Playing Around With chatGPT
Scott Mcleod,
Dangerously Irrelevant,
Dec 14, 2022
Your chatGPT and AI update for today: like much of the rest of the world, Scott McLeod is playing around with chatGPT (I think he should comment on the output rather than just generating it). InfoQ offers an authoritative summary description of the technology. Alternatively, Richard Byrne's less authoritative but much shorter summary. There's also a Zotero group for articles about chatGPT (most of which has also been linked here). Finally, the gap in copyright law around generative models.
Related: Michael Webb examines GPT and plausible untruths: "when you realise that actually it's 'just' predicting the next word in a sentence, like some supercharged auto-complete tool, some of its weakness make more sense." It's arguable that nobody knows how it works. "Any innovations unique to it are secret. OpenAI could well be trying to build a technical and business moat to keep others out," writes Toby Walsh. Also, Melanie Mitchell channels Nick Bostrom while asking what it means for AI to align with human values: "it's not clear whose values we should have machines try to learn." There's a full-length book (173 page PDF) exploring various views on the issue.
Finally, people dismiss chatGPT and similar products saying things like "it's just statistics and machine learning." Exactly. Wait until real AI takes hold, as described in this article on deep learning and product design. But more to the point, AI is finally good at stuff, and that, writes Rebecca Heilweil, is the problem. "GPT is a stark wakeup call that artificial intelligence is starting to rival human ability, at least for some things." Like writing essays. But as Heilweil points out, students were using aids and ghost-writers before GPT. The issue now is that everyone can do it, not merely the wealthy and well-connected. Image: OpenAI.
Today: 3 Total: 3 [Direct link] [Share]
