philip lelyveld The world of entertainment technology

7Apr/24Off

AI keeps going wrong. What if it can’t be fixed?

... Today’s large language models (LLMs) have learnt to recognise patterns but don’t understand the underlying concepts. They will therefore always produce silly errors, says Marcus. ...

We don’t do very much to adapt to AI (or climate change) because we can’t quite believe upheaval until it arrives....

Even tech optimists find themselves caught out. Meta’s head of AI, Yann LeCun, told world leaders on February 13 that a text-to-video generating AI service was not possible: “Basically we don’t know how to do this.” A few days later, OpenAI revealed its text-to-video model, Sora. If you loudly say that AI will never be able to do something, there’s the chance that someone in Silicon Valley is laughing. ...

Like crypto, AI has identifiable flaws. LLMs such as OpenAI’s can’t digest all human knowledge. They are trained on sets of available data — words, images and audio, but not the direct interaction with the physical world. Even if you pump in more data, can you address the limitations? ...

“The burden of proof lies with the people making the extraordinary claims . . . No one is saying AI is hype, we’re saying that your claims of AI are hype.” ...

Gary Marcus suggests performance may get worse: LLMs produce untrustworthy output, which is then sucked back into other LLMs. The models become permanently contaminated. Scientific journals’ peer-review processes will be overwhelmed, “leading to a precipitous drop in reputation”, Marcus wrote recently. ...

(Rasenberger adds that ChatGPT and others would face restrictions on how they used copyrighted data: they would not, for example, be able to provide text in the style of a certain author.) ...

Recently scepticism has got a bad name, because of how easily its followers have veered into conspiracism: doubting credible information about climate, vaccines and Ukraine. AI scepticism has so far avoided this fate. ...

See the full story here: https://www.ft.com/content/648228e7-11eb-4e1a-b0d5-e65a638e6135

Comments (0) Trackbacks (0)

Sorry, the comment form is closed at this time.

Trackbacks are disabled.