At Google, I asked why they were fixated on building THE LARGEST model.
Why are you going for size?
What function are you trying to achieve?
Why is the thing you were upset about that you didn't have THE LARGEST model?
They responded by firing me 🤷🏼
Why are you going for size?
What function are you trying to achieve?
Why is the thing you were upset about that you didn't have THE LARGEST model?
They responded by firing me 🤷🏼
Comments
Losers are running the biggest companies in the world. Thoughtfulness is not a trait they need. We need more women and poc people in these spaces for this very reason
More importantly AI data centers should have mandatory self sufficiency for power & water. Ideally they're helping innovate nuclear/desal while advancing AI. Potential sustainable profit from those investments too.
I get that it'll use more device power, but I suspect many queries would be fine with a much weaker model. Just predict if you need to off-load to something more powerful.
I’ve been doing heaps of local AI experimentation recently and it’s extremely capable, especially around the 70b size which is reachable to many folks already.
Microsoft on desktop
Amazon on e-commerce
Same reason med-schools restrict the annual number of physicians trained
The American Medical Association deliberately restricts the number of physicians as well as specialists that train every single year
Same with cost (we can always locate a few trillion to bailout bankers and the Pentagon)
Flawed decisions create barriers, not cost
Clippy 2.0 with spinning head.
Not only are LLMs too big, being too big is what makes them plagarism machines.
https://achewood.com/2006/04/28/title.html
Any LLM trained on Internet data is trained on 90% trash and can’t be trusted.
No but seriously, LLM value is based on the strength of the signal in the noise, not the amount of noise you have.
Ugh my arm chair expertise is flaring up again.
(This is sarcasm, you're the best!)