Note: the latest pixel fold is about the same size and weight as the pixel 9 pro.
It’s actually kinda incredible that it is that light and thin while folding.
Note: the latest pixel fold is about the same size and weight as the pixel 9 pro.
It’s actually kinda incredible that it is that light and thin while folding.
3.4bn is their gross - we have no idea what their operating costs are since they refuse to share them.
Some estimates say they’re burning 8 billion a year.
The latest releases ChatGPT 4o costs $600/hr per instance to run based on the discussion I could find about it.
If OpenAI is running 1k of those models to service the demand (they’re certainly running more since queries can take 30+ seconds) then that’s 200M/yr just keeping the lights on.
I am not! I hadn’t heard of it until just now. Very neat project.
No I’m just someone with a particular interest in corvids who feels represented by a god of Thought that is a bird
What a mediocre story
Oh I’m 1000% in agreement with you. I think Copilot for programming is more expensive than it’s worth right now, both for my employer and for Microsoft.
OpenAI et al have done nothing to address the fundamental issue of hallucinations. In code hallucinations are pretty quickly evident: your IDE immediately throws up error highlights whenever the code complete fucks up.
The latest open AI model is to chain together a computational centipede to try and create reasoning structures out of stochastic processes. It takes longer and still doesn’t fix the issues. In their own demo video there are clear bugs with the “code” their 4o model writes.
Ish.
You’ll have assertions that are entirely new or different, other pieces of setup or teardown. It really is one of the best use cases for GH’s Copilot that I’ve run across.
In my day to day the intellij autocomplete is what I prefer.
Yeah my usage of it is similarly limited. But the plagiarism engine is more useful than it is annoying in my experience. Especially in writing kdoc or unit test variations. Write one, write the name of the next, have autocomplete fill it out with the expected conditional variation
The plagiarism engine effect is exactly what you need for a good programming tool. Most problems you’re ever going to encounter are solved and GenAI becomes a very complex code autocomplete.
An LLM constructed only out of open source data could do an excellent job as a tool in this capacity. No theft required.
For writing prose it’s absolutely trash, and everyone using it for that purpose should feel ashamed.
TSMC is an exceptionally high paying job in Taiwan and they can’t go anywhere else.
You’re not abusing extremely high skill workers who can easily get a new job elsewhere. Not if you want high profit margins.
What they’ve done is flattened and encoded every aspect of the doom game into the model which lets you play a very limited amount just by traversing the latent space.
In a tiny and linear game like Doom that’s feasible… And a horrendous use of resources.
ML means you need a beefy GPU. That could always be a secondary addition though - add it in later as an external GPU and call it good.
I bought an expensive chair 8 years ago and it’s as good as the day I bought it. I’ll easily get another 8 out of it and it will likely last 30+ years of heavy use.
Which makes it cheaper than buying a $120 chair every 3 years.
If a device isn’t using a local detection of the wake word it would have a constant stream of data sent back to the developer… Which is super obvious.
It also wouldn’t be able to respond “Your device is offline” when the Internet is down.
It’s not a thing and it doesn’t happen.
It’s a dangerous command - I’d rather not run it by accidentally hitting the f
key a second time.
If it’s not using GCM then it must be long polling, unless signal servers are set up to use a 3rd form of push (APNS for iOS, GCM for Android)
I mean how bad can the fabs be - they’ll be running ASML EUV machines right?
Also brake dust is carcinogenic ❤️❤️❤️
The battery life is definitely worse because of it but it doesn’t bother me much