reddit refugee
Since the first time I heard about FSD I’ve been wondering why Tesla (or others) doesn’t set up a system where drivers opt-in (no opt-in by default) to sending anonymized driving data to help train the model. The vast majority of the time, it’s probably modeling OK driving. At least no accidents. But the shitty driving and accidents are also useful as data about what to avoid.
Maybe they’re already doing this? But then I wonder why their FSD is getting shittier rather than improving. One would think with more driving data, good and bad examples, would only help.
I personally don’t tinker much with the OS. I want it to stay out of the way and let me do things. In the case of Bazzite, everything I need for gaming is just there and works without me lifting a finger.
I like the safety and simplicity immutables bring.
If I’m doing something out of the ordinary, a temporary container usually suffices.
It’s really made the switch from Windows as a daily driver much easier.
Been following this company’s development for over a decade now. I really want them to succeed but I have major doubts.
Yeah I’ve had to go back and fix (re prompt) some things like this in the past.
The headline is that it helps me code things far faster than if I was doing it myself. And sometimes saves me 100% of the work.
I used it the other day to spit out a ~150 line python script. It worked flawlessly on the first try.
I don’t know python.
All these downvotes make me wonder if people know that Saudi Arabia is the chair of the UN gender equality forum.
LOL
Yeah, agreed. That’s not what I asked though.
This response is a bit of a misdirection since we all discuss shit that isn’t the most important all the time.
I can give better advice in a similarly easy to consume manner, applicable to most.
Invest in a low cost target date fund. Look at Vanguard target date funds for examples and pick a year close to your expected retirement date.
Pay your highest interest debt before lower.
Both of these pieces of advice make you more money than doing what Ramsay says and are equally easy to understand.
Given what I think I know about LLMs, I agree. I don’t think they’re the path to AGI.
The person I replied to said AGI was never going to emerge.
Are you assuming LLMs are the only way humans could ever try making an AGI? If so, why do you assume that?
It sounds hyperbolic but if you assume it will reach human-level intelligence and will have the ability to update its own code, you very quickly have something much smarter than us. Whether it will want to help or hurt us is an unknown. Whether we can control something that’s smarter than us (and getting smarter every second) is unlikely, IMO.
“You’ve chosen the wrong phone” is peak lemmy.
Second only to “ditch Windows, use Linux. I don’t care about your use case”.
Glad I’m not the only one. I read it 3 times and still don’t know what’s being said.
Is it still full tilt?
Looking at CDC graphs of excess deaths, it appears we’re back at baseline. That is, assuming I’m reading this correctly which is very much not a sure thing.
Okay 2 things.
Edit: I see you have some info on the Rustdesk point elsewhere in the thread. I’ll read up on that part so don’t feel like you have to repeat yourself here.
No rustdesk but recommend RDP for remoting?
I’m confused on both recommendations.
Everything you suggest seems self evident because you supply the evidence yourself.
It should be obvious to anyone who knows the first thing about how cell phones work.
Remember, you replied to me. You seemed quite proud of your comment and all your links. Unfortunately you completely missed the point.
What if I flip both at once?