Category: AI, Computation, and Technology

RT @curtdoolittle: @RussellJohnston @templexaciounes RE: AI HALLUCINATIONS – WE ACTUALLY WANT THEM?
Correct. Though interestingly… most h…

RE: AI HALLUCINATIONS – WE ACTUALLY WANT THEM?
Correct. Though interestingly… most hallucinations are due to the probabilism inherent in the ‘temperature’ of a given LLMs selection of a sequence of terms.

This would end if we took the opposite position (musk, tesla, driving) where we are testing against real world models (operational possibility) instead of linguistic (analogical) probabilities.

However, just is as there is a difference between testimony (truth) which we want from AIs, and the ideation (ideas) which we want from AIs, there is a difference between whther a hallucination consists of unwanted error or hallucination consists of wanted ideation. 😉

In other words we do want hallucination – just only when we want ideation..

Reply addressees: @RussellJohnston @templexaciounes

Correct. I use AI to find holes in my prose (-), and make it more accessible (+). That means I lay out the logic of an argument in my usual parsimony. And I ask it to compose it for a postgrad audience. Then I ask it for what I’m missing. It usually takes a few iterations and corrections but that’s it. Mostly I have to correct the natural imprecision of the AI’s given that they’re llm’s using probabalism to select words.

Reply addressees: @templexaciounes

RT @LukeWeinhagen: @bryanbrey @gregisenberg @curtdoolittle @WerrellBradley @ThruTheHayes Emergent dependence.

Failure to integrate the new…