The AI-collapse pre-mortem - Bert Hubert's writings
a day ago
- AI predictions from 2023 were largely accurate but underestimated the ensuing hype magnitude.
- LLMs are not truly intelligent but perform 'absolute miracles' in tasks like grammar correction and code debugging.
- The Chinese Room Argument remains relevant for debating AI intelligence vs. simulation.
- LLMs are compared to a chess-playing pigeon—impressive that they function at all, but often still useless.
- AI is not on a clear path to general intelligence, yet it won't disappear even if the bubble bursts.
- Non-LLM AI, like Whisper for transcription and AlphaFold for protein folding, shows super-human performance.
- AI in radiology and other fields will surpass human capabilities, especially in resource-limited setups.
- The AI bubble collapse does not invalidate the technology's achievements; it's not a measure of long-term potential.
- AI's progress in pattern matching (e.g., DNA splicing prediction, language translation) is unmatched.
- Post-collapse, an open mind is needed to discover genuinely useful applications beyond current hype.