ACCount36

8 ms·

31.
by ACCount36 1y ago
A sufficiently capable AI would be able to plug itself in properly too. One more reason to be wary of pushing for better capabilities.
32.
by ACCount36 1y ago
No, training a smaller model off a more capable larger model (or an ensemble of models) is the "usual" distillation. "Self-distillation" refers to distilling from a model into a copy of itself. Which is of limited use -
33.
by ACCount36 1y ago
"Accessibility CAPTCHA" is a well known partial CAPTCHA bypass. Solving an audio-only CAPTCHA with AI is typically way easier than solving some of the more advanced visual challenges. So CAPTCHA designers are discouraged from leav
34.
by ACCount36 1y ago
And then you use self-distillation to wire the improved prompts back into the LLM. Bam, free metacognitive skills.
35.
by ACCount36 1y ago
They're solving for everyone who needs captchas solved. It's a very old service, active since 00s. Somewhat affiliated with cybercrime - much like a lot of "residential proxies" and "sink registration SMS" serv
36.
by ACCount36 1y ago
Where's the externality here? Isn't "vitality of bees" that this method claims to improve actually supposed to be desirable to beekeepers themselves?
37.
by ACCount36 1y ago
As always: if those ideas are so good, why aren't they used? If existing practices are somehow radically worse, I would expect the first entity to adopt better practices to obtain a significant advantage - and the competition to copy t
38.
by ACCount36 1y ago
Given that "training AI on books you own" was ruled fair use, the "purpose" DRM is serving here is preventing fair use. Which is the kind of thing you would expect it to do.
39.
by ACCount36 1y ago
You think too highly of humans. Humans are not reliable. For every "no human would make this kind of mistake", you can find dozens to hundreds of thousands of instances of humans making this kind of mistake.
40.
by ACCount36 1y ago
LLMs are only capable of performing a finite amount of computation within a single forward pass. We know that much. They are also known to operate on high level abstracts and concepts - unlike systems operating strictly on formal logic, and
41.
by ACCount36 1y ago
Do you reason? "LLMs do not reason at all" casts that into doubt immediately.
42.
by ACCount36 1y ago
It is shallow. But as long as what you're asking it of is the kind of material covered in high school or college, it's fairly reliable. This generation of AI doesn't yet have the knowledge depth of a seasoned university profe
43.
by ACCount36 1y ago
Not every book is readily available as a digital copy. Things like textbooks, older technical books or just books that weren't too popular can be easier to source as physical books and scan destructively. A lot of digital copies are al
44.
by ACCount36 1y ago
Machine learning is 20% programming, 30% math and 50% demon summoning.
45.
by ACCount36 1y ago
DRM keys. This is exactly the kind of attack that's used to extract DRM keys, which are normally made completely inaccessible to the user by malicious device vendors.
46.
by ACCount36 1y ago
Exposing motion vectors is a prerequisite for a lot of AI framegen tech. If you could tap that?
47.
by ACCount36 1y ago
Which is why even "unlocked" bootloader doesn't let the user load his own code into TrustZone. The name "TrustZone" is rather ironic. It's most commonly used to run DRM code the user should never ever trust.
48.
by ACCount36 1y ago
Phones are portable, and thus more likely to suffer from a physical attack. But that's about it. It is, and always was a flimsy excuse to the strip user of control over his own device. "Secure Boot" isn't actually there
49.
by ACCount36 1y ago
One of the best use cases for LLMs. They excel at this kind of translation-adjacent task.
50.
by ACCount36 1y ago
Reportedly, Jared Isaacman, the rumored Musk's NASA pick of choice, wanted to cut down SLS/Orion and axe Gateway - and was opposed to science cuts. Jared Isaacman is out, and what we're seeing actually happen now is the oppos
51.
by ACCount36 1y ago
Not actually a problem? Have you seen what NASA is doing lately? SLS. Orion. Gateway. Ambitionless Artemis. JPL's disaster of an MSR proposal. NASA reeks of rot and decay. It's not in a good place, and hasn't been in a long t
52.
by ACCount36 1y ago
You're onto something. The issue is, if you do small targeted cuts, you'll spare the very people you want to cut. Because they're the best at playing office politics and finding ways to justify why they shouldn't be cut.
53.
by ACCount36 1y ago
SLS and Gateway should be the first to die, not the last. But politicians just can't stop funding their pork barrels.
54.
by ACCount36 1y ago
How many tons of payload would that Starship land? How many could Saturn V? If you want to go beyond planting a flag, you need to be thinking of how to land hundreds of tons of equipment and industrial infrastructure on the Moon. Saturn V i
55.
by ACCount36 1y ago
People who should be fired are the very people who would be the best at justifying why they shouldn't be fired. Not at all by a coincidence. "Slicing at random" could actually outperform most other methods, as long as it&#x27
56.
by ACCount36 1y ago
MSR as NASA's JPL envisioned it was a disaster. But there was a chance that someone like Rocket Lab could actually make it work without breaking the bank.
57.
by ACCount36 1y ago
That's just human capability elicitation. Having "skin in the game" doesn't somehow make a human surgeon more capable. It makes the human use more of the capabilities he already has. Or less of the capabilities he has -
58.
by ACCount36 1y ago
I can't see this ever replacing general purpose Arm cores, but it might be viable in LP-optimized always-on processors and real time control cores.
59.
by ACCount36 1y ago
Real world isn't "reproducible". If a robot can't handle weird and unexpected failures, it wouldn't survive out there.
60.
by ACCount36 1y ago
[flagged]

More ›