I've never had an issue with the anthropomorphization of digital processes. When someone says "if you enter this kind of command, the tool knows that it needs to look up this thing over here", it's an adequate metaphor of what's happening, and the language flows well. People get a decent visualization of boxes and arrows in their heads that captures what the processes are doing.
It's different when you're anthropomorphizing an LLM. Please don't, be very technical and precise about what it does. There's a kind of uncanny valley of behavior that makes the metaphor collapse.
Isaac Kuo likes this.
clacke: exhausted pixie dream boy 🇸🇪🇭🇰💙💛
in reply to clacke: exhausted pixie dream boy 🇸🇪🇭🇰💙💛 • • •Because of the hype, all of the ways media and irresponsible salespeople are talking about LLMs, and all of the allusions to AGI, I know that a decent chunk of the audience will not get that anthropomorphization is a metaphor.
I know that there are really people out there making decisions, not just people on the street, but CEOs and engineers, that seriously believe that LLMs can think, reason and feel.
Anthropomorphizing LLMs feeds into the misconception and makes people make bad decisions based on a misunderstanding of what LLMs can actually do. We all need to take responsibility for presenting accurately what the process is and what it can do, and perhaps even more importantly what it can't do.
Isaac Kuo likes this.
doboprobodyne
in reply to clacke: exhausted pixie dream boy 🇸🇪🇭🇰💙💛 • • •Re. Not anthropomorphizing LLMs
I'm a sucker for this. I'll apologise to an inanimate object if I walk into it.
I find useful practical tips for myself in following this to be:
1. Use the verb "I prompted" rather than I told or I asked.
2. State that the program "output" rather than it replied.
3. I don't discuss "confabulation" because it's an anthropomorphization (the reality is that the computer program is doing exactly what it is instructed to do by the user), but if I was compelled to anthropomorphize, I would use "confabulation" rather than hallucination.
#LLM #AI #GAN #programming #language #linguistics #metacognition #philosophy #computers #anthropomorphization
clacke: exhausted pixie dream boy 🇸🇪🇭🇰💙💛 likes this.