A very pro-AI account on both Bluesky and X posted about a "disturbing" Stanford paper on LLMs' failures at reasoning

[Happy Hoosier]

29. I work in software development. Such Anthropomorphic language is common.

First of all, "they" is a just a pronoun that can be used n the plural. When I say "they," I am referring to the AI models. I use the same to refer any group of things. Example: Q: Where are the cans of paint we bought? A: THEY are in the garage.

Secondly, we refer to software "trying" to do stuff all the time. It's a shorthand. Non-conscious actors cannot have any intentions, of course, but software CAN be written to have objectives and to attempt to achieve those objectives. This doesn't imply conscious intention. For example, some software my team has developed, "tries," that is "attempts" to resolve ambiguities in GPS measurements, to achieve a highly accurate fix. It calculates probabilities, and picks a particular fix and then "tries" to evaluate this decision to see if it was correct. Not intention here. It's math. All software is math. But so are our brains. There WILL be a time when the complexity of this process is so complex that something like consciousness and intention will emerge. I think that because I am a methodological naturalist (I don't think souls exist). But we're not there yet.