Local?

How much value do people find in running local ai? Unless privacy is a huge concern, I just don't understand why people are clamoring to get dedicated hardware.

Not only is the process incredibly expensive (the ones in this video easily exceed $40,000), but it requires constant tinkering to get a useable result.

I'm all for getting rid of subscriptions, but I'll be waiting a while before I buy hardware at this point.

Source: https://www.youtube.com/watch?v=4l4UWZGxvoc

Stats of Facts?

One of my most controversial takes is that statistics aren't real. When I was watching Star Wars, at one point c3po was giving the odds of surviving an asteroid field much to Han Solo's irritation.

But Han had a reason to be angry, the predictability of his attempt can only be compared if you assume that the former pilots had the same factors as Han. Statistics are so much more complicated than most people realize, to the point where the same statistics can be used to present both sides of an argument.

Elon has no possible way of saying xAi has a 10% likelihood of success. It is not possible to have data about something that does not exist. It rings as hollow as saying that self driving will be here next quarter. When experts all across the field agree that artificial general intelligence is a way off, and that our current methods will not get there, Elon still promises Mars for a bit of extra cash...

Source: https://www.businessinsider.com/xai-all-hands-agi-superintelligence-funding-success-optimus-space

Consequences

It's funny how things don't turn out how we intended. First the US blocks chip production from China so they start producing their own chips. Then China starts producing ai models that use less compute resources, and now people are smuggling in Nvidia chips.

Trying to limit your competitors isn't the way to win, and it's clearly not working.

Source: https://www.youtube.com/watch?v=lh0CrcPt3OM