@gsvaca you may be mistaken on the nature of ranking changes
it’s not something that someone hands down to you to simply program; you have to wring the initial insight out of the substrate through your own investigation and scientific inquiry
i don’t know how to estimate the nyt articles readership but i would be willing to bet that the circulation of negative social media buzz is much higher
amazed @ very smart people concern trolling about whether or not its "responsible" to allow people to post pseudonymously online
it should be axiomatic
@ChrisWGuthrie i don't think this is good since dissidents must be protected and i find this to be a higher priority than preventing the marginal cybercrime
take down the porn and gore with ML tagging, and inference the identities of your users if you must but make it clear from day 1
@aquariusacquah@rglpwx I guess in this case the point is that I’ve seen a notable economist and a few OpenAI guys doing the “I’m just asking questions” thing about whether anons should even be allowed
@aquariusacquah@rglpwx there seems to be a significant cultural screen at openai for ppl who are rationalist / effective altruist / etc. ive heard them go off about gwern at neurips
@aquariusacquah@rglpwx so it’s kind of interesting watching them do a normie lib dance where they don’t want to question the sensibilities of the Times but have also probably been SSC readers for years and respect Scott Alexander a lot
but I agree most of this discourse is lukewarm
unemployed finance midwits - “Bitcoin is tulip mania, it’s pure pump and dump, elon will go to jail—“
actual hedge fund executives - “Bitcoin is valuable because people want Bitcoin”
@Real_Lets_Talk@thesravaka it’s not even that everyone should optimize for speed and take all the joy out of it. it’s just that I’ll get bored if it’s too slow now lol
there is something here and it's not just a leftist thing. americans are definitely more averse to hiring help than other cultures but i'm not sure why
>in the year 20xx, I am peacefully raising cattle on my homestead
>all of a sudden the carbon police raid my property and shoot my cows
>mfw i am named a carbon terrorist and sent to exile in the off world colonies
the answer as always is Technology
there should be a computer in your home making power consumption decisions based on your preferences and data feeds of instant price https://x.com/AlanMCole/status/1361688819335069699
@zcknln even without forward capacity markets, shouldn’t there be simple feedback loops like for example: turning down electric heating temp when prices are high, leaving more capacity on the grid
rather than rolling blackouts
@Teleonomic so complicated though. I’m hugely bullish on active feedback intervention like RNS or other BMI tho. Is there any human system that’s poorly modulated & tacklable by intelligence in your view?
@valueless_user makes sense yeah — this is also for the benefit of human operators to understand the system as modular components. can think of many ways to gain efficiency that will ultimately destroy engineering productivity
@postpostpostr@BroductManager yes lol
people see in the tech industry whatever they want to see in it
you will find people working on any problem you can conceive
r u using Theranos or Tesla as the exemplar
@SpeaksNanda bc its an enormously powerful tool to reason about decision making under uncertainty
that being said i dont understand the quasi spiritual bayesianism either
@erf_of_why based. I tried to plug that data into FB prophet once but eventually got lazy re: checking it every day. the fees on predictit were too brutal to make real money
@valueless_user@HeikoPierre@R1Jack CCS for coal plants adds LCOE of around $30/mwh which is not great for an already losing industry
the question now is really how cheap CCS can get and if it can make gas viable in a carbon tax environment
@95thoughts@wolftivy@mattparlmer lol horrifying
and also sounds way harder than bottom up but what do i know
the difference between understanding two cell types and plating them together vs how to get rid of entire structural components?
@valueless_user@alth0u To be honest control theory is a lot harder than machine learning lol
But the constellation control systems are relatively simple — I think they mostly don’t move
@95thoughts@wolftivy@mattparlmer there is probably some bio electric trigger to get cells to build these structures yeah? also what exactly does vascularity add to the texture/flavor vs heme
bear in mind I have no idea what I’m talking about
@powerbottomdad1 we shouldn’t compete with the UI/UX of excel
just boost tf out the backend or build a near exact clone that has compatibility with xlsx
@TeddyRaccovelt it can be argued whether memes are invented or discovered (just like math!) but i feel that you can accurately call a sufficiently spread idea a group mind
@eigenrobot@ollybot_redux pls explain. ive been seeing them for years & still no clue what’s happening . one of the most inscrutable affiliation emojis
"when there is a gold rush sell shovels"
it's a bigger business than you think to tutor people in pure leetcode type algorithmic problem solving
and as unis found out long ago the expected income boost is higher than e.g. raising SAT 100 pts you can charge them way more https://x.com/grantadever/status/1362786933210894336
@balensphere@averykimball the strongest and smartest humans are still finite and weak
the former should avoid wrestling with a silverback and the latter should avoid the lovecraftian secrets
when I joined my team at {big_tech_co} i did a little intro post and said a thing about how i like election gambling
then my skip immediately mogged by posting his PredictIt rank (two levels higher than me)
@jasoncrawford@mattdemonte lol from where I’m standing it looks kind of fundamental but I guess it’s all relative
im more interested in pharma companies failing to do more basic bio research
http://gather.town is so good for conference and work purposes that im wondering why there aren’t any analogous social networks
there should be a twitter scale http://gather.town
not to mention the demand side of the equation: people are drawn online by their hunger for dissenting and controversial arguments, of which they’ll find a world class supply on twitter
it’s hard not to keep changing your mind under these conditions
@alth0u afaict engagement data is far more important than content level understanding in news feed recommender systems
but twitter has an uncanny tendency to group similar tweets together so idk
@WilliamGrobman i honestly thought the initial acceleration was from the wheels
after seeing this image and thinking about it that doesn't make any goddamn sense and i feel stupid
@divinix_chi@WilliamGrobman no matter how fast the conveyor is moving you can't make the plane "effectively stationary"
all you're doing is spinning the wheels super fast and applying some rolling friction
@divinix_chi@WilliamGrobman imagine holding down a toy car gently on a treadmill. does it take a lot of force to move it forward with your hands even if the wheels are rapidly spinning?
@Hellachans your logic seems correct but the answer backward
if the wheels don’t matter and we generate thrust via jet engine then we can create acceleration. the wheels just spin freely
prediction market governance enthusiasts are really excited about a form of leadership structure that hasn’t even been used successfully at a small company
@antirobust I thought Walter Isaacson did a pretty bangup job at both physics and computing (Einstein and The Innovators respectively)
but he may be the exception that proves the rule
context: the common neuroscientific theory is that immediate memory (<30s) is current synaptic activations in the brain, short term memory (~1 day) is a process in the hippocampus, and long term memory (~indefinite) is actual synaptic weights
GPT-3 has a significant "long term memory" as in it memorizes large tracts of the internet
but if you had to have it track a coherent conversation across several pages of text it would be helpless
@sadmachiavel the job of bettors in prediction market based governance is to reliably predict the conditional outcomes regardless of which outcome they actually want
low skill bettors and market manipulators will run out of money in the conditional prediction market paradigm
@averykimball@kartographien and? are we not machines that regurgitate new combinations of the content we've seen all our lives?
not to mention the distillation of genomic learnings from ancestor generations
@rglpwx@QuantumSeany I would argue that this unsupervised learning autoregression goes on in the backdrop of life (while reading or conversing) at all times. but yeah it's secondary learning whereas the primary may be closer to reinforcement e.g. interacting with social agents and observing responses
@rglpwx@QuantumSeany however i don't think there's any reason to believe that the way humans get to intelligence is the only way to do it. I bet a sufficiently large autoregressive LM could become intelligent even without ever having to interact with anything
@rglpwx@QuantumSeany and at some point the most efficient way to predict the next token in a physical reasoning problem is to derive it from first principles
@rglpwx@QuantumSeany in my personal theory of mind i demarcate consciousness/agency from intelligence
the former characterizes an OODA loop and the latter is a kind of brutish metric -- can you come to the right answer?
@hyperdiscogirl greatest tweet ever for me personally. created a market panic, price run up, and subsequent sell off after the fiasco. the stonks dropped quite a bit and I bought a ton of it. now they're worth 15x what they were then
@rglpwx@QuantumSeany why assume GPT can't do adversarial situations? it would have to develop agent interaction models in order to get anywhere in story text completion. that's what makes it write "good" stories
good in scare quotes bc it's all relative
@reweirding the median person is probably fatter but also has access to weapons grade cosmetics and stuff
but on the far end the most beautiful people have gotten much better looking. Same in speed or strength
weyl and scott arguments are hilarious because they both boil down to “good technocracy is good and bad technocracy is bad”
neither has any insight into how to get it right more often (but at least Scott admits it)
@lisatomic5 even something like twitter gives u tremendous powers to change your life for the better but we mostly use it to keep smashing the dopamine button
@SeyoneC but yeah joking aside I think the top candidates are have multiple first author papers at ICML/NeurIPS by the time they graduate. pretty insane but that’s where we’re headed
@abelianraisin honestly the institutions I was applying to all have very long time horizons and giant endowment pools so I’m guessing it has more to do with the field getting more competitive
honestly I used one of these and was frankly amazed by the lack of security. there’s someone being paid pennies in Malaysia on the other end who could give a shit https://t.co/O8exSUUkG6
@RobertCinci@eigenrobot wouldn't say she's a bad writer but definitely kind of frenetic & somewhat rewrites the same essay over and over
no clarity of logic but great vibes
@NewWaveRave1@eigenrobot oh i'm sure they're closed to non mutuals but he probably still gets added to tons of high volume GCs that are hard to keep up with. to each their own
this is just "bravery debate" type stuff
the argument isn't really about whether AI x-risk is real but rather if too much resources are being allocated to it, which is an unanswerable question https://x.com/Noahpinion/status/1364735783597350913
@ChrisCroy big fan of OpenAI's approach, where they give up on the x-risk component and just say ok the best result will be if we beat everyone else to AGI
@justjoshinyou13@ChrisCroy i very much doubt that MIRI is even a serious player in the AI research game, even in the Alignment/Safety sector
it's hard to attract top scientists somewhere they can't publicize their results
also worth mentioning that one of the most public and vocal proponents of AI x-risk research simultaneously builds electric cars and reusable rockets and probably isn't using it as "an excuse to worry about nothing"
@BecomingCritter@pervexists69 there's a difference between depression (which seemingly half the country is diagnosed with) and giving up on life / being suicidal
latter feels much rarer
@GuilleAngeris 90% of applicants to CS departments these days mark their primary interest as ML and it's getting pretty bleak it seems out there
but I think what the reddit guy here is describing is a surefire 100% candidate and not the average student that gets in
people pitch starting companies as the antidote to credentalism but realistically the Stanford grad entrepreneur attracts all the top talent and VC funding