so that a company can directly use or sell the knowledge.
