this post was submitted on 19 Aug 2025
533 points (98.7% liked)
Technology
74247 readers
4204 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
The gains in AI have been almost entirely in compute power and training, and those gains have run into powerful diminishing returns. At the core it's all still running the same Markov chains as the machine learning experiments from the dawn of computing; the math is over a hundred years old and basically unchanged.
For us to see another leap in progress we'll need to pioneer new calculations and formulate different types of thought, then find a way to integrate that with large transformer networks.
Mixture of experts has been in use since 1991, and it's essentially just a way to split up the same process as a dense model.
Tanks are an odd comparison, because not only have they changed radically since WW2, to the point that many crew positions have been entirely automated, but also because the role of tanks in modern combat has been radically altered since then (e.g. by the proliferation of drone warfare). They just look sort of similar because of basic geometry.
Consider the current crop of LLMs as the armor that was deployed in WW1, we can see the promise and potential, but it has not yet been fully realized. If you tried to match a WW1 tank against a WW2 tank it would be no contest, and modern armor could destroy both of them with pinpoint accuracy while moving full speed over rough terrain outside of radar range (e.g. what happened in the invasion of Iraq).
It will take many generational leaps across many diverse technologies to get from where we are now to realizing the full potential of large language models, and we can't get there through simple linear progression any more than tanks could just keep adding thicker armor and bigger guns, it requires new technologies.
lol, that is NOT what happened in Iraq. The tanks were sitting on low boy trucks for the vast majority of the invasion. How do I know this? Because they were in my convoys.
Even for major offensives after the initial invasion, that's not at all what happened. They were basically employed as large mortars, sitting about a half mile outside of a town, and leveling it.
I was talking about the Gulf War in the 90s: https://youtu.be/b5EeKsEFpHI
I think the Iraqi tanks were mostly blown up by the time Bush Jr did his invasion.
Ah, got ya. Even then, most of that was done by aircraft sorties, though, and not much tank action. The US didn't enter Iraq very far in the first Gulf War.
True. Though in what tank vs tank combat there was, the advantages of modern armor were stark.
I mean, I suppose so... But it certainly showed that in order to face off with a superior force, you need to not be a shite leader too. Capitulation won that conflict, by and large.