orbital-decay

3 ms·

31.
by orbital-decay 1mo ago
You don't have any backups at all if you never tried restoring and testing them
32.
by orbital-decay 1mo ago
Implanting a foreign CoT should drop the performance due to the reward-hacked CoT language mismatch, or in any case it will give replies different from the suspected teacher, which is precisely what happens here. However it's just a si
33.
by orbital-decay 1mo ago
Name training is shallow and should never be relied upon. Claude sometimes identifies itself as Qwen or DeepSeek when asked in Chinese. I've seen it identify itself as GPT-3 (that version in particular) and Reddit Anti-Evil Operations
34.
by orbital-decay 1mo ago
It's mostly coming from two out of many American companies that happen to have the highest concentration of people from a very specific movement born in California long before the current AI wave. In fact Anthropic's entire intern
35.
by orbital-decay 1mo ago
Mounting, wedging, steering uphill, forcing the head into a turn, etc. Tripping and especially firearms were used as a last resort, there's plenty of ways to do that without permanently injuring or killing the horse.
36.
by orbital-decay 1mo ago
That's one of the reasons why you should never trust a single word from Anthropic and OpenAI (Sam Altman also blamed them back in the day of R1, in a pretty convenient moment). If you know anything about Claude, DeepSeek, jailbreaking,
37.
by orbital-decay 1mo ago
Sure but he's pretty successful at creating a moral panic. Without a proper response people like him will succeed at a lot more than that. What are you going to do about it, besides commenting?
38.
by orbital-decay 1mo ago
Yes, he's the exact reason people are distrustful. He's a crank who learned about reward hacking and made a new religious movement out of it, pretending it's a world-ending issue and deliberately avoiding much more serious is
39.
by orbital-decay 1mo ago
English isn't the first language for me as well so I don't see any problem with that
40.
by orbital-decay 1mo ago
J-space is just one convenient projection (out of many) to look at these well known phenomena
41.
by orbital-decay 1mo ago
Not even a maybe... https://breezewiki.com/
42.
by orbital-decay 1mo ago
Is the color science correct? How did you test it? This will affect you regardless of the subset of features you use. A superficial resemblance to Lightroom isn't what makes a good RAW processing app.
43.
by orbital-decay 1mo ago
Nothing stops you from caching known states and workflows, or simply making the "fast" part of your interface fixed. I think this could be genuinely useful for one-off cases for which no interface exists, or simply for interface p
44.
by orbital-decay 1mo ago
It's well known that CoT needs a narrow problem or several independent narrow problems to perform well. With too broad of a task it often turns into a burden and hurts performance, making it hyperfocus on one thing where a non-reasonin
45.
by orbital-decay 1mo ago
Google doesn't care. Even massive efforts with humongous communities behind them are ignored for years. Try an incognito tab with a random element from a random game, for example this Fallout 4 quest: https://www.google.com&
46.
by orbital-decay 1mo ago
The next logical step is to express execution graphs in a formalized language.
47.
by orbital-decay 1mo ago
This sounds unconvincing, because a) pelican test is subjective, there's simply nothing to leak as it has no available direct answers and maybe an extremely faint preference signal, and b) the same small models actually do perform well
48.
by orbital-decay 1mo ago
This reads like typical LW FUD. 1. CoT is a pretty mediocre debugging proxy due to multiple phenomena, even if you train it for readability. You can have some direct intuition about how misleading it is by trying to hijack it to follow a fi
49.
by orbital-decay 1mo ago
Talking, sure. But have they created any good pelicans to train on? Pelicanmaxxing is clearly not a thing, which can be demonstrated by asking it for something else, or migrating to another domain (voxels, 3D modeling, and other things an L
50.
by orbital-decay 1mo ago
Worth keeping in mind that the article is written by a competitor.
51.
by orbital-decay 1mo ago
Window manipulation is trivial, most window managers are even natively programmable or at least support complex rulesets that don't need scripting for such basic tasks, I'm automating KWin in a similar manner for example. There&#x
52.
by orbital-decay 1mo ago
For all their readability benefits under direct sunlight, prolonged exposure also damages or kills them. You need at least a UV blocker for outdoor use, but the heat cannot be avoided.
53.
by orbital-decay 1mo ago
> wonder if you ever used Autohotkey to unfuck annoyances in software. Sure I did, but what particular annoyances in what software that you want to take with you into another ecosystem? I just don't get it, as you're being va
54.
by orbital-decay 1mo ago
Well, since you never even said what AHK feature you need it's impossible to suggest anything. Yes I used to use AHK years ago, but GUI automation is fairly alien and roundabout way of doing things on Mac and Linux. That's why I c
55.
by orbital-decay 1mo ago
I remember that time very well, and not in a good way. We were all running dubious janky obfuscated software without giving it a second thought. These days trust is not enough, you need at least some way to verify (but also a lot more). Ple
56.
by orbital-decay 1mo ago
Not sure why you think it's gaming related but fine, I just don't think AHK can do anything special. OP sounds rather like he's trying to find gotchas and justify staying on Windows, not genuinely find the replacement, which
57.
by orbital-decay 1mo ago
AHK is a pretty Windows ecosystem-specific crutch and everything it does is done in different ways in other systems. TC can be run on Wine absolutely everywhere, I used to do this on Linux for a number of years and it's perfect includi
58.
by orbital-decay 1mo ago
33ft/10m seems less than the multipath error in many cases (buildings, tall forests)
59.
by orbital-decay 1mo ago
Just FYI, Honda Goldwing (the archetypal Huge Luxurious Expensive 1800cc Motorcycle For Your Mid-Life Crisis) has about the same noise level as any ordinary car. I had one for years. Please don't lump everyone up under "they"
60.
by orbital-decay 1mo ago
That's pretty model-specific, for example DeepSeek of the v3/R1 era would already start losing coherence occasionally at t=0.7 with no other samplers

More ›