open source AI models have their work cut out for them. google's i/o conference keynotes in years past were all about Android with machine learning mentioned here and there, search mentioned here and there, and other properties getting honorable mentions. this year's conference keynote it's the exact opposite. it's all AI. Android had a short AI focused presentation for good measure, but that was it. google's narrative shift is clear.
Discussion
Everything in the computer engineering and electrical engineering graduate seminars involves AI right now as well. It's all the rage.
Why are you predicting difficulty with open source models? Is it the massive investment in the closed models?
The massive head start and amount of capital that Google has going for them. I want open source models to succeed, but it's going to be a large uphill battle.
We may see the capability gap between closed and open models expand for a while, but I think the investment in the closed models will actually drag the open model capability up as well, even if it lags. Engineers quit and start their own companies, advance open source models in their free time, etc.
Another way to understand the vast differences in capabilities between BigTech AI and opensource AI, is that BigTech can fake the results and make theirs look much better than it really is.
Open source "AI" can't do this kind of fakery to the same degree, by the nature of being open source. People can verify its capabilities.
Yeah Google has faked the performance of a lot of their tech in past years. It is unprofessional and pathetic and they aren't called out for it enough.
My take on "AI" is that it is largely a BigTech rug pull as they bring in tons of money from stock market investors, but there's really not much practical use or profit from "AI".
"AI" is largely fake. Don't buy into the hype. It only enriches BigTech at your expense.
It seems to me to just be a nice jump in algorithmic pattern recognition with some human input.
It's got some nice uses (for example, a photographer who would like to outsource their editing could theoretically use AI to speed up their editing process without paying someone to do it) but largely I agree.
In that understanding, some "AI" is just a subset of software. The parts that work and are useful can be understood as software with more of the steps automated.
But BigTech companies were having a hard time making more sales with "just software". Something new called "AI" ... but really just a slight variation of the software we already had was used to create investor excitement and hype.
My prediction is that hype will cool as little practical value over just regular software is realized. The people who have bought into the "AI" hype are likely to lose a lot of money if they stay invested.
That's my speculation.
Good points. Hype term that'll die down after a while or get so integrated into our regular lexicon that it makes no difference.
Llama 3 would like to have a word.
I don't know much about Meta's project and how open it is?
It is fully open, you have to KYC-lite to download it, but derived models don't have that and are usually better anyways. These open source models are a real gift and compete well against the closed source Big Tech ones. Speaking from experience
At some point when I can find the time I'd love to play with them. Thanks.