this post was submitted on 23 Sep 2024
90 points (95.0% liked)
Wikipedia
1663 readers
365 users here now
A place to share interesting articles from Wikipedia.
Rules:
- Only links to Wikipedia permitted
- Please stick to the format "Article Title (other descriptive text/editorialization)"
- Tick the NSFW box for submissions with inappropriate thumbnails
- On Casual Tuesdays, we allow submissions from wikis other than Wikipedia.
Recommended:
- If possible, when submitting please delete the "m." from "en.m.wikipedia.org". This will ensure people clicking from desktop will get the full Wikipedia website.
- Use the search box to see if someone has previously submitted an article. Some apps will also notify you if you are resubmitting an article previously shared on Lemmy.
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
In particular, that they signed a 20-year-long deal. I'm guessing that gets them lower prices, and it is Microsoft, so I would expect them to weasel out of that contract somehow. But still just wild to me that they would even consider such a long-running contract for technology that's been around for barely a fraction of that.
Machine Learning tech has been around for decades. It just became sorta a little bit useful (outside of super specialized uses like photo post-processing or chess engines) around 2017 or so, and since then has exploded in capability, at a speed that the public is kind of incapable of comprehending.
It went from the approximate intelligence of a toddler to a smart high schooler in about 4.5 years, in the last 1.5 years it has nearly reached the level of a STEM PHD in several areas, if the latest whitepapers are to be trusted. And of course it's been much faster than us at reading, interpreting context and summarizing information accurately for a couple years as well.
DeepMind cracked the Protein Folding problem, and that's old news.
I honestly don't think people are ready for how fast AI has been improving.
Even if the AI bubble pops spectacularly and Microsoft aren't training any new models in two years, Microsoft still have massive datacentres that need to compute loads of things, and loads of things they'd like to compute, but don't because they don't have the capacity, but might end up doing if they're suddenly not training AI models on lots of machines. Having free electricity for twenty years is useful for a tech company with or without AI, and nuclear is a good way to get back on track for being carbon neutral soonish like they were going to be before all the recent AI stuff.