@kconrad@mastodon.social
Professor of English @ University of Kansas. Exploring science, technology, education, literature & culture. Particularly interested in generative AI.
LLMs are a selection of a handful of words from billions of possibilities. They are selected one by one until they arrive at a destination that aligns with what we ask of it. But at no point does it understand the words it is writing, in the same way the Macrocilix Maia doesn’t have any conception of looking like bird shit and how that benefits. It simply happens, much like LLM generated sentences! No awareness needed.
Hallucination is Inevitable: An Innate Limitation of Large Language Models
https://arxiv.org/abs/2401.11817
"In this paper, we formalize the problem and show that it is impossible to eliminate hallucination in LLMs. [...] By employing results from learning theory, we show that LLMs cannot learn all of the computable functions and will therefore always hallucinate. "
Elk is a nimble Mastodon web client. You can login to your Mastodon account and use it to interact with the fediverse.
Expect some bugs and missing features here and there. Elk is Open Source and we're actively improving it as a community project. Join us and let's build it together!
If you'd like to report a bug, help us testing, give feedback, or contribute, reach out to us on GitHub and get involved.
To boost development, you can sponsor the Team through GitHub Sponsors. We hope you enjoy Elk!
