Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="Heilemann" data-source="post: 9698201" data-attributes="member: 7017813">SMLs are just LLMs but smaller, no? And (generally speaking) their size mirrors in their performance, no?I actually agree that LLMs are going to turn out to be just one aspect of intelligence, and will only get us so far (clearly our skull does not contain a massive datacenter, and yet we can do things they cannot). But I fail to see how SMLs solve that problem and are 'older and have more potential'?I think it's fair to say that language models are the only actual 'intelligent' machines we have today, and they can obviously do many, many things which were previously simply not possible. Whether they meet your exact, and undefined, criteria for 'solving problems' doesn't seem to stop them from doing a lot of real world work.Month after month the amount of work going into making them more efficient, easier to work with, more aligned and capable is just factually undeniable. Are they perfect? Far from it. And I don't think you'll find anyone who knows what they're talking about saying so, so playing that card is tired at best.And moreover, the capabilities they've already unlocked for people to talk to computer, understand them, work with them, is not just about Silicon Valley grow-grow-grow mentality. The unlocks in accessibility alone are absolutely astounding.Harping on about water (there are <a href="https://www.microsoft.com/en-us/microsoft-cloud/blog/2024/12/09/sustainable-by-design-next-generation-datacenters-consume-zero-water-for-cooling/" target="_blank">closed circulation water systems</a> already in use in Microsoft data centers) and power consumption (real worry, but we have and should use clean energy much more, AND the intense pressure on the industry is bringing down inference/token costs in an absolutely massive way) and on and on (while yes, overall energy consumption is up due to the growth of AI overall right now, but 'how do I boil an egg' costs vastly, vastly less to answer now than it did two years ago).There are also, thanks to the investments made into LLMs now, examples of <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5211933" target="_blank">training LLMs purely on copyright free material</a> successfully.And this technology is proving in fact to be very difficult to moat, and as such you're seeing countries and people previously unable to compete get a massive leg up. (Be less concerned with whether they're pouring billions into model training, and more with whether they're being taxed appropriately to spread the wealth. VOTE!)There is just an infinite amount of nuance to this essentially brand new technology development, and the "down with this sort of thing!"-ism is fine, if it's served with nuanced arguments rooted in (updated) facts and an open mind.There is much about AIs that I am skeptical or critical of, but to dismiss it entirely is certainly a mistake in my book.PS: I love using it for roleplaying, and I find it very capable when used right; will show some of my work soon <3</blockquote>

Verification