bo1024

11 ms·

151.
Adversarial Binary Search (or, how to fleece Ballmer)(bowaggoner.com)
5 points by bo1024 2y ago|2 comments
152.
by bo1024 2y ago
Hi all, there was a lot of interest on the post from a few days ago about how to solve for the equilibrium. Hope you enjoy!
153.
by bo1024 2y ago
It's reassuring that such things still exist. My mental model of consumer hardware is that they take devices like the ones you describe, and just add wifi, bluetooth, telemetry, ads, and an app.
154.
by bo1024 2y ago
You're already using your own randomness to pick the pivot at random, so I don't see why the shuffle helps more. But yes, if your randomness is trustworthy, the probability of more than O(n) runtime is very low.
155.
by bo1024 2y ago
They say this: > We have IP-based rate limiting in place for many of our endpoints, however these crawlers are coming from a large number of IP addresses, so our rate limiting is not effective. Do you have something else in mind? Just sh
156.
by bo1024 2y ago
It would be nice to have more of a peer-to-peer infrastructure (torrent inspired) for serving big resources.
157.
by bo1024 2y ago
Yeah, the word is "uniform", e.g. a uniform family of circuits is one where there is a Turing machine where, for each n, it outputs the circuit for inputs of size n.
158.
by bo1024 2y ago
In these situations one can also ask about the "adaptivity gap", which is the difference in how many tests are required if one is allowed to be adaptive -- make a test, look at the result, then decide what to test next -- versus n
159.
by bo1024 2y ago
I think the article does a nice job connecting these bugs to other historical, non-AI bugs that result in a racially biased system.
160.
by bo1024 2y ago
I believe Lewis was there at the time, so I think this is firsthand from him. Whether you believe him or not is up to you, of course.
161.
by bo1024 2y ago
Source: "Going Infinite" by Michael Lewis. Chapter 9. > Gary never said when he would leave, or how he would leave—which was an issue, as the Bahamas had taken his passport. On Sunday night, without a further word to anyone, h
162.
by bo1024 2y ago
Sure, "the CIA" can do it, but my point is that some particular person actually had to coordinate a bunch of people and actions to make this happen, quite quickly after the collapse and before it was even clear whether SBF would b
163.
by bo1024 2y ago
I think there is a very interesting untold story in how the U.S. proactively and aggressively pursued this case. If I recall Michael Lewis' book right: - FTX (international) was a somewhat separate company from FTX US; SBF was convince
164.
by bo1024 2y ago
I thought the argument was that adding functional AM receivers to EVs is difficult?
165.
by bo1024 2y ago
This is very well put. I think the next point of discussion is what fraction of "software developers" "need" to know how to code in the deep way you describe.
166.
by bo1024 2y ago
I agree, at least in theory (not sure if any existing blockchains could adequately solve the problem today). Someday, you could imagine essentially running gitlab on a blockchain. Actually, it's not a stretch to include access control
167.
by bo1024 2y ago
Do you need to measure VO2 max specifically? If your form of exercise is jogging/running, you will easily be able to tell if you're improving by tracking distance and time of your runs. If you want to be precise, you can e.g. race
168.
by bo1024 2y ago
If the authors don't draw conclusions from that, I wouldn't either. For example, suppose the main impact of exercise on cancer risk were simply the correlation with time spent outdoors in the sun. Then you would expect to see a ch
169.
by bo1024 2y ago
They did say "simple", not "easy"...
170.
by bo1024 2y ago
Yes, most of experienced runners' running is at a conversational pace, often literally (this is why running groups are popular). To explain why harder isn't better: improvement in running comes only a little from muscular strength
171.
by bo1024 2y ago
One thing that surprised me, as a juror, was that nobody ever simply submitted a piece of evidence. Each of the hundreds of exhibits was presented with a person on the stand attesting to its provenance, under penalty of perjury.
172.
by bo1024 2y ago
I think this is a great start to make more private LLM use accessible. Excited for the possibilities of where this can go.
173.
by bo1024 2y ago
And they have agreements with the providers not to use your chats for training, and to delete them after 30 days (I believe).
174.
by bo1024 3y ago
(Not who you asked, but) one important point here is definitely that BPP is about yes/no decision problems, and approximation algorithms are generally a different beast.
175.
by bo1024 3y ago
> Perhaps: Am I incorrect in assuming that randomized 2-approximation is somehow transformable to 'correctness in the decision problem 2/3 of the time'? Yes, that's the issue. For another example, Max-Cut is NP-hard.
176.
by bo1024 3y ago
BPP is the complexity class for decision problems that a randomized algorithm can solve in polynomial time, in the sense that on every input (worst-case), the algorithm is right with at least 2/3 probability (over its own randomness)
177.
by bo1024 3y ago
They are generally esoteric conjectures similar in spirit to P != NP. For example, the third paper uses the assumption "E requires exponential circuits". Here E is the class of problems solvable in exponential time, or 2^O(n) time
178.
by bo1024 3y ago
> my job is to enable higher ups to do their job Ouch, I like to think it's the other way around.
179.
by bo1024 3y ago
Yeah, with forwarding, I am seeing DKIM still passes but SPF fails.
180.
by bo1024 3y ago
What do you mean by the email delivery problem? Do you mean setting up DKIM, SPF, and DMARC? It's not that hard, compared to setting up an email server in the first place.

More ›