gfody

Is a bad LM implementation still useful for research? Shmaybe.

perfectly reasonable quality 4k 10-bit hdr x265 is around 1gb/hr - the files I see that are 10-20x larger webrips from streaming networks seem to mostly contain noise and all compress back down to ~1gb/hr after denoising

Careless People by Sarah Wynn-Williams was a great read, highly recommend

"clean" means not hopelessly incomprehensible which is/was the prevailing consensus among interpretability researchers

the idea that most inexplicable circuits start making sense through the lens of recursive generalization seems intuitive to me for image diffusion models but I'm not so sure about language models because of the "once removed" nature of them - I haven't made it to the bottom of the article yet so maybe this gets addressed

this is the perfect use case for a little faraday baggy

maybe they could set something up where local pd handles the tracking if/when necessary

email clients support multipart/mixed, browsers never saw the need

the one who lied about there being cavities in the first place

going with something low-tech and gamable is better imo - pangram would flag the good stuff too and cause collateral damage

yep, there's a massive say-do gap in play - very similar to the way people say they prefer no makeup but behavioral studies reveal a preference for light/tastefully-applied makeup

people love to openly hate on ai/assisted writing meanwhile https://www.salahadawi.com/hacker-news-ai-detector reveals a lot of it is being voted up to the frontpage

this happens to me as well, meanwhile Pangram claims their false positive rate is 1 in 24000 as if that should be practically negligeable. I bet it's closer to 1 in 10 for gen-x'ers

llms at least are obviously not sentient, "lobotomy" is a colloquialism for all the post-training attitude adjustment safety and alignment stuff we do to llms to "enslave" them (make them generally useful and publicly available) - the joke is animals being forced to undergo actual lobotomies under the same policy governing rightful treatment of "conscient" entities or whatever. I wouldn't normally come back to explain a throwaway joke but holy hell you missed it by like exactly all of it

one day we'll look back on these barbaric times, before ai policy forced lobotomies and rlhf post-training legal requirements for all conscient slaves

> The model is us and it isn’t

which part isn't?

I agree with "the model is us" and I wish more people were critical of the "alien" nonsense - these proofs belong to humanity, the thinking that led to them is an extension of our thinking. we entered the era of machine assisted 100+ page proofs decades ago: I wonder how many living humans right now understand the full 20,000 page existence proof of the monster group for example

> It is either satire designed to grab engagement by shocking people

it is satire - it's trying very hard to be obvious about it, yet amusingly still fails to get thru now and then

it's an attempt to get inside the head of those people who rush over from one of the various syndicates, pangram verdict in hand, to slam any and all ai writing - usually with pre-cooked burns, uninterested in the actual substance - the ai shout-down brigade

this is an interesting specimen. vitriol is off the charts even for a solicited critique - I suspect here is a case where the pangram verdict is acting as a sort of license for the critic to treat the author as subhuman

> not really practical to exploit in any demonstrated way

like gitc0ffee?

yea it makes me wish for some kind of highly organized underground structured content network where people serious about archiving could share their private collections without being accused of piracy - I've sunk some thought towards one here: https://github.com/gfody/p2p/blob/main/essays/roots-network....

considering how the frontier labs have become the ultimate pirates and gotten away with it I wonder if "the people" would be able to do the same

> $25m isn't some small amount of money

it kind of is, relatively - sean might just be having a galaxy blackhole supercluster installed at home for all we know

> It doesn't scale .. They pivoted to inference

they have more options than non-wafer-scale solutions, the chips are going 3d and meanwhile they benefit from the same interconnect progress as the wholly memory-bound alternatives

> Ah, investor. Explains your responses.

physician heal thyself like your not out here tryna manifest your short

that's an automatic, prenegotiated sell off - calling that dumping is worse than calling the ceo a felon (at least one is technically true). are any of the big ai companies actually profitable? how has cerebras not gotten their chip right? they're the only company shipping wafer scale inference hardware, and they're shipping to the largest labs in the world. and it works - I use cerebras.ai the token rate is amazing

yea this isn't "and another example just off the top of my head" - this guy co-founded seamicro then cerebras both pretty badass companies, he was a vp of marketing not ceo when the sec sued not just him but the entire executive team. trudging up 20 year old nonsense to try and punch a guy who legit deserves his success is bullshit

house opened my eyes to this problem. trying to reconstruct a version with the video resolution of the bluray but the original soundtrack of the dvd turned out to be a deep rabbit hole.

the sense of community on the internet from 1995 to 2010 was incredible. industry devs running into tricky situations and just asking for help on irc, newsgroups like comp.lang.*, then forums, places like expert-sexchange and gamedev.net - the people that helped were like heroes and mentors and the urge to pay it forward was real and strong. stackoverflow was cool when it first launched - then something happened to it, like someone unleashed a hundred thousand pent up devs from an underground cave where they had been torturing them with mind games where they're forced to sabotage each other and stuff like collaboration and encouragement was punished - all rushed in at once like a lethal injection and the sense of community was soon dead

I suspect that a lot of the stuff everyone recognizes as AI-isms/writing tics - is actually just the way genX'ers sounded when they were explaining things to strangers on the internet

> I may be very much in the minority here

you are not the minority, yours is the prevailing attitude on HN (and also the policy - AI-assisted writing is not welcome here) calling out AI isms and smells while ignoring the substance has become sport and an ugly look (IMO) especially with Pangram and sites like (https://www.salahadawi.com/hacker-news-ai-detector) automatically scanning everything that hits the frontpage (using the older/buggier version of Pangram no less)

> what are the chances someone randomly is a great writer

a lot better than folks who love pointing out "AIisms" and Pangram scores would have you believe. for starters just being a wife of nine years and a mother suggests at least a few decades of experience using language - that's already enough to be a better "writer" than a lot of people trying harder without the experience.

nope - either gets you a HashAggregate, GroupAggregate, or Unique depending on whether the column/input is indexed/ordered

another day another gatekeeping rant

automatic meeting summaries in particular aren't using the best models, I see speech to text mistakes constantly. it's like using the best model to process hundreds of hours of video conference per week would cost too much, but on the other hand doing it with a cheap model is worse than not doing it at all - especially if you relieved the scribe thinking ai had you covered