Why would somebody intuitively know that a newer, presumably improved, model would hallucinate more? Because there’s no fundamental reason a stronger model should have worse hallucination. In that regard, I think the news story is valuable - not everyone uses ChatGPT.
Or are you suggesting that active users should know? I guess that makes more sense.
I don’t pay Tesla for the vehicle (I don’t even supercharge), so it doesn’t actually affect their bottom line if I sell it. Buying a new car does affect my bottom line though.
Boycotts are only effective when they change purchasing behaviors, and this purchase is long done. I do encourage everyone who asks if I like my car to look into specific alternatives instead however.
And that’s really the only reason I have one - it was the only long range electric car available when I bought it. Now I would choose any of the many other options first.
That’s still faster than me though…
If you’re feeling out of breath, drink a thorium potion!
That’s increasingly untrue as we build tools to inspect and modify those weights. Nearly every open model has homegrown LoRA support, and most of them have abliteration code that lets you erase undesirable behaviors.
Should be illegal, but they are doing it legally by exploiting a loophole. Disgusting.
This one projects the image onto a bunch of vibrating rubber bands. I don’t think there’s a long term market for this. It’s a clever idea though that fixes the hazards of a solid diffuser (like those godawful fan-based volumetric displays).
Agreed. I use it in my daily workflow but you as the senior developer have to understand what can and cannot be delegated, and how to stop it from doing stupid things.
For instance when I work in computer vision or other math-heavy code, it’s basically useless.
Code written by AI is really poorly written. A couple smells I’ve noticed:
On the other hand, if you’re in a green field project and need to throw up some simple, dirty CSS/HTML for a quick marketing page, sure, let the AI bang it out. Some projects don’t need to be done well, they just need to be done fast.
And the autocomplete features can be a time saver in some cases regardless.
Yeah it was a quick google search. Do you have better numbers available?
Even an 80” tv only uses around 150W, if my research is correct. Surely this must be thinking about massive displays.
I disagree. While intellectual property legally exists, ethically there’s no reason to be protective of it.
Information should be a shared resource for everyone, and all these open weights models are a good example of that in action.
Presumably “small business” means self-employed or other employee-owned company. Not the bureaucratic nightmare that most companies are.
This is great stuff. If we can properly understand these “flows” of intelligence, we might be able to write optimized shortcuts for them, vastly improving performance.
Nah I do similar stuff. I think very few people actually trace their own lines of thought, so they probably don’t realize this is how it often works.
As long as open source AI keeps up (it has so far) it’ll enable technocommunism as much as it enables rampant capitalism.
I would agree with that if the cost of the tool was prohibitively expensive for the average person, but it’s really not.
Yes, but when the price is low enough (honestly free in a lot of cases) for a single person to use it, it also makes people less reliant on the services of big corporations.
For example, today’s AI can reliably make decent marketing websites, even when run by nontechnical people. Definitely in the “good enough” zone. So now small businesses don’t have to pay Webflow those crazy rates.
And if you run the AI locally, you can also be free of paying a subscription to a big AI company.
No but it is tone deaf (heh) to use Claude, a non-self hosted AI, and Suno, another non-self hosted AI to literally sing the praises of why you shouldn’t use corporate software.
Especially when open source and open weight models exist to do both of those things, albeit at a lower quality.