Skip to content

Piero Bosio Social Web Site Personale Logo Fediverso

Social Forum federato con il resto del mondo. Non contano le istanze, contano le persone
  1. Home
  2. Categories
  3. Uncategorized
  4. How LLMs Learn from the Internet: The Training Process#LLMs https://share.google/R7w0FBUiIByrclGmk

How LLMs Learn from the Internet: The Training Process#LLMs https://share.google/R7w0FBUiIByrclGmk

Uncategorized
1 1 6
Log in to reply

Feed RSS
How LLMs Learn from the Internet: The Training Process#LLMs https://share.google/R7w0FBUiIByrclGmk

Gli ultimi otto messaggi ricevuti dalla Federazione
@pierobosio@soc.bosio.info
Post suggeriti
  • cassolotl@eldritch.cafeundefined

    Mozilla have a vibe-gathering survey out about AI.

    Uncategorized mozilla firefox thunderbird llms fuckai noai antiai
    8
    1
    0 Votes
    8 Posts
    10 Views
    oblomov@sociale.networkundefined
    @drchaos @cassolotl yeah trying to fill the form twice works even from the same browser and machine, I doubt it's serious (still, filled it.)
  • gert@poliversity.itundefined

    Language models cannot reliably distinguish belief from knowledge and fact

    Uncategorized llms epistemology knowledge
    1
    0 Votes
    1 Posts
    3 Views
    gert@poliversity.itundefined
    Language models cannot reliably distinguish belief from knowledge and factAbstract-----------«As language models (LMs) increasingly infiltrate into high-stakes domains such as law, medicine, journalism and science, their ability to distinguish belief from knowledge, and fact from fiction, becomes imperative. Failure to make such distinctions can mislead diagnoses, distort judicial judgments and amplify misinformation. Here we evaluate 24 cutting-edge LMs using a new KaBLE benchmark of 13,000 questions across 13 epistemic tasks. Our findings reveal crucial limitations. In particular, all models tested systematically fail to acknowledge first-person false beliefs, with GPT-4o dropping from 98.2% to 64.4% accuracy and DeepSeek R1 plummeting from over 90% to 14.4%. Further, models process third-person false beliefs with substantially higher accuracy (95% for newer models; 79% for older ones) than first-person false beliefs (62.6% for newer; 52.5% for older), revealing a troubling attribution bias. We also find that, while recent models show competence in recursive knowledge tasks, they still rely on inconsistent reasoning strategies, suggesting superficial pattern matching rather than robust epistemic understanding. Most models lack a robust understanding of the factive nature of knowledge, that knowledge inherently requires truth. These limitations necessitate urgent improvements before deploying LMs in high-stakes domains where epistemic distinctions are crucial.»#ai #LLMs #epistemology #knowledgehttps://www.nature.com/articles/s42256-025-01113-8
  • jubalbarca@scholar.socialundefined

    Fediverse folks, especially from the UK!

    Uncategorized llms
    1
    0 Votes
    1 Posts
    6 Views
    jubalbarca@scholar.socialundefined
    Fediverse folks, especially from the UK! The Lib Dem spokesperson for science and technology has a short feedback form on AI to get public thoughts on the subject. If you have five minutes to fill it out, please do so: I think it'd be good for politicians in her position to be hearing more from small scale creators and academics and suchlike on the problems we're seeing with these technologies.https://docs.google.com/forms/d/e/1FAIpQLScZiot9vGHvhOOt-1KX068gZSUwkvdE5vFQSRWTHBEoVIei3Q/viewformBoosts welcome!#AI #LLMs
  • remittancegirl@mstdn.socialundefined

    In reference to #LLMs, I want to really underscore the difference in speaking and writing.

    Uncategorized llms
    2
    0 Votes
    2 Posts
    10 Views
    remittancegirl@mstdn.socialundefined
    Also, there is an enjoyment, a compulsion, a jouissance, in the act of speaking that does nothing to help prompt any consideration of the chain of signifiers we use. This is great for psychoanalysis. Great for in-group social bonding, but not so good for communication that can bridge gaps.Writing, which is usually slower, which comes with many more rules than speaking - spelling, grammar, etc. essentially slows down the process and gives opportunity for consideration of impact and resonance 2/