The oldest marketing play in AI is to build an impressive special-purpose system (e.g. play this one game) then leverage the illusion that this task-specific capability has general relevance in order to raise money, sell something else, negotiate an acquisition, etc.
Because task-specific skill and general intelligence are orthogonal to each other, disappointment always follows.
@fchollet *cough*deepmind*cough*
@fchollet It wasn’t even pretense about It. Everything unrelated to It was marketed as It.
@fchollet Could be worse, like creating a data labeling company and calling it an AI company.
@fchollet Counter example: general language models have been great to demo a specific purpose and also have far-reaching applications. Ultimately, it depends on how much manual bias is inserted into the tool & the statistical “closeness” of other tasks.