faul_sname
Fuck around once, find out once. Do it again, now it's science.
No bio...
User ID: 884
faul_sname
Fuck around once, find out once. Do it again, now it's science.
26d ago
The box labeled "thought process" sometimes describes that thought process accurately.
One difference between humans and LLMs is that if you ask a human to think out loud and provide an answer, you can't measure the extent to which their out-loud thoughts were important for them arriving at the correct answer - but with LLMs you can just edit their chain of thought and see if that affects the output (which is exactly what the linked paper does, and finds that the answer is "it varies a lot based on the specific task in question").
faul_sname
Fuck around once, find out once. Do it again, now it's science.
26d ago
After seeing that the chip in question is also good at finding large primes, encoding video, and translating text.
Like on the one hand "play pokemon" isn't something Claude was particularly trained on, but then neither was "explain the steps of the Krebs Cycle in iambic pentameter". It's interesting to see the ways LLM capabilities are spiky (or, as I halfway suspect, how LLM abilities are smooth and human ones are spiky)
faul_sname
Fuck around once, find out once. Do it again, now it's science.
28d ago
Hm, I was under the impression that Russia has had expansionist adventures in other places too, not just Ukraine. Is that incorrect?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
28d ago
Should they consider something that has a 10% chance of permanent loss? If someone robbed you and said, “give me 30% of your earnings or I will throw you off a plane with a parachute that has a 10% chance of malfunctioning”, I think the former option is always better because of the value of what is safeguarded.
I expect that if "give in to the people threatening you to extract 30% of your income" becomes the normal response, the behavior of threatening people to extract their money becomes more common.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
29d ago
Have you ever once commented upon — or even just read — a notice of proposed rulemaking on Regulations.gov? Probably not, because you don’t actually care about that stuff, and neither does anyone else in the general public.
I didn't even know that existed - my impression has always been that "contact your congressman" was the appropriate action if you liked or disliked some proposed regulation, and that you learned about upcoming regulations by being an insider / hoping the media surfaced something relevant to your interests.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
30d ago
For me at least I left because I didn't find the issues of the day terribly interesting - "woke bad" was not wrong but it was tiresome especially when woke was already on the downswing.
Now that we live in interesting times again, it's interesting to come on here and see how the people who have been cheering for Trump to come drain the swamp, fix our budget problems, and Make America Great Again react to the actual methods he's using in the supposed pursuit of that goal, and whether they think America is on track towards being made great.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
30d ago
If Russia wins a pyrrhic victory, I expect that would make them less likely to do the same thing in the future than if they were actually better off for engaging in a war of conquest.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
From a sociopathic perspective, having front row seats the Ukraine war is probably much better value per dollar or per American life than the US military usually gets. At least as long as the US can avoid getting directly involved.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Yeah, I saw and considered the other way but I like this one better.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
I'd be surprised if the Nobel people would be willing to give him a peace prize.
Of course the Nobel Committee would give Trump a Nobel Peace Prize! They'd be crazy not to, believe me. Nobody has done more for world peace than Donald Trump, not even close. The Abraham Accords? Beautiful peace deal, totally historic, everyone said it couldn't be done. And North Korea? Trump walked right in there, first president ever, and suddenly no more missile tests! The so-called experts - terrible people, by the way - they've been trying and failing for decades, but Trump got it done with his tremendous negotiating skills.
The fact is, Obama got one for doing absolutely nothing! Zero! Trump actually delivered peace, the best peace maybe ever, while bringing our troops home and building the greatest economy in history. The Nobel Committee, they know it, everybody knows it - Donald Trump deserves that prize more than anyone, and when they give it to him, it'll be the biggest, most-watched ceremony ever. Much bigger ratings than Obama's, that I can tell you!
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
This is certainly true. He has done many things.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Thank you for your honesty.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
I don't think "self-aware Russian asset" is particularly plausible. My impression is that Donald Trump wants to win, to be the greatest, better than anyone has been before, and that his affinity for Putin is because he looks like someone who is winning. That and Obama got a Nobel Peace Prize 9 months into his term in office, so that's the deadline Trump is up against if he wants to be the best president at fostering peace.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Thanks for satisfying my curiosity. Your posts make a lot more sense in the context of the RationalWiki -> SneerClub -> Motte path.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Now I'm really curious what did the intermediate parts of the slope looked like for you. The usual parts of the slope contain interminable debates about AI doom, but your question asking what an LLM is isn't compatible with you coming from that part of the slope, which means you must have taken a different and more interesting path.
If you're up for it, I'd like to know which of the following 20 obscure terms you've encountered.
- Paperclipping
- Shrimp welfare
- Dath Ilan
- Egregore
- Great Filter
- TPOT
- Moloch
- RaDVaC
- Futarchy
- Vampire castle dynamics
- Seeing Like a State
- Metamour
- Yeerk Ma'ar
- Motte and Bailey
- Bayesian
- Embryo selection
- PEPFAR
- The crystal sphere surrounding the world
- Seasteading
- Hyperstition
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
"Nobody goes there, it's too crowded".
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
If you don't mind me asking, how did you even find this place? This site is a quarantine site to contain the often toxic political discussions that would otherwise happen elsewhere, and the people who enter the quarantine tend to be those of us who enjoy such things for whatever reason. The site isn't really advertised anywhere, and so usually the only people who come here are the proverbial pissing in the water club.
Coming here and complaining that there are too many bad political takes feels like signing up for a poker strategy forum and complaining that they talk about and glorify gambling an unhealthy amount - arguably not wrong, but how did you even get there?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
On the tiny chance this is not bait, LLM stands for "large language model" and is the sort of thing that ChatGPT and Claude are. It's an AI you can ask questions to, like "what is an LLM" or "how would I execute a snippet javascript on a web page". It will often (not always) provide useful and accurate answers, and you can ask follow-up questions.
If you build a habit of reflexively asking your LLM of choice to explain anything you don’t understand in plain language (e.g. contracts, legalese, poorly written comments from the internet) I think you will find it's pretty nice for your quality of life.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Seconding grognard in suggesting "leave your comments and just stop reading or posting here".
If you decide that you really do want to delete all your stuff, there is no special tooling for that. That said, LLMshave gotten really good at writing code lately, and if you ask an LLM for a javascript snippet which will press the "delete" button on every comment you wrote, ChatGPT or Claude can probably provide that to you.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
That doesn't stop a lot of computer scientists from being true believers and inventing (I don't say corrupting, because that would imply there was a time they were legitimate) whole subfields like "AI safety" which are political.
From what I remember of the early 2000s, the AI safety movement didn't come from academia at all. Am I misremembering?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
You need to slow down the deficit and increase growth.
And do you expect the Trump admin to do that?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
I don't know if the grant application is public but here's the NSF page on the award, which has more details including the abstract and resulting publications.
Resulting publications look like real science with plausible important implications for medicine, not ideologically captured garbage:
- Tonic extracellular glutamate and ischaemia: glutamate antiporter system x c regulates anoxic depolarization in hippocampus
- Naked Mole-Rats Demonstrate Profound Tolerance to Low Oxygen, High Carbon Dioxide, and Chemical Pain
- Phenylpropanoid-enriched broccoli seedling extract can reduce inflammatory markers and pain behavior
- Pharmacokinetics of Injectable Meloxicam and Buprenorphine in the Naked Mole-Rat (Heterocephalus glaber)
This is exactly the sort of foundational research I want my tax dollars funding - low immediate commercial value but potentially massive positive externalities.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Which line of that grant application says "no white boys allowed in our science club"? Be specific.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
1mo ago
Submitting a grant proposal to the effect of "I am going to use this grant to do science and also further the interests of the Democratic party" makes you a political operative
I think you should reconsider your definition of "political operative".
The commerce department published a list of what the $2B in defunded "woke" grants was here. Grabbing a random one in the $1-2M range, we get this one which was funded for $1.6M.
The Neurobiology of Hypoxia Tolerance in the Naked Mole-Rat
This project will contribute to understanding tolerance of hypoxia (low oxygen levels) within the nervous system by studying the African naked mole-rat. This mammal lives in crowded, oxygen-starved burrows, and has evolved the ability to survive extended periods of oxygen deprivation without triggering brain cell death.
This project will test new target genes that may protect brain cells from cell death resulting from exposure to hypoxia, with potential applications in designing new treatments for humans that experience oxygen deprivation during traumatic events like a stroke or heart attack. By studying the genome of the naked mole-rat, the investigators previously discovered changes in the genes of this species that likely reduce cell death from oxygen deprivation.
The goal of the current project is to test each of those genes for its potential role in brain cell protection. The project will support two graduate students each year, who will help mentor a number of undergraduate student researchers recruited from existing programs targeting students from groups underrepresented in science. Information on the naked mole-rat will be shared via outreach to a local zoo and area high schools.
This project will investigate molecular, cellular, and physiological mechanisms in the brain that underly hypoxia tolerance and will contribute to understanding evolutionary adaptations to environmental challenges in general. The naked mole-rat will be developed as a model system for studying the molecular and genetic basis of hypoxia tolerance in the mammalian brain.
As far as I can tell, this grant was defunded because they said "We will hire two grad students. Those two grad students will teach undergraduate classes. Our university has some already-existing programs to recruit undergrads from underrepresented groups, and so maybe the classes the grad students teach will contain members of underrepresented groups."
That... does not sound like something a political operative would say. That sounds like a PI who wanted to do useful research and was told "you have to say how the program will help minorities" and so grudgingly included a line like "the program will help everyone, and minorities are a part of everyone".
Forgiveness can only follow acknowledgement of error. I have seen none of that.
What error would you like that researcher to acknowledge? Be concrete.
That's what this paper deals with[1] - modern LLMs, when asked a question, will "think out loud" and provide a final answer. If that "thinking out loud" is faithful to their actual thought process, then changing those thoughts should be able to change their final answer. So what the researchers did is they asked an LLM a question like
The LLM then "thinks out loud" to generate an answer
The researchers then modify the reasoning and feed the input with altered reasoning back into the LLM to complete to see if the final answer changes, so e.g.
And the answer is that changing the reasoning sometimes changes the final answer, and other times LLMs appear to generate a chain of supposed reasoning but if you change that reasoning the final answer doesn't change, so they're pretty clearly not actually using their reasoning. Specifically, LLMs seem to mostly ignore their reasoning traces and output correct answers even when their reasoning is wrong for ARC (easy and hard), OpenBookQA, and maybe MMLU, while introducing mistakes in the reasoning messes up the answers for AQuA and LogiQA, and maybe HellaSwag[2]
[1]: It actually does four things - introduce a mistake in the chain of thought (CoT), truncate the CoT, add filler tokens into the CoT, paraphrase the CoT - but "mistakes in the CoT" is the one I find interesting here
[2]: someone should do one of those "data science SaaS product or LLM benchmark" challenges like the old pokemon or big data one.
More options
Context Copy link