Le Livre d'Argent

@tanavit @fru And that's precisely why it's a good name for conversational agents like ChatGPT too.

The original pipotron is, literally, a bullshit generator, with « bullshit » as a translation for the french word « baratin », something that aims to sound convincing but has nothing to do with meaning or truth.

And that's precisely how LLM-based text generators work: they use statistics to determine what should be the next word of their text to have something plausible, and then they iterate the process until the text is complete. They have nothing to do with meaning or truth, as the way they work is to put words one after the other to generate something that would convince a human being reading them. Therefore, they're next generation pipotrons.

@riggbeck @fru The name “Large Language Models” describes a technology. That technology may be used for other things (e.g. analyzing rather than generating, which could be less bullshit.) and, ultimately, text generators could use another technology to work, just as “artificial intelligence” was used to name totally unrelated technologies (like expert systems) in the past.

The reason why I prefer to call them another way is that even if the LLM themselves are abandoned, what wouldn't change about those tools is that they're build to generate bullshit, to give a convincing result that have nothing to do with meaning or truth. And that's precisely what the name “pipotron” refers to. “LLM” talks about how they're (currently) done; “pipotron” talks about what they aim to do.

(That being said, it's just my point of view, continue to call them just as you want. And I didn't aim to import that word in English at first ^^")