mquander

10 ms·

61.
by mquander 3y ago
> But in one sentence, try asking ChatGPT to reverse a string of 20+ digits, like 473936482738338373926. https://chat.openai.com/share/79a7ab39-bd3c-42e8-b4bb-452ded... > The reversed string of '473936482738
62.
by mquander 3y ago
This seems like total nonsense to me. Compare: "If you do everything the way the average runner does it, you should expect average performance. The problem here is, average performance means that you won't win a race... So if you
63.
by mquander 3y ago
For people who agree with the libertarian stance, the optics aren't horrible, of course, they are good. The question is whether there are enough of those people to sustain an organization like the EFF.
64.
by mquander 3y ago
OK, acknowledged.
65.
by mquander 3y ago
Why does being a minor celebrity mean you shouldn't donate your kidney? You don't think most Americans are "psychologically suited" to donate a kidney? Again, I don't think you would apply this level of paternalism
66.
by mquander 3y ago
Your summary literally doesn't even sound more stressful than the median American adult life. He has a job? He got married? He has taken psychiatric drugs before? Some people don't like him? There exists social drama sort of nearb
67.
by mquander 3y ago
> So in total, a donation produces about 10-20 extra quality-adjusted life years. How much would you say the donor benefits?
68.
by mquander 3y ago
Game theory says that each party should receive their Shapley value. In the case of having one guy who owns the USB drive and N guys who are capable of decrypting it with equal proficiency, then the one guy's Shapley value is N/(N
69.
by mquander 3y ago
Unsupported assertion - simplest explanation is that the guy's credentials just got hacked and someone else is using his account.
70.
by mquander 3y ago
It's frustrating to me that if this journalist was writing about what e.g. Donald Trump said, they would feel morally compelled to do a bunch of fact checking, so that they can also print that the stuff he said is wrong and bad. But wh
71.
by mquander 3y ago
I assume by "commercial websites" you mean specifically "websites whose whole purpose is to have information in the website that you view in return for running ads?" Generally speaking, if I have a website where I have i
72.
by mquander 3y ago
AutoGPT doesn't work because GPT-4 is too stupid to make good plans, not because it's missing the secret sauce of "agentiness." If some future software was really great at observing, planning, and reasoning, then an Auto
73.
by mquander 3y ago
OK, but the hard part was the "intelligent" part, right? If you have the "intelligent" thing, and you want to get an "intelligent agent", all you have to do at most is wrap it in a shell script that runs the eq
74.
by mquander 3y ago
As long as the business model of journalism is running on fumes, there isn't going to be a renaissance in journalism quality. Truth-seeking is time-consuming and will always be inherently trading off against other local incentives, lik
75.
by mquander 3y ago
The examples you're looking at are coming from systems that have undergone long courses of reinforcement learning specifically teaching them to be banal.
76.
by mquander 3y ago
It doesn't need to be contagious to create a crisis, just correlated, e.g. if there is a big macroeconomic shock that makes many students unable to pay their loans all at once.
77.
by mquander 3y ago
What do you mean, you haven't seen anything else like it? At least O(100) people are working full-time on research programs trying to mitigate AI x-risks. The whole founding team of Anthropic left OpenAI to do their own thing specifica
78.
by mquander 3y ago
I just looked again and I didn't see that claim, can you verify? https://arxiv.org/pdf/2306.08997.pdf If as per the linked critique, some of the questions in the test set were basically nonsense, then clearly they
79.
by mquander 3y ago
Since you are here, when I was reading the paper I wondered -- when they show the "zero-shot solve rates", does that mean that they are basically running the same experiment code, but without the prompts that call `few_shot_respon
80.
by mquander 3y ago
> In a way it makes perfect sense that gpt4 can score 100% on a test gpt4 also grades. Even this is overstating it, because for each question, GPT-4 is considered to get it "correct" if, across the (18?) trials with various pro
81.
by mquander 3y ago
I think it's appropriate to be extremely critical. The paper is basically useless. The thing that they actually measured is "can GPT-4, when given a 'question' with lots of additional information and many tries with smal
82.
by mquander 3y ago
> Utilitarianism has trouble dealing with cheating because any single act of cheating seems to cause benefits for the cheater greater than the damage a single cheater does. It clearly doesn't? There's no sensible mechanism by w
83.
by mquander 4y ago
FWIW, I couldn't answer your question, because when you said Roro was rotating "alongside" Z I didn't know that meant the same thing as rotating around Z. I also agree with GPT that the question seems very underspecified
84.
by mquander 4y ago
The links in the README just go to other documents, but the repo seems to have most of the code for the components the documents are describing.
85.
by mquander 4y ago
The subfields of programming I know the most about are game development, networking, and web development, and in all of those, it's not the case that only 1% of the code is in the "edge case" where performance matters at all.
86.
by mquander 4y ago
> Look, most modern software is spending 99.9% of the time waiting for user input, and 0.1% of the time actually calculating something. If you're writing a AAA video game, or high performance calculation software then sure, go crazy
87.
Anomalous Tokens: a mysterious failure mode for GPT(lesswrong.com)
11 points by mquander 4y ago|2 comments
88.
by mquander 4y ago
I didn't make it to the end of the essay, but near the beginning I saw this which looked like it couldn't possibly be right: > In 1824, only eight years after the declaration of independence from the Spanish empire, the union o
89.
by mquander 4y ago
> It isn't "critically evaluate" but rather "in this stack of 100 resumes, find four that stand out sufficiently to bring in for interviews." The "stand out significantly" check doesn't take 10 min
90.
by mquander 4y ago
jedberg, I know you are a smart guy, so I wonder if you can explain this thing that has always mystified me. The difference between a good hire and a bad hire is really big, right? So why would you be like, there's no way I am reading

More ›