xyzzy123

47 ms·

751.
by xyzzy123 4y ago
I think of ASML sometimes and its relationship to hundreds of years of Dutch work in optics.
752.
by xyzzy123 4y ago
Like most engineering choices it has upsides and downsides. I'm personally "pro choice" here and in favour of allowing system operators to decide how much and what kind of automation makes sense for them. TLS vulns in general
753.
by xyzzy123 4y ago
Almost like it has a kind of dyslexia when it comes to "looking inside" tokens. If you instead ask it to write a Python program to do the same job, it will do it perfectly.
754.
by xyzzy123 4y ago
Don't students prepare for tests by studying past instances of them? "Teaching the test" (aka overfitting of human students at the expense of "real" learning) is a common complaint about our current education system
755.
by xyzzy123 4y ago
I think it's not just run cost - there's also a risk of ad revenue destruction, that has to be worrying.
756.
by xyzzy123 4y ago
If you live out in the woods, grow much of your own food and don't do many things that require money, I would argue you are quite well equipped to handle the most likely societal catastrophes. What would be even better is knowing a few
757.
by xyzzy123 4y ago
Is your "regulating" ai smarter, dumber or the same as the thing it's regulating? Dumber isn't going to work, the agent will be able to outmanouver it. If the "control" AI is smarter, what is keeping that one a
758.
by xyzzy123 4y ago
Risk, I think. For most people this is a higher risk strategy long term than just "paying the mortgage". Depending on your life stage and overall position it can make sense but it's not for everyone.
759.
by xyzzy123 4y ago
Setup matters a lot, eg if you have a separate room to go to or a standalone unit in the back yard. Kids are "sticky" (once they get into your office for any reason it can be difficult to get them out again). Number of kids matter
760.
by xyzzy123 4y ago
All indications are that you can scale LLMs up & down pretty smoothly, I bet there will be lots of workflows in future using AI on pc-or mobile scale hardware which can call out to cloud AI for "deep thinking" or to double che
761.
by xyzzy123 4y ago
I don't think this is anything new in "cyber land" grab any vps, take a pcap & watch the logs, the locks will start rattling right away. Twitter has always been a toxic cesspit of misinformation & influence campaign
762.
by xyzzy123 4y ago
Aren't all those things already blocked/banned in China? (since 2009). (I feel like a joke might be wooshing over my head here).
763.
by xyzzy123 4y ago
Yep, if you look carefully a lot of the demos don't actually work because the LLM hallucinates tool answers and the framework is not hardened against this. In general there is not a thoughtful distinction between "control plane&qu
764.
by xyzzy123 4y ago
LLMs can be implemented to be formally deterministic but if you ask them to solve a specific problem instance you have not seen before, you cannot generally guarantee they will do so reliably. So you're correct in a pedantic sense but
765.
by xyzzy123 4y ago
If you give it a reasonably well specified and well understood problem, solvable in a page or two of code and let it run in a loop (giving feedback) it will write and debug the code by itself. The output might not be "great" but i
766.
by xyzzy123 4y ago
I've played with this a lot, and can report (just like everyone else) it works eerily well. The main obvious challenge is getting things done within the context window, you have to fit your "tool instructions" in there, plus
767.
by xyzzy123 4y ago
cat prompt.txt && find . -type f -name *.py | xargs -I{} sh -c 'printf -- "---\n\n" && cat {}' && echo "--END--"
768.
by xyzzy123 4y ago
Sorry, yes that wasn't clear. Scott Alexander coined most of them in "I can tolerate anything except the outgroup" [1] Red & blue are exactly what a U.S reader would think they are. "Grey" is an attempt by Scott
769.
by xyzzy123 4y ago
I feel the difference is that to a first approximation, ai alignment folks tend to be grey tribe and ai ethics people tend to be blue or rainbow tribe. The schism isn't a technical disagreement about ai, it's a disagreement betwee
770.
by xyzzy123 4y ago
It's an "antipattern" (and usually a strong sign that something is badly wrong) but it's not illegal... "sinful" is a useful hint that something is "norm" related.
771.
by xyzzy123 4y ago
Unfortunately this is almost the least profitable market segment imaginable and isn't going to be addressed well by capitalism. I don't mean this in a "capitalist bad" way, it's mostly great but there are certain in
772.
by xyzzy123 4y ago
I had fun doing this until I had kids. I have a rack with 10gbe, ups, kubernetes a zfs storage server, multiple vlans, 4 unifi APs & locally hosted controller and all sorts of self-hosted stuff. My heart breaks slightly as I watch thing
773.
by xyzzy123 4y ago
Feels like most people have a hard time imagining that a god-like superintelligence could be completely indifferent to human affairs.
774.
by xyzzy123 4y ago
I don't think much of the "takedown" so far. Interested to discuss specifics if I'm not getting something. Cadhla (who works for EcoHealth alliance BTW) points out that it was Wuhan University who registered the rice seq
775.
by xyzzy123 4y ago
I think we might be talking slightly past each other. It's absolutely true that for many values of X, being "unaligned" or "uncertain" on X does some harm. What I'm saying is that human social dynamics mean tha
776.
by xyzzy123 4y ago
The Dogma usually includes statements like "people who don't care about X are part of the problem" and considerable social effort is spent on recruiting more "soldiers" and influencing group norms. The unaligned are
777.
by xyzzy123 4y ago
Did you notice the bit where it's a cDNA clone of an RNA virus and do you understand the implications of that? This is an engineering product. As for politics, sure there is a lot of that going around but you can verify this finding yo
778.
by xyzzy123 4y ago
The argument "they would have used a published backbone and reverse genetics system" is wishful thinking. You might interested in this paper: https://www.biorxiv.org/content/10.1101/2023.02.12.528210v2 &
779.
by xyzzy123 4y ago
> However, this is exactly the sort of virus that they study at the BSL4 lab in Wuhan, quite legitimately Bat coronavirus work including chimera creation was done at WIV at BSL-2, not BSL-4.
780.
by xyzzy123 4y ago
Doesn't the supply of stuff also matter? How about people's expectations of future prices? How about global-market linked prices of economic inputs, foreign money supply creation? Etc. I think I could come up with 50 reasons why

More ›