@mmasnick The beginning and middle were excellent, and overall the article is a great antidote to the hype around LLMs. I had some issues with the article towards the end.

I’m unconvinced that everyone, or even most people, will end up confusing chatbots for people, at least not with the current line of research involving LLMs. Sure, some tech people might claim that they think there’s no difference between chatbots and humans, but for the time being I attribute this to tech bros boosting a tech bubble they benefit from, rather than anything more malicious.

And I do think it will be a bubble this time around: given the poor performance of those chatbots on tasks requiring logic, let alone those requiring real-world knowledge, they will fail to meet the vast amount of hype that’s been building behind them. (There is a huge risk that this will lead to larger amounts of spam, phishing, and short-form disinformation, but much of that problem is an issue of sheer quantity rather than inability to distinguish between man and machine.)

I am also not convinced the article’s discussions of fascism in this context are likely; in fact, I almost want to invoke Godwin’s Law, though I admittedly do not have either the knowledge or the confidence in my social awareness to conclusively do so When it comes to fascism I am far more worried about real world strong men like Putin, Orbán, and on our soil, DeSantis, who are a clear, and current, authoritarian threat.

@mmasnick There is compelling evidence that ChatGPT *was* trained on books under copyright, because it can be prompt-engineered into emitting them as outputs.

E.g., https://medium.com/@neonforge/chatgpt-copyright-concerns-and-potential-legal-consequences-for-openai-56feb6974c27