Posts are bits of writing published here and not in other venues. For external publications, see my cv or Google Scholar
Until we answer this question, we cannot carry out an ethics of AI.
The world is suffuse with general intelligences, but science is still stuck. Generality is a red herring on the path to superintelligence.
Language is Not a Compression of Something Else
The popular view that perception is lossless and language is lossy is simply false: language can make more distinctions than perception. This has implications for those desirous for models to reason "in latent space".
Relative to pre-linguistic models, LLMs reduce epistemic risk for cultural heritage research.
Image Captioning in Hostile Conditions
We test 10 large VLMs on a difficult collection of historical photographs.