Exactly. Giving a helping hand to every bad actor, rogue state or hostile superpower is reckless and naive. Stupid thing to have done

Wordpockyclypse

Wordpocalypse?

Oh good. Someone did the subtext. Thank you.

https://www.itjungle.com/2023/01/23/it-is-time-to-have-a-group-chat-about-ai/

And that’s why the United States Government should immediately publish how to build hydrogen bombs.

You’ll be able to identify when the technology is being used badly from the handy mushroom clouds.

Keeping dangerous things in the hands of a few isn’t great, I’d rather have them in the hands of nobody, but it still beats having them in the hands of everyone with a GPU.

That’s quite an interesting video from Regent Law School, reminding us all that the “goals” (if any) of our interlocutors should be correctly apprehended for successful communicative exchanges. I expect that interactive LLM software has the overall goal of providing an environment of interactions with human users, that makes the machine’s textual responses to user queries, appear human as well. Part of this would be NLP parsing of the query, followed by response production with corresponding domain-specific vocabulary, syntax, and grammar. In the absence of a model of cognition (or domain-specific sub-models thereof), the LLMs can only guarantee that their outputs are syntactically similar to what a human would produce, within the query’s subject area (domain), but without much guarantee of accuracy beyond accidental correctness. This is where I think that a sub-goal of persuasiveness may intervene in current LLM efforts (either explicitly programmed-in, or resulting “inadvertently” from particular aspects of the training dataset), that directs responses towards a written style of apparent authoritativeness (sophist rethoric), that effectively hides the lack of a cognitive backend, making the system’s outputs appear more credible to users than they actually are. Humans do the same thing all the time (eh-eh-eh!).

Tom M

“Don’t talk to an AI”, because it can, well, will be used against you – reminded me instantly on this https://www.youtube.com/watch?v=d-7o9xYp7eE “Don’t talk to the police”