orbital-decay

7 ms·

61.
by orbital-decay 1mo ago
Yes, although admittedly I haven't tried recent ones which have a bunch of synthetic data in them (mainly to aid the reasoning), and are usually only available after mid-training. One look at the logits/output distribution and it&
62.
by orbital-decay 1mo ago
The corpus is already massively multilingual, it's really large, and has nothing to do with it. Models are very specifically engineered to respond in a default register when asked in English. There were models trained differently, fo
63.
by orbital-decay 1mo ago
It's not fundamental at all, that's what randomized sampling is for. Try tinkering with a base model and you'll be surprised how diverse it is. The semantic collapse happens in post-training that is using the current methods.
64.
by orbital-decay 1mo ago
That's a problem of mode collapse in post-trained models, which AI labs don't seem to be interested in, or at least their attempts to fix it are not visible, and there's relatively little research around it. Base models are d
65.
by orbital-decay 1mo ago
Very different software. Paint.NET is mostly a composition/annotation tool oriented towards simplicity and speed, Krita revolves around digital painting.
66.
by orbital-decay 1mo ago
My bad, it wasn't the intention. Then you should probably be aware that none of this is new.
67.
by orbital-decay 1mo ago
If you paid any attention they've been acting this way for years exactly to get open research banned, based on what bizarrely looks like a religion (developed over the recent 2 decades, with most of religious attributes). Monopolies, m
68.
by orbital-decay 1mo ago
>If anything, NAT has saved millions of wildly insecure devices running unpatched old Windows versions from getting pwned the second they connect to the open internet. I don't think NAT being a janky ad hoc firewall does anyone a
69.
by orbital-decay 2mo ago
>Deepseek invented the paradigm of prompt caching Caching was always here, you don't need to do anything special to get it on a single user local backend running a base model or a chatbot in the first place. Among commercial provide
70.
by orbital-decay 2mo ago
>Hold Alt while trying to select to make a rectangular/block selection. I don't see how block selection helps on cut text. In GUI you simply select the partial element and copy it, with either mouse or keyboard. You can also
71.
by orbital-decay 2mo ago
>I also can't cleanly nest and stack GUI applications in tmux like I can with TUI programs. Sounds like a window manager problem
72.
by orbital-decay 2mo ago
Every small feature is used by some subset of users, but if you don't trim your software from time to time then feature creep will eventually make it incoherent. Yes, it involves even small details like this one. I don't think lea
73.
by orbital-decay 2mo ago
> For example, why did Edge and possible chrome, remove support for Back via the backspace key? Literally, why remove something that exists and that some people might use? Because web pages don't preserve element focus well, web w
74.
by orbital-decay 2mo ago
Professional and power user software is always keyboard-oriented when it makes sense, I don't see it being ignored. Microsoft even addressed the criticism that Windows was largely not possible to use from keyboard, and massively improv
75.
by orbital-decay 2mo ago
Since you mentioned Windows, Microsoft made Ribbon which uses Vimperator-style navigation to make toolbars and buttons discoverable as long as you know one hotkey (Alt). It can also be hidden until triggered with a hotkey. This is miles ahe
76.
by orbital-decay 2mo ago
>A TUI can't prevent the user from copying text by its very nature. Funny but I struggle with copying anything in most recent TUIs. Fancy padding or multiplexer borders and multiple lines of text? It breaks. Incomplete text in a s
77.
by orbital-decay 2mo ago
>That sort of design has fallen out of favor with newer frameworks, unfortunately. The new preferred style seems to be a wireframe that the dev chooses which parts fill in, and often only the barest of essentials makes the cut. Which o
78.
by orbital-decay 2mo ago
>But I want to oppose a recurring argument in favor of TUIs that in my opinion does not have a solid foundation. To paraphrase various commenters: TUIs should be preferred because they are keyboard-driven. Yes. One problem with this ar
79.
by orbital-decay 2mo ago
This looks pretty neat but also distracting and not very convenient. A useful follow mode should provide a view into the next several dozen seconds of road conditions. This projection forces road sections after sharp turns to go off screen,
80.
by orbital-decay 2mo ago
This problem is legitimately hard and needs high cognitive abilities. Even the biggest generalist models struggle with memes and lingo salad that sound immediately intuitive for an out-of-the-loop human, and I'm talking about text comp
81.
by orbital-decay 2mo ago
I'm starting to think that your condescending "go ask a chatbot" is not intended as an insult to imaginary annoying AI bros unable to comprehend High Matters That Have Been Definitely Solved Once And For All By Those Smart Pe
82.
by orbital-decay 2mo ago
Science has little to do with consciousness. If it can't even be defined without using immeasurable concepts like "intentionality" then it's subject to arbitrary interpretation.
83.
by orbital-decay 2mo ago
Yes, I've been doing this since Claude 2 (1.x really).
84.
by orbital-decay 2mo ago
Have you... tried giving it to Claude? Although a model would likely need a different guide, aimed at its particular issues.
85.
by orbital-decay 2mo ago
Then I suggest you do the same for rigorous definitions that don't rely on consensus, and maybe also addressing my point about non-frozen weights directly.
86.
by orbital-decay 2mo ago
>See e.g. 4.8, Dynamic Flow: LLMs basically can't have this, as their weights are frozen. So consciousness = neuroplasticity? Requires neuroplasticity? Sounds pretty arbitrary, just like the entire link is an arbitrary set of ill-
87.
by orbital-decay 2mo ago
>Whether AI is conscious or not is independent of the social acceptance of the idea, unless you believe in thoroughly discredited and quite radical forms of social constructionism. Are you saying I need to have certain beliefs (discred
88.
by orbital-decay 2mo ago
Meanwhile that's exactly what we see now, do we not? Whether AI is conscious will be determined by the social acceptance of that idea, because how can you prove or disprove it without using a ton of questionable assumptions? How can yo
89.
by orbital-decay 2mo ago
It defines consciousness as an >observer-dependent model, subject to social consensus. Which is actually more pragmatic than most conflicting philosophical definitions involving qualia and other handwavy concepts.
90.
by orbital-decay 2mo ago
But... why? Computer use is probably the least efficient way to do this. There are agentic mods and plugins that are a lot better. See also SkyrimNet, people have been building agents for gaming and roleplay long before tool use was even a

More ›