polygamous_bat

13 ms·

31.
by polygamous_bat 2y ago
> Also of interest, he has won the Abel Prize in 2021, making it a rather unique combination of winning the top honors in both theoretical/abstract math & CS The overlap between theoretical CS and math is way larger than most pe
32.
by polygamous_bat 2y ago
> It's too late for the person to receive any benefit. They won't know or care. Funerals are rarely for the dead, despite what we tell ourselves they are mostly for the living. The benefit of the program, as I see it, is the re
33.
by polygamous_bat 2y ago
I commented this somewhere else, but word in the ether is that OLMo is not actually that good of a model given its size and compute budget. I am not entirely sure why, and it’s still good to have the full recipe for at least one model out i
34.
by polygamous_bat 2y ago
From what I heard through the grapevine, OLMo is not nearly the best model for its size or compute budget. Apparently something didn’t quite go right and AI2 didn’t have the money to train until they got it right.
35.
by polygamous_bat 2y ago
Think of the term b in y = Wx+b. W is called weight, b is called bias.
36.
by polygamous_bat 2y ago
Gary Marcus has been calling the end of the AI bubble since 2012 or something.
37.
by polygamous_bat 3y ago
GPT4 is the SOTA, for better or for worse, so it would be just stupid to not use it just for PR reasons.
38.
by polygamous_bat 3y ago
I think you may enjoy reading about Mcnamara’s fallacy, coincidentally also related McNamara of the Vietnam war: https://en.m.wikipedia.org/wiki/McNamara_fallacy I hope it makes you reconsider your position about disre
39.
by polygamous_bat 3y ago
A lot of it might be coming from deteriorating social bonds. A propensity to see your neighbor as a number, a statistic, rather than a fully fleshed human being with the same human needs and weaknesses as you.
40.
by polygamous_bat 3y ago
> Eugenics by free market That hit harder than I expected, for someone whose loved ones were crippled by primarily schizophrenia. Even two generations back, such people would be cared by family or community, but now they have practically
41.
by polygamous_bat 3y ago
AFAIK RT-2 doesn’t quite work outside of Googles micro-kitchen, where they collected about a 1000 hours of data.
42.
by polygamous_bat 3y ago
There are two factors to this. 1. If your partner is not in a PhD program themselves, then you can rely on them during what I call “crunch time”, and there are a few of those in every PhD when you’re doing 16 hour days 7 days a week just to
43.
by polygamous_bat 3y ago
Generally when companies seek to avoid liability, they push to have their employees use telephone to communicate so that compromising emails don’t come out during the discovery period of a lawsuit. I imagine this is one of such cases.
44.
by polygamous_bat 3y ago
> Why does the stock market reward idiot shit like this? Goodhart’s law, unfortunately. Whatever metric the stock market rewards gets gamed like there is no tomorrow (or next quarter, here.)
45.
by polygamous_bat 3y ago
> He got caught up in it all and you bend, bend, bend until the next thing you know.. This denies him agency. Probably a better hypothesis is that he was a raging narcissist who thought he could swindle everyone with none the wiser, with
46.
by polygamous_bat 3y ago
No no, he is rich so of course he is just a kid, lost in the sauce, accidentally squandering a few billion. Now, 16 year old male from a low income household, those are men, and should know the consequence of their actions better. /s T
47.
by polygamous_bat 3y ago
Exactly this. I don’t know many lower income folks affording on call doctors, and those who can afford it can also afford an extra $15 charge on their bill, god knows how much they’re paying already anyway.
48.
by polygamous_bat 3y ago
Free street side parking, I imagine? I would like to see that.
49.
by polygamous_bat 3y ago
Except in lower Manhattan those bus lanes are taken over by clueless tourists or entitled folks driving around. Pretty much an everyday occasion near Union square.
50.
by polygamous_bat 3y ago
> Every single one of my passwords is unique and randomly generated and at least 32 characters, none of them are getting brute forced unless there is a sudden gigantic leap in quantum computing. One of the threat models that I consider i
51.
by polygamous_bat 3y ago
Really no wonder he ended up in “decentralized AI” aka crypto grifting. It must be like returning home after a long day.
52.
by polygamous_bat 3y ago
https://en.wikipedia.org/wiki/Emad_Mostaque Read his Wikipedia page and tell me he doesn’t sound like your run of the mill crypto scammer. > He claims that he holds B.A. and M.A. degrees in mathematics and computer
53.
by polygamous_bat 3y ago
AFAIK this was the default, copy paste impact statement by ICML template.
54.
by polygamous_bat 3y ago
I think the two modes of LLM discourse: “they’re conscious!/they’re just next token predictors with impressive datasets” comes largely from two different groups of people: those who learned about LLMs before learning about ML fundament
55.
by polygamous_bat 3y ago
I've always used dust as a replacement, and so I am curious to know if you have tried both tools: do you have thoughts on what makes ncdu better?
56.
by polygamous_bat 3y ago
How is building a dataset easier than using a prebuilt dataset?
57.
by polygamous_bat 3y ago
> We need to operate society for the good of humanity, not for the benefit of 'the market' and 'profits'. Good of humanity is a hard to achieve/track goal, so we traded that in for an easier to track metric calle
58.
by polygamous_bat 3y ago
> Not sure why it makes sense to complain about the EU and not the companies. Unfortunately a non-negligible number of people in tech also have libertarian leanings, with a default “gubmint bad!” position, which makes them easy prey for
59.
by polygamous_bat 3y ago
To a far, far smaller extent than banks, but yes.
60.
by polygamous_bat 3y ago
One primary difference is between a discrete (think chess/go moves) and a continuous (think steering wheel input) control space. Even theoretically, the error in one is bounded linearly while the other can have an exponential error wit

More ›