Philpax

35 ms·

751.
by Philpax 2y ago
It'll get increasingly difficult for open models to keep up with proprietary frontier models as the costs involved increase. Yes, there will always be older open models available, but at some point, they just won't be as competiti
752.
by Philpax 2y ago
Yes: https://www.mercurynews.com/2025/02/14/autopsy-openai-whistl... If it becomes a natsec concern, the government will simply ignore or redefine the laws. The current US government has already demonstrated
753.
by Philpax 2y ago
The latest autopsy came down pretty firmly on it being a suicide. Even if it could've been an assassination, there isn't much motive: his claims were about copyright violation, which neither OpenAI or the government would much car
754.
by Philpax 2y ago
There are many researchers, but a limited number with the ability to make the revolutionary leaps necessary. To take your analogy: there were many physicists who knew an atomic bomb was possible, and many who even attempted it, but the Manh
755.
by Philpax 2y ago
The end goal is something capable of human-level intelligence in most, if not all, fields, and from there, it's not that difficult to continue scaling it up across speed / scale (and perhaps intelligence.) See Dario Amodei's
756.
by Philpax 2y ago
I assume the implication is that they'll build AI capable of automated drone surveillance. Which is silly, because other Chinese companies have already done that, and America is pioneering it (cf Anduril)
757.
by Philpax 2y ago
I wouldn't rule it out iff the other candidates can't stump up the resources to play the game. I very much doubt AGI will come out of Canada or Australia; Europe is making some overtures, but it'll take some time to get the m
758.
by Philpax 2y ago
famously they were very much used
759.
by Philpax 2y ago
It's a nice sentiment and I agree, but I'm not sure we'll see this applied to AI going forward. These training runs cost a lot of money - at some point, every player in the game will realise that they need to charge somethin
760.
by Philpax 2y ago
Kind of a strange sentiment? It seems like you're willing to take a paycut, so there's no shortage of jobs that will pay you a sustainable, but not FAANG, income to work on something respectable. I get wanting to take a break afte
761.
by Philpax 2y ago
Man, I remember going to swapmeets as a kid in 00s Australia. Online shopping killed it then, but I'm happy to see it make a comeback in one way or another - I have fond memories of meeting other enthusiasts and admiring hardware old a
762.
by Philpax 2y ago
[2023], but that doesn't matter much. Fantastic article! Great to see what development was like at the start - I'd always heard anecdotes, like its humble beginnings and the MINIX argument, but I'd never had context for them
763.
by Philpax 2y ago
Baffling definition. How did you develop this impression?
764.
by Philpax 2y ago
Wow, you weren't kidding: same phone and it feels like I'm using a 15-year-old Android. That's rough .
765.
by Philpax 2y ago
Well, I don't know about that being relevant [0], but at least it isn't blatantly offensive to the concept of providing information like this proposal is. [0]: https://en.wikipedia.org/wiki/Gemini_(protocol)
766.
by Philpax 2y ago
I mean, the goal is to get it to a state where it can do those things. There are benchmarks for autonomously resolving issues that are being hill-climbed as we speak: https://www.swebench.com/ I don't know how far aw
767.
by Philpax 2y ago
That looks really nice, but it's not quite what I'm looking for - thanks anyway, though!
768.
by Philpax 2y ago
which is a lot of steps compared to just clicking the relevant window in a taskbar :(
769.
by Philpax 2y ago
Way, way too obtrusive, and not dense enough. I like it in concept, but it's just not practical, especially on a laptop screen.
770.
by Philpax 2y ago
This also looks like a great option! I have a vague memory of trying this before and finding it lacking, but I might need to re-evaluate.
771.
by Philpax 2y ago
This looks much closer to what I was looking for, thank you! Will try when I can :)
772.
by Philpax 2y ago
The issue isn't the scripting language, it's having a common interface for users, especially for those who can't use standard browsers. Rendering the application yourself will break every accessibility tool that relies on b
773.
Inside the Kodak Factory and Eastman Museum [video](youtube.com)
3 points by Philpax 2y ago|0 comments
774.
by Philpax 2y ago
Does anyone have any advice for making the most of the Dock? I find it pretty unhelpful coming from an older Windows / Linux background: I just want easy access to the windows that are open on my current workspace on my current monitor
775.
by Philpax 2y ago
Holy shit, that's incredible. You should publicise this more! That's a fantastic resource.
776.
by Philpax 2y ago
You're thinking of https://www.youtube.com/watch?v=DcYLT37ImBY and https://github.com/PWhiddy/PokemonRedExperiments . > It also almost always runs from battles instead of changing Pokemon or fig
777.
by Philpax 2y ago
The second half of this comment reaches the opposite conclusion to what I would have expected from the first half.
778.
GPT-4.5 spotted in ChatGPT web build(twitter.com)
3 points by Philpax 2y ago|0 comments
779.
by Philpax 2y ago
A LLM can be reused for other use cases. Your agent can't.
780.
by Philpax 2y ago
But how much effort do you have to put in to build an agent that can play a specific game? Can you retarget that agent easily? How well will your agent deal with circumstances that it wasn't designed for?

More ›