AI

> But artificial intelligence is one of the biggest threats to human
> civilization. It's the first technology in history that can make
> decisions by itself and can therefore take power away from us.

> The thing is that it's not a law of nature that in all cases, under
> all conditions, a distributed system, a decentralized system is always
> more efficient than a centralized system. This was true in the second
> half of the 20th century. Now, AI is changing the situation. -- Harari

----
>Today's attack on Community is using AI to answer questions, suggest code, and do everything in its power to get in between people so that the world stops relying on community and cooperation, and becomes fully dependent upon capitalism. [Source:](https://social.bau-ha.us/@raganwald/112428620133593570)

----
> Why AI can't do math: "They are fine-tuned for determining probabilities, not doing rules-based calculations. Likelihood is not accuracy, and language is more flexible, and forgiving, than math."

----
>... draw a distinction - weak in theory, stronger in practice I think - between “classical” AI/ML modelling (i.e. statistics) and Generative AI/LLMs.

<ol>
 <li>LLMs are <a href= "https://www.youtube.com/watch?v=UakqL6Pj9xo&amp;ref=blog.relyabilit.ie">
 not (yet?) AGI</a>. <a href= "https://ajithp.com/2024/09/30/ai-reasoning-and-lrms/?ref=blog.relyabilit.ie">
 Reasoning ability is still developing.</a>
 </li>
 <li>They’re also <a href= "https://medium.com/@alain94040/llms-are-not-just-autocomplete-a-simple-proof-a4880dd25a5b?ref=blog.relyabilit.ie">
 not just autocomplete</a>. <a href= "https://ajithp.com/2024/09/30/ai-reasoning-and-lrms/?ref=blog.relyabilit.ie">
 That paradigm fails to capture important components of their behaviour.</a>
 </li>
 <li>We <a href= "https://www.linkedin.com/pulse/when-llm-experts-say-we-dont-know-how-pallav-sharda-2tpyc/?ref=blog.relyabilit.ie">
 don’t understand</a> a significant amount about how they work.
 </li>
 <li>They are (incidentally or necessarily) <a href= "https://substack.com/home/post/p-151006948?ref=blog.relyabilit.ie">
 excellent search engines</a>, since exact syntactic/token matches are no longer necessary to retrieve.
 </li>
 <li>They are most reliable when they retrieve tokens which are already in the training data, or are clearly derivable from an
 existing distribution. <a href= "https://www.lesswrong.com/posts/Fr6eJkjYWG9Mw6XQc/how-good-are-llms-at-doing-ml-on-an-unknown-dataset?ref=blog.relyab
 Unknown datasets give less reliable performance</a>.
 </li>
 <li>Hallucination is <a href= "https://arxiv.org/abs/2401.11817?ref=blog.relyabilit.ie">not
 an accidental or incidental property</a> of how they work.
 </li>
</ol>

[Source for above](https://blog.relyabilit.ie/llms-wont-save-us/)  This is a really good article.

Create new association:
AI