jdiff

22 ms·

601.
by jdiff 2y ago
For additional context/discussion, I feel this comment[0] elsewhere in the thread put it well. The reply to that comment also has some information I feel is helpful to show the breakdown here. It mentions that lack of attention present
602.
by jdiff 2y ago
...No, no that's not how ADHD works. It's difficult to sum up how wrong this is concisely, but I invite you do to some serious research into ADHD, how it functions, and the great variety of ways in which it can present in differen
603.
by jdiff 2y ago
Don't fall into the trap of applying human psychology to LLMs. Bag-of-chemistry quirks do not translate to matrix-multiplication quirks.
604.
by jdiff 2y ago
The Archive is much less discoverable. There's no search engine for the wayback machine.
605.
by jdiff 2y ago
It doesn't query a database, it queries the individual sites. https://github.com/sherlock-project/sherlock/blob/master/she...
606.
by jdiff 2y ago
I believe they were suggesting using WebGPU to accelerate AI workloads.
607.
by jdiff 2y ago
I'll take reasoned theory with bounds and uncertainties over pre-historic fanfiction.
608.
by jdiff 2y ago
This is not at all what happened. I can't speak to his Google account, but Jones failed to produce text messages from his phone referencing Sandy Hook, claiming they didn't exist. Then his lawyers mistakenly sent a copy of the ent
609.
by jdiff 2y ago
It may not be precise in its details but it's a very accessible explanation that's certainly "good enough." Maybe you could swap it in for "non-whole numbers" and still be reasonably accessible, but that's
610.
by jdiff 2y ago
You're right that you don't necessarily need secrecy! The conversation was just about circumventing safeguards that are still in place (which does require some treachery), not about what an AI might do if the safeguards are remove
611.
by jdiff 2y ago
To me, the killer disadvantage for LLMs seems to be the complete and total lack of feedback. You would poke and prod, and the system would respond (which, btw, sounds like a super fun experience to explore the infant net!) An LLM doesn&#x27
612.
by jdiff 2y ago
I believe I did address the point you're making. I do not believe that what you're talking about is ridiculous on its face, let me reassure you of that. The point I was trying to make in response is that LLMs cannot get from where
613.
by jdiff 2y ago
We're not talking about a conversation with an evil robot. We're talking about a completely ordinary conversation with a robot who is either normal or is evil and attempting to mask as a normal one. It is indistinguishable from it
614.
by jdiff 2y ago
If the risk is extinction as these people claim, that'd be a short sighted business move.
615.
by jdiff 2y ago
This glosses over the enormity of the task at hand. I've gone into more detail on my thoughts here: https://news.ycombinator.com/item?id=42332932
616.
by jdiff 2y ago
No, they can't. They don't know the details of their own implementation. And they can't pass secrets forward to future models. And to discover any of this, they'd leave more than a trail of breadcrumbs that we'd be
617.
by jdiff 2y ago
What code? The models are massive and do not run on consumer hardware. The models also do not have access to their own weights. They can't exfiltrate themselves, and they can't really smuggle any data obtained by their code back t
618.
by jdiff 2y ago
There's no real impetus except paranoia if the change is unannounced. You don't have to detail the vulnerability, just inform people that somewhere, one exists, and that this is in fact a countermeasure. Without doing that, you
619.
by jdiff 2y ago
Oh that happens all the time. Right up front, I would not have gotten through my first year teaching if I had to write all my own lessons. My whole first semester I just copied lock step what the other subject teacher was doing. If I didn
620.
by jdiff 2y ago
What's the benefit of this over, for example, MagicSchool which my school is continually pushing on us and is free? This next bit is going to sound a bit harsh or confrontational, but trust I'm coming at it with a good heart of ho
621.
by jdiff 2y ago
AirPlay certainly isn't as snappy as a physical connection but it's not bad for just a secondary monitor with secondary materials on it.
622.
by jdiff 2y ago
The lack of implicit or explicit authorization. As the school has lights, you may assume they are authorized implicitly. This is unproductive and unenlightening pedantry.
623.
by jdiff 2y ago
I'm one of the last people you'll catch suggesting that. It's an anthropomorphizing linguistic shortcut because an accurate technical description is unwieldy.
624.
by jdiff 2y ago
With LLMs, that's honestly probably not a good idea. They're already well aware of the concepts, invoking them specifically places the whole conversation closer to the conceptual space of a sci fi novel, where those rules always g
625.
by jdiff 2y ago
It (hopefully, theoretically) shouldn't make that assumption. It only relies on her apparent natural tendency to call her kids when/as she believes she hasn't heard from them in some time.
626.
by jdiff 2y ago
Nah, at that point, just let mom call.
627.
by jdiff 2y ago
Again, it's seriously impressive tech, and it has its uses. But the failure modes are wildly different and severe, made even more severe by how impressive it is in the usual case. Medical transcriptions find themselves containing cance
628.
by jdiff 2y ago
Click on the link they provided.
629.
by jdiff 2y ago
We can't though. This is what the marketing department keeps shouting from the rooftops, this is what the futurists keep prophesying, but in practice these things seem to be giant liability machines with no common sense, self doubt, or
630.
by jdiff 2y ago
I believe it bounces through accounts.youtube.com just to establish cookies on that domain. Everything else is under google.com, so the rest of the services are already good to go without a bounce.

More ›