2026-02-11 17:54 UTC
@sbrl's 2 laws of artificial intelligence:
1. AI will always hallucinatinate
2. AI will always be biased
AI is like a dictionary. It stores the important things in the training dataset in its memory. Then it matches against them when queried. Anything outside of what it learnt matches results in an inaccurate answer.
Data is generated by humans, who are inherently biased. We can't account for biases we don't know about.
What matters is not really AGI and media hype, but how we understand how models work, apply them appropriately, and continually work to understand and deal with bias as best we can - just like with irl.
#AI #ArtificialIntelligence #AIEthics #GenAI
Replies (1)
-
@km6ecc@mastodon.radio 2026-02-11 18:37
@sbrl 1) AGI will have to wait another generation at least, the scaling has stopped cold. (And I am not too sanguine in the long run.) 2) What people have not quite realized yet is that scaling and tuning are helping the best performing models we are able to make SCALE DOWN to run on potatoes. Basically we are stuck with liar potatoes who hallucinate. Soon so cheap to operate that, why not? The tuning is pushing toward something more like "weak AI" and it will replace quite a lot of human work.