Just by the by, if you want a really great SFF thriller about weaponized rhetoric, I highly recommend _Lexicon_ by Max Barry. On top of being a great thriller, at the very end (in the "summary memos") it invents a new metaphor for true love in the fullest sense of "new."
"The OpenAI explanation of why their Codex tool has to be restrained from talking about goblins gives us a clue. The chatbot style isn’t a result of the transformer neural network picking up latent features of the entire training set; it’s a feature set at a much higher level, in the “system prompts” and the optimisation of the actual product in testing and manual tweaking, rather than anything happening at the level of vector spaces."
This is wrong, I'm afraid, there's a very large human-feedback reinforcement component in all these models where annotators have labelled massive response corpuses as racist, helpful, cheerful, pornographic, difficult-to-understand etc. etc. and that's incorporated directly into the training process via e.g. https://cameronrwolfe.substack.com/p/ppo-llm. Much of the "chatbot style" is actually baked into the network weights.
A couple of weeks ago I was half-listening to a podcast when I noticed how often the interviewee would start a response with some form of "That's an excellent question...," or other compliment, presumably as a brief stall for time while they formulated their answer. I assume a lot of automated podcast interview transcripts have made it into the training sets by this point; I'm now fond of the possibility that LLMs developed this tic because humans in conversation need a second to gather their thoughts.
Poppy is an amazing young woman! Has she ever tried poetry? If she is good at patterns and mimic, I suggest she take some classics and use the form and rythym, tone, to write her own. I used to do it back in the day for kicks.
Just by the by, if you want a really great SFF thriller about weaponized rhetoric, I highly recommend _Lexicon_ by Max Barry. On top of being a great thriller, at the very end (in the "summary memos") it invents a new metaphor for true love in the fullest sense of "new."
Oh, I'll have to check that out, I loved Nation State
Sorry I mean Jennifer Government, “nation state” was the web game he created around it which I also loved
"The OpenAI explanation of why their Codex tool has to be restrained from talking about goblins gives us a clue. The chatbot style isn’t a result of the transformer neural network picking up latent features of the entire training set; it’s a feature set at a much higher level, in the “system prompts” and the optimisation of the actual product in testing and manual tweaking, rather than anything happening at the level of vector spaces."
This is wrong, I'm afraid, there's a very large human-feedback reinforcement component in all these models where annotators have labelled massive response corpuses as racist, helpful, cheerful, pornographic, difficult-to-understand etc. etc. and that's incorporated directly into the training process via e.g. https://cameronrwolfe.substack.com/p/ppo-llm. Much of the "chatbot style" is actually baked into the network weights.
I see what you mean, but that is still coming about through human intervention and deciding what to reward.
A couple of weeks ago I was half-listening to a podcast when I noticed how often the interviewee would start a response with some form of "That's an excellent question...," or other compliment, presumably as a brief stall for time while they formulated their answer. I assume a lot of automated podcast interview transcripts have made it into the training sets by this point; I'm now fond of the possibility that LLMs developed this tic because humans in conversation need a second to gather their thoughts.
Poppy is an amazing young woman! Has she ever tried poetry? If she is good at patterns and mimic, I suggest she take some classics and use the form and rythym, tone, to write her own. I used to do it back in the day for kicks.
https://francinemckenna.com/poetry/
She loves poetry and has memorised an astonishing amount. I don't know if she writes poetry but suspect I wouldn't be told
I teach a first-year college writing class, and I'm thinking "write like a chatbot" could be a fun activity.