nmca

105 ms·

61.
by nmca 2y ago
You get what you incentivise, not what you pay for.
62.
by nmca 2y ago
Thank you very much for the reference!
63.
by nmca 2y ago
A wonderful treatise on the same topic, “Reinforcement Learning Bit by Bit”, for anyone looking for a more advanced treatment of explore/exploit. https://arxiv.org/abs/2103.04047
64.
by nmca 2y ago
I already get a driverless taxi a few times a week
65.
by nmca 2y ago
Seems very promising! Couple of qs: 1) How high is the pubkey overhead / isn’t this a lot of cryptography? 2) How do migrations work?
66.
by nmca 2y ago
> Object storage is an amazing technology. It provides highly-durable, highly-scalable, highly-available storage at a great cost. I don’t know if this was intended to be intentional funny, but there is a little ambiguity in the expressio
67.
by nmca 2y ago
Sports betting should be banned because knowing the outcome of sports games is in advance is not socially valuable. Gambling on many other things e.g elections, crime rates, ceo performance, etc should be allowed because predictions there a
68.
by nmca 2y ago
Note the selection effect in “a mediocre graduate student” (that got to work with Terry Tao)
69.
by nmca 2y ago
It was asked to explain step by step.
70.
by nmca 2y ago
I have spent some time doing this for these benchmarks — the model still does make mistakes. Of the questions I can understand, (roughly half in this case) about half were real errors and half were broken questions.
71.
by nmca 2y ago
hm - do random end times solve that particular issue?
72.
by nmca 2y ago
It has functions which replace macros. One would use it instead of latex because it is immeasurably faster and more sane.
73.
by nmca 2y ago
please use this opportunity to reflect on whether ARC measures reasoning skills :)
74.
by nmca 2y ago
worth noting that longevity escape velocity is well-posed but typically badly miscalculated. You care about the right tail of ages, not the mean, and the tail has moved slowly if at all. Tracking rate of increase of max or 99th percentile w
75.
by nmca 2y ago
https://arxiv.org/abs/1811.07871
76.
by nmca 2y ago
A critic for the critic would be “Recursive Reward Modelling”, an exciting idea that has not been made to work in the real world yet.
77.
by nmca 2y ago
someone’s site (I forget whose) has a built in inflation adjuster so that they write dollars(x, year) and get a future-proof value. Very cute!
78.
by nmca 2y ago
I had the same temperature bug recently despite changing the AC, hot air was coming from the vents. Not a hot day. Very odd.
79.
by nmca 2y ago
thankfully we can just wait and see here. concretely, I predict time from first multimodal llm that can reliably read a chessboard and analogue clock without finetuning (obviously not reasoning) until ARC is solved is <4 months
80.
by nmca 2y ago
can you propose a concrete prediction here?
81.
by nmca 2y ago
ARC is a noble endeavour but mistakes visual/spatial reasoning for reasoning and thus fails.
82.
by nmca 2y ago
Prediction markets on the outcome: https://manifold.markets/JacobPfau/will-the-arcagi-grand-pri...
83.
by nmca 2y ago
This is already true of many consumer vehicles and thus less exacerbated by Waymo than you imply
84.
by nmca 2y ago
suppose we design a reasonable experiment here, like taking a random sample of teenagers and mitigating for anonymity concerns. are you seriously claiming that this would show an increase after the ban? or to put it another way --- the phen
85.
by nmca 2y ago
ublock origin and Firefox mobile are a lovely combination!
86.
by nmca 2y ago
A different criticism of mapreduce-like technologies, and one of my all time favourite papers in any field, is: Scalability, but at what COST? By Frank McSherry. If you enjoy dist-systems bashing or honestly engineersing in general it'
87.
OpenAI – transformer debugger release(github.com)
362 points by nmca 3y ago|118 comments
88.
by nmca 3y ago
This response represents some confusion on your part and not a paradox or counterargument to rationality. very very roughly the ideas you need are: 1. survivorship bias 2. self-interested optimisation is not guaranteed to lead to a global m
89.
by nmca 3y ago
I like methods of this flavour, pioneered as I understand it by Fabio Petroni. Very elegant, particularly because you can change the distribution over substrings in O(params) time instead of O(index size). What's funny about them is th
90.
by nmca 3y ago
Few counterpoints: * AI is better for juniors than seniors. It doesn't do legwork; it just provides knowledge. * Blind is a toxic hellscape and redditors are broadly pretty dim. Plausibly AI is a problem for your employment in like ~2

More ›