Yes, but not in a way sufficient to resolve this question positively.

— edited by ugandamaximum

Community median seems too low. Trump didn't lose by much in 2020 and likely wouldn't have lost if not for covid. The Republican base seems to love him so he should get their nomination if he does run. I would predict ~50% if it is a given he is running as a Republican. The 25% discount is based mostly on the significant chance that he'll decide not to run and less on the chance that he'll run as an independent. I also wouldn't discount the possibility of actual electoral fraud benefiting the Rs if he runs under that platform. This arguably happened in ...

Looks like their government is threatening to block youtube.

Apparently this is nominally related to two German-language RT channels being blocked.

— edited by ugandamaximum

The debate itself is setup in such a way that it would be extremely unlikely for a challenger to win: >Both sides will first agree on two judges with strong analytical skills, relevant experience, no previous endorsement of either side, no relevant political biases, and who declare they will examine both hypotheses equally. As specified the judge criteria is going to be next to impossible to fulfill, which is why right afterward they suggest that the rules can be bent in certain cases. Fair enough. >The debate will be based on all currently available...

There are two major problem with the new design for those of us who like to browse large numbers of questions:

1) Density of questions is reduced, more scrolling needed to see equivalent number of questions.

2) Question titles broken up with data preview instead of simply listed one after the other, making question-to-question browsing less readable.

To solve this you can have an option to collapse the data previews and / or shift them to the right of the question text.

@ersatz Who says you need neurons or emulated neurons to make machines as smart as humans? Simulating the biological architecture of a human brain is a different problem than the one specified in the res.

@(Matthew_Barnett) Couldn't quickly find info on computing power for latest attempt, a decade old japanese attempt that did 2e12 digits took about 7 petaflop hours "with verification." The new record is only 30 times as big (is that right??) so you might figure ~210 petaflop hours if everything scales which I'm not sure about. According to [this paper](https://arxiv.org/pdf/2005.14165.pdf) gpt3's biggest version took ~3600 petaflop days or 86400 petaflop hours to train. So compute time to train the biggest GPT-3 version should be over 400 times as expe...

@GoldbergMachinations And tweak it higher again because both of his parents were fairly long-lived especially given that they were born at the start of the 20th century. Genetics are important for longevity. Even with the man's notorious lifestyle I'd give a >95% chance that he survives in good health long enough to contend for the nomination.

— edited by ugandamaximum

It seems the biggest unknown might not be the performance of the machine on well-specified questions but the ability of the researchers to ask questions exploiting any differences between the machine's architecture and the human brains. Maybe they put a bunch of compute into machine intelligence toward this end too. The question does not specify how much knowledge the expert interviewers conducting the test will have of the candidate machine intelligence or its architecture, nor how far in advance they can prepare the test questions. Unless this so-ca...
@(RyanBeck) Targets like this should be the very tip of the iceberg for an all-out strategic nuclear war. It is most likely that quick-delivery weapons including silo'd and sea-launched would be used primarily to strike nuclear weapons delivery platforms and storage facilities. But in addition to strictly military targets that might entail collateral damage if struck, you also have at least the following: --Airports capable of hosting strategic aircraft. In practice this includes big airports in and around major cities. --Ports for shipping, the most ...
There are a few reasons to think we won't discover any alien technosignatures by 2050: 1) Search effort severely resource constrained 2) Previous searches turn up zilch 3) Even if we find a technosignature, the odds of it being unambiguous enough to meet res criteria are not clear. Maybe we've already found one. 4) To paraphrase Eliezer Yudkowsky, the minimum IQ required to destroy the world drops by a point every 18 months. This is an ongoing issue for humanity, and advanced civilizations may tend toward annihilation or primitivization right around ...

Given Fauci's recent endorsement to investigate lab origins one would imagine that at least one secret intelligence assessment using resolution-qualifying language probably already exists. Whether any such assessment will be made public in the next year seems like a good 50 / 50 at this point. Actually, 50 / 50 might be a tad optimistic but the points spread is good.

— edited by ugandamaximum

@(FNORD) Apparently the main reason for the new design is to look professional, with a secondary goal of mobile-friendliness (not that I had any problem browsing the old version on mobile) and maybe tertiary goal of better data displays (which IMO is the only actual improvement of the 3). So letting people choose the old design at another URL as the new design used to be available in preview or by other option wouldn't interfere with any of those considerations and would also be easy, since the old design already exists. I already asked for a link to t...
My feeling is that it makes most sense to keep your AGI secret as long as possible, since the advantage is magnified if adversaries aren't aware and trying to respond, and also because such an announcement could be extremely destabilizing. Of course, this might be much easier said than done, so the real question may be how long until the info gets leaked, and how credible would a leak be? What are some reasons why a rational organization with a national-security interest in AGI would want to announce their capability? I'm having trouble coming up with a...

@metani The strongest argument against lab release is that there's no non-circumstantial evidence of lab release, which isn't a very strong argument.

@Matthew_Barnett Until Simintov's whereabouts become public knowledge this is still a question about the unknown.

@(johnnycaffeine) It's not nearly so clear cut that China would lose a war, even with determined resistance, US allies and no nukes. Two other things to keep in mind are that Taiwan may surrender very quickly from the start of hostilities, and it's not guaranteed that the west would actually respond in a determined fashion which would involved striking mainland China with incredible risk of uncontrollable escalation. The main consideration keeping China from starting a war and the US from absolutely responding in a determined fashion is that either opt...

One obvious lucrative use for transformer language models generally and GPT series specifically is hard-to-detect adversarial botting for battles of public opinion. I would be shocked if trials of this haven't already started.

Unfortunately I think it would be hard to prove GPT is being used for this in the first place and even harder to get a firm dollar value on any results.