Over the last year I have read a bunch of articles about AI, by which people mostly mean LLMs, and when I recently needed to quickly share the most formative pieces, these four articles are what jumped to mind first:
- ChatGPT Is a Blurry JPEG of the Web by Ted Chiang – I think it is notable that this piece comes from a fiction writer: he tells a story with a deft weaving of analogy and technical detail that gets at the spirit of the thing. A good starting point for why all of this is less impressive than it looks.
- Pop Culture by Ed Zitron – coming from the business and industry side of things, Zitron digs into a June 2024 report from Goldman Sachs on AI productivity, financial returns, and power grid demands. (The power usage is frankly horrifying.)
- When ChatGPT summarises, it actually does nothing of the kind by Gerben Wierda – a bit of a detailed deep dive into what actually happens when you ask a LLM to summarize an extended text. Turns out, summarizing is hard!
- ChatGPT is bullshit by Hicks, Humphries, and Slater – an actual academic philosophy paper to round things out. I like this one because it digs into the question of “hallucinating”, which is a term that they argue gives this technology too much credit for actual intelligence. Also I appreciate the use of “bullshit” as a formal philosophy term.
Bonus article: The LLMentalist Effect: how chat-based Large Language Models replicate the mechanisms of a psychic’s con by Baldur Bjarnsen – another “how does it work”, this time from the perspective of a software developer, with a very clear metaphor as in the title. This one gets into how AI hype happens, and why it grips in to particular tendencies in human psychology.
Together these all paint a very specific (and bleak) picture, and other things I’ve read mostly just fill in the blank spots. I do also recommend the tumblr Why AI for that filling in and for excellent shit-posting. Because you might as well get a few laughs in.
