Lemmy is at a particularly extreme level of violence endorsement. If the people on here said the things that they usually say in public, in actual society and not their little bubble, they’d be rightfully locked up.

If I were her attorney I’d push to have her sentence commuted to life without parol …

That’s exactly what her attorney is asking for.

The thing is, AI companies are terrified of model collapse because if they train AI systems on AI output, the model becomes useless.

That’s actually not a fact, more of a possibility. As far as am I aware, reinforcement learning from AI feedback (RLAIF) is currently employed by some of the most capable frontier AI companies; Anthropic openly admits to it, and their models are arguably the best.

A term that is often thrown around nowadays is recursive self-improvement (RSI), and it seems that there are at this time more knowledgeable proponents of RSI for intelligence development than opponents who raise the risk of model collapse.

But yes, training on today’s sloppified public internet would probably only have downsides.

This overstates the findings by a lot, and borders on misinformation.

How AI use relates to student performance is complex, varying with how students use it and what for. PISA data show that students who use AI for specific schoolwork tasks, such as summarising texts, drafting or conducting research, attain lower scores in science than students who do not use AI. In these scenarios, students who do not use AI outperform those who do, on average, by around 20 points – equivalent to being around one year of schooling ahead of their peers. However, weekly users of AI for general purposes have similar performance outcomes to non-users. Students who regularly use AI for general purposes, and have learning opportunities at school to develop AI literacy skills, tend to have slightly higher scores in science. However, these opportunities are more common among socio-economically advantaged students, risking a growing AI divide.

URL to OECD Summary

Edit: I concede that “heavily use” exonerates you, but your comment still feels a little disingenuous to me.

That is simply counter-factual. LLMs can absolutely do things that have not been published before, and they can access knowledge that is not in their training data through RAG. Intelligence is debatable, but reasoning models can reliably imitate “thinking” to a degree, and recursive self-improvement might lead to an autonomous advancement of intelligence. That LLMs allegedly “only regurgitate information” is simply wrong.