Sources: chatgpt

One of several antidote articles showing the limitations of LLMs in action. This is a good example of not so much technical limitation as cognitive limitation, a useful counterpoint to the many commenters who get easily captivated by the apparent facility of LLMs with natural language, tending ineluctably toward anthropomorphism and onward to predicting the enslavement or extinction of humans by our new AGI overlords.

Cory Doctorow applies his usual cutting analysis to the AI hype that has overwhelmed the common-sense of most commenters, especially tech enthusiasts and a few credulous academics. ChatGPT is new and impressive so it shouldn’t be surprising that media are overreporting; what is surprising is the unself-conscious and unreflective attention from people who should know better. The technology is new and we don’t know what it is for yet and, as usual, we have to wait to find out. What isn’t going to happen immediately is the extinction of humans by chat-bot AGI, but what probably will happen is corporate hacks will figure out ways to use it to put pressure on labour and inflate their ability to extract capital from credulous legislators.

This is exactly the kind of thing Cory is good at reminding us to be careful of and this is a pretty good shot across the bow of AI hype credulity.

There is so much writing on ChatatGPT and LLMs that it is hard to keep a list — something I started to do in a desultory kind of way but quickly got overwhelmed. This example is one I like. It is provocative without falling for casual anthropomorphism, end-of-the-worldism, or goofy enthusiasm. It does, however, go in for some goofy testing of ChatGPT with literary device experimentation and some conceptual guidance by way of Claude Lévi-Strauss. This is a good one to include for open-minded fun and education.