There is currently much debate about how impressive frontier AI models really are and what role they can play in research, including in academic philosophy. To get some evidence, I asked GPT-6 Astra, OpenAI’s new model, to review my book, ‘The Social Roots of Delusions
I have just put this on my work group chat. To encourage anybody labouring under the delusion that AI cannot produce first-class undergraduate work to get with the times…
On the annoying AI-isms, I recently asked Claude to supply a paragraph for a post I was writing and passed it through Pangram, which thought it was 100% AI written. I told this to Claude and asked it to rewrite the paragraph following guidelines we had established to avoid AI-isms, and the Pangram thought the result was 100% human written. I didn't use either paragraph, it was just a fun experiment.
Nice idea, and unsurprising. Also good practice to see what models produce on topics where you're an expert. (I still have this very strong aversion to read LLM content not produced in my own chats!)
Being both a 'fan' of your work and of ever-improving AI models I still could not read that properly. I can't explain it better than to say that even the best outputs lack 'crunch' or 'texture' in some way.
What I have noticed is that something that AI is still *really* bad at is high stakes social communication. It can small-talk (low stakes social) or contradict one's philosophical arguments (high stakes nonsocial), but try to ask it to write a kind and authentic "sorry, we won't hire you, but we appreciated meeting you and wish you luck" letter in the context of academic hiring, and it's not only cringe-level bad, it's actually offensive.
I have just put this on my work group chat. To encourage anybody labouring under the delusion that AI cannot produce first-class undergraduate work to get with the times…
On the annoying AI-isms, I recently asked Claude to supply a paragraph for a post I was writing and passed it through Pangram, which thought it was 100% AI written. I told this to Claude and asked it to rewrite the paragraph following guidelines we had established to avoid AI-isms, and the Pangram thought the result was 100% human written. I didn't use either paragraph, it was just a fun experiment.
Nice idea, and unsurprising. Also good practice to see what models produce on topics where you're an expert. (I still have this very strong aversion to read LLM content not produced in my own chats!)
This feels less like a review and more like a very detailed summary. I wonder what you’d get if you prompted the llm to give its opinion.
Being both a 'fan' of your work and of ever-improving AI models I still could not read that properly. I can't explain it better than to say that even the best outputs lack 'crunch' or 'texture' in some way.
Honestly, I think you (still) write better.
Very impressive. No longer feel a need to read the book itself.
What I have noticed is that something that AI is still *really* bad at is high stakes social communication. It can small-talk (low stakes social) or contradict one's philosophical arguments (high stakes nonsocial), but try to ask it to write a kind and authentic "sorry, we won't hire you, but we appreciated meeting you and wish you luck" letter in the context of academic hiring, and it's not only cringe-level bad, it's actually offensive.