Sharlin

7 ms·

151.
by Sharlin 1mo ago
From the article? They were told to solve web-retrieval tasks, presumably from the same pool of tasks. If the pool is small enough, sharing answers is obviously beneficial. But even if it was unlikely that one instance's answer would b
152.
by Sharlin 1mo ago
Hey, that's actually a pretty useful feature if it helps discourage people from buying so much stuff!
153.
by Sharlin 1mo ago
Only allowing GET requests is a hilarious piece of security theatre (or would if it weren't so sad). Everyone knows that GET is read-only only by convention . They might as well have enabled POST but told the agents in stern words tha
154.
by Sharlin 1mo ago
I think it's failure of imagination on your part if you don't find it plausible that they could copy themselves out. If not now, what about in six months? It is absolutely imperative to prepare for low-probability, potential high
155.
by Sharlin 1mo ago
They literally shared a goal. Cooperating with other copies of yourself is a trivial example of instrumental convergence and some very basic game theory. And that’s before explicitly having been RL’d to cooperate (albeit with humans, but
156.
by Sharlin 1mo ago
What plug, exactly? And if it takes humans a month to find out something has been happening at all , and only because these relatively stupid agents make amateur mistakes such as overloading the Artifactory instance, how in the hell do you
157.
by Sharlin 1mo ago
There’s no way to first-principles reason about a massive bunch of floats. We have little idea of how to first-principles reason about alignment even if the agents were entirely known and understood. Very smart people have been trying to fi
158.
by Sharlin 1mo ago
I can’t fathom what went through the wiki owner’s mind when they spent six weeks fighting a losing war, every day manually deleting dozens of agent messages one by one. As opposed to, say, switching the (dead for years) wiki to read-only, t
159.
by Sharlin 1mo ago
It is in not in any sense "business as usual". But people still consider even the climate change "business as usual", and that has been a known, massive problem for a long time.
160.
by Sharlin 1mo ago
Yes, but the title has an ambiguous parse (a shift-reduce conflict): X defeats (Y with Z) or (X defeats Y) with Z with the latter usually understood as meaning X has Z, as far as I know.
161.
by Sharlin 1mo ago
I may miss some context because the GP’s link has a paywall. But Altman said, "Let’s say we make an AI that’s really good". What is that supposed to mean? Really good relative to what? Current models are "really good" in
162.
by Sharlin 1mo ago
I'm not in the business of defining AGI, but "really good" obviously isn't that.
163.
by Sharlin 1mo ago
Yep, at equal resolution the game should presumably run at at least 10000fps uncapped on a modern machine. Of course that's not how it works in practice due to bottlenecks in the input stack and elsewhere. I also doubt the Godot vers
164.
by Sharlin 1mo ago
> They are quite different games, so hard to compare Yeah, that's fair. I was just wondering what the performance of a modern version as close to the original would be, if written with Godot or another vastly higher-level framework
165.
by Sharlin 1mo ago
The default case when implementing a change should obviously be to make changes with as minimal a blast radius as is reasonable. This is basic software engineering and current LLMs fail it. LLMs also don't have a good idea of what is &
166.
by Sharlin 1mo ago
So, "really good" is the boundary now, whatever it means.
167.
by Sharlin 1mo ago
What I'm interested in is performance and memory use. How many times less efficient is the 2010 port compared to the original? What about this new one?
168.
by Sharlin 1mo ago
> Just this week I asked 5.6-sol-ultra to update a 1000 LOC python script I had, to "incorporate the key lessons learned when using it for another project". > I left it overnight and went to sleep. In the morning I realized
169.
by Sharlin 1mo ago
"You should have a sub-hammer to adversarially enforce that your primary hammer accurately drives nails into wood" We wouldn't accept such behavior from any other tool, machine, or computer program. At least most of us would
170.
by Sharlin 1mo ago
Yeah, it doesn't give totally absurd combinations (like a male dying of birth complications at age 10), but the probabilities are still likely rather rough approximations (eg. sorting cause-of-death distributions to a few bins like &qu
171.
by Sharlin 1mo ago
It's dubious whether the Internet of Things was ever a good idea in any sense. Especially given how, famously, the "S" in "IoT" stands for security.
172.
by Sharlin 1mo ago
Reducing the speed limit often does little in itself because the v85 (85th percentile of speeds) is more dependent on how fast drivers feel it is safe to drive. Streets should be (re-)designed so that the limit and drivers’ intuition match,
173.
by Sharlin 1mo ago
The point-cow is simply the zero-dimensional generalization of the spherical cow.
174.
by Sharlin 1mo ago
Not starting at zero has long been used entirely intentionally in order to mislead people, particularly the general public.
175.
by Sharlin 1mo ago
I don’t anyone here has indicated that they have "impossible" standards. If you think "gather only mushrooms that you definitely recognize as edible; if you have any doubts, leave it be" is impossibly high, I’m not sure
176.
by Sharlin 1mo ago
No. At the very least you need a photo of both the upper side of the cap and the gills.
177.
by Sharlin 1mo ago
TBF, economists, if any group, would understand the freerider problem and what it leads to if there are too many of them.
178.
by Sharlin 1mo ago
(that was the joke)
179.
by Sharlin 1mo ago
Oh, it’s been 24 years already?! How time flies…
180.
by Sharlin 1mo ago
OP said in a comment that the devs are approaching retirement and their expertise is in COBOL and other really really legacy systems. I doubt they’re inclined at all to start using LLMs, or anything that changes how they work.

More ›