Need to let loose a primal scream without collecting footnotes first? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid: Welcome to the Stubsack, your first port of call for learning fresh Awful youāll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cutānāpaste it into its own post ā thereās no quota for posting and the bar really isnāt that high.
The post Xitter web has spawned soo many āesotericā right wing freaks, but thereās no appropriate sneer-space for them. Iām talking redscare-ish, reality challenged āculture criticsā who write about everything but understand nothing. Iām talking about reply-guys who make the same 6 tweets about the same 3 subjects. Theyāre inescapable at this point, yet I donāt see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldnāt be surgeons because they didnāt believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I canāt escape them, I would love to sneer at them.
(Semi-obligatory thanks to @dgerard for starting this.)
You gotta love how in the announcement the guy is so blatantly āhey they said and did such nice things for me that I just got a throw them a bone, and if releasing the leader of a notorious drug bazaar who tried to put out a hit on one of his employees is what they want then they can have it!ā
Sidenote: AFAIK, even with this pardon, Ulbricht still ended up spending more time in prison than if he took a plea deal he was reportedly offered:
He was offered a plea deal, which would have likely given him a decade-long sentence, with the ability to get out early on good behavior. Worst-case scenario, he would have spent five years in a medium-security prison and been freed.
Gotta say, this whole situationās reminding me of SBF - both of them thought they could outsmart the Feds, and both received much harsher sentences than rich white collar criminals usually get as a result.
Ah yes that will be good for international relations and the morale of law enforcement and anti cybercrime people. Lol it is all so stupid.
This and the releasing of the jan 6 people who assaulted cops (one cop who testified against them got a shitton of messages they got early release) is going to do wonders. Not that it will shake the belief of a lot of people that the repubs are the party of back the blue and law and order.
I know a lot of people are looking for alternatives for programs as stuff is ennshitfying, rot economying, slurping up your data, going all in on llms etc https://european-alternatives.eu/ might help. Have not looked into it myself btw.
trump just dumped half trillion dollars into openai-softbank-oracle thing https://eu.usatoday.com/story/news/politics/elections/2025/01/21/trump-stargate-ai-openai-oracle-softbank/77861568007/
youād think itās a perfect bait for saudi sovereign wealth fund, and perhaps it is
for comparison, assuming current levels of spending, this will be something around 1/10 of defense spending in the same timeframe. which goes to, among other things, payrolls of millions of people and maintenance, procurement and development of rather pricey weapons like stealth planes (B-21 is $700M each) and nuclear-armed nuclear-powered submarines ($3.5B per Ohio-class, with $31M missiles, up to 24). this all to burn medium-sized country worth of energy to get more āimpressiveā c-suite fooling machine
CIDR 2025 is ongoing (Conference on Innovative Data Systems Research). Itās a very good conference in computer science, specifically database research (an equivalent of a journal for non-CS science). And they have a whole session on LLMs called āLLMs ARE THE NEW NO-SQLā
I didnāt have time to read the papers yet, believe me I will, but the abstracts are spicy
We systematically develop benchmarks to study [the problem] and find that standard methods answer no more than 20% of queries correctly, confirming the need for further research in this area.
(Text2SQL is Not Enough: Unifying AI and Databases with TAG, Biswal et al.)
Hey guys and gals, I have a slightly different conclusion, maybe a baseline 20% correctness is a great reason to not invest a second more of research time into this nonsense? Jesus DB Christ.
Iād also like to shoutout CIDR for setting up a separate āDATABASES AND MLā session, which is an actual research direction with interesting results (e.g. query optimizers powered by an ML model achieving better results than conventional query optimizers). At least actual professionals are not conflating ML with LLMs.
following on from this comment, it is possible to get it turned off for a Workspace Suite Account
- contact support (
?button from admin view) - ask the first person to connect you to
Workspace Support(otherwise youāll get some made-up bullshit from a person trying to buy time or Case Success or whatever, simply because they donāt have the privileges to do what youāre asking) - tell the referred-to person that you want to enable controls for āGemini for Google Workspaceā (optionally adding that you have already disabled āGemini Appā)
hopefully you spend less time on this than the 40-something minutes I had to (a lot of which was spent watching some poor support bastard start-stop typing for minutes at a time because they didnāt know how to respond to my request)
Thanks. I simply switched to Fastmail over this bullshit. (āSimplyā mileage may vary)
- contact support (
so the new feature in the next macos release 15.3 is āfuck you, apple intelligence is on by default nowā
For users new or upgrading to macOS 18.3, Apple Intelligence will be enabled automatically during Mac onboarding. Users will have access to Apple Intelligence features after setting up their devices. To disable Apple Intelligence, users will need to navigate to the Apple Intelligence & Siri Settings pane and turn off the Apple Intelligence toggle. This will disable Apple Intelligence features on their device.
IDK how helpful this is, but Apple intelligence appears to not get downloaded if you set your ipad language and your siri language to be different. I have it set to english (australia) and english (united states). Guess Iāll have to live without āgaolā support, but that just shows how much Iām willing to sacrifice.
oh boy: https://social.wake.st/@liaizon/113868769104056845 iOS devices send the contents of Signal chats to Apple Intelligence by default
e: this fortunately doesnāt seem to be accurate; excuse my haste. hereās the word from the signal forums
also, my inbox earlier:
24661 N + Jan 21 Apple Developer ( 42K) Explore the possibilities of Apple Intelligence.
hackernews: Weāre going to build utopia on Mars, reinvent money, and construct god.
also hackernews: moving off facebook is too hard :( :( :(
they will take facebook there with them. none of their space escapism will solve their problrms because they take them along. these mfers will do anything but go to therapy
deleted by creator
til that thereās not one millionaire with family business in south african mining in current american oligarchy, but at least two. (thielās father was an exec at mine in what is today Namibia). (they mined uranium). (it went towards RSA nuclear program). (thatās easily most ghoulish thing iāve learned today, but iām up only for 2h)
thereās probably a fair couple more. tracing anything de beers or a good couple of other industries will probably indicate a couple more
(my hypothesis is: the kinds of people that flourished under apartheid, the effect that had on local-developed industry, and then the āwider worldā of
opportunitiesprey they got to sink their teeth into after apartheid went away; doubly so because staying ZA-only is extremely limiting for ghouls of their sort - itās a fixed-size pool, and the still-standing apartheid-vintage capital controls are Limiting for the kinds of bullshit they want to pull)there are more it seems https://www.ft.com/content/cfbfa1e8-d8f8-42b9-b74c-dae6cc6185a0
that list undercounts far more than I expected it to
thereās gotta be way more, but frankly idk even where to begin to look
Banner start to the next US presidency, with Wiener Von Wrong tossing a Nazi salute and the ADL papering that one over as an āawkward gestureā. 2025 is going to be great for my country.
Incidentally is āWiener Von Wrongā or āWernher Von Brownnoseā better?
Perhaps āWanker von Clownā?
Ooo, I like that.
in that spirit: Loserus Inamericus
(I donāt know if that scans, I have no latin skills and I donāt feel like breaking out information to check)
Itās term time again and Iām back in college. One professor has laid out his AI policy: you should not use an AI (presumably Chat GPT) to write your assignment, but you can use an AI to proofread your assignment. This must be mentioned in the acknowledgements. He said in class that in his experience AI does not produce good results and that when asked to write about his particular field it produces work with a lot of mistakes.
Me, Iām just wondering how you can tell the difference between material generated by AI then edited by a human, and material written by a human then edited by an AI.
Here is what I wrote in the instructions for the term-paper project that I will be assigning my quantum-physics students this coming semester:
I canāt very well stop you from using a text-barfing tool. I can, however, point out that the āAIā industry is a disaster for the environment, which is the place that we all have to live in; and that it depends upon datasets made by exploiting and indeed psychologically torturing workers. The point of this project is for you to learn a physics topic and how to write physics, not for you to abase yourself before a blurry average of all the things the Internet says about quantum physics ā which, spoiler alert, includes a lot of wrong things. If you are going to spend your time at university not learning physics, there are better ways to do that than making yourself dependent upon a product that is a tech bubble waiting to pop.
I was talking to someone recently and he mentioned that he has used AI for programming. It worked out fine, but the one thing he mentioned that really stuck with me was that when it was all done, he still didnāt know how to do the task.
You can get things done, but you donāt learn how to do them.
This must be mentioned in the acknowledgements
wat
I know!
https://xcancel.com/kailentit/status/1881476039454699630
āWe did not have superintelligent relations with thatā¦ā
Reposting this for the new week thread since it truly is a record of how untrustworthy sammy and co are. Remember how OAI claimed that O3 had displayed superhuman levels on the mega hard Frontier Math exam written by Fields Medalist? Funny/totally not fishy story haha. Turns out OAI had exclusive access to that test for months and funded its creation and refused to let the creators of test publicly acknowledge this until after OAI did their big stupid magic trick.
From Subbarao Kambhampati via linkedIn:
"šš§ šš”š š¬šššš² šØš©šš¢šš¬ šØš āš©šššš ššš šš šØš®š° š“ššš šš šŖššššššššš š©šššššššš šŖšššššššā hashtag#SundayHarangue. One of the big reasons for the increased volume of āššš ššØš¦šØš«š«šØš°ā hype has been o3ās performance on the āfrontier mathā benchmarkāsomething that other models basically had no handle on.
We are now being told (https://lnkd.in/gUaGKuAE) that this benchmark data may have been exclusively available (https://lnkd.in/g5E3tcse) to OpenAI since before o1āand that the benchmark creators were not allowed to disclose this *until after o3 *.
That o3 does well on frontier math held-out set is impressive, no doubt, but the mental picture of āš1/š3 šššš šššš ššššš ššššššš šš šššššš šššš, ššš šššš šššššššššššš šššššššššš šš šššššššš ššššāāthat the AGI tomorrow crowd seem to haveāthat šš±š¦šÆšš šøš©šŖšš¦ šÆš°šµ š¦š¹š±ššŖš¤šŖšµššŗ š¤šš¢šŖš®šŖšÆšØ, š¤š¦š³šµš¢šŖšÆššŗ š„šŖš„šÆāšµ š„šŖš³š¦š¤šµššŗ š¤š°šÆšµš³š¢š„šŖš¤šµāis shattered by this. (I have, in fact, been grumbling to my students since o3 announcement that I donāt completely believe that OpenAI didnāt have access to the Olympiad/Frontier Math data before handā¦ )
I do think o1/o3 are impressive technical achievements (see https://lnkd.in/gvVqmTG9 )
š«šššš šššš šš šššš šššššššššš šššš ššš ššš ššššš šššššš šš šš ššššš ššššššššššāššš š ššššāš ššššš šššššš āšØš®š° š»ššššššš.ā
We all know that data contamination is an issue with LLMs and LRMs. We also know that reasoning claims need more careful vetting than āšøš¦ š„šŖš„šÆāšµ š“š¦š¦ šµš©š¢šµ š“š±š¦š¤šŖš§šŖš¤ š±š³š°š£šš¦š® šŖšÆš“šµš¢šÆš¤š¦ š„š¶š³šŖšÆšØ šµš³š¢šŖšÆšŖšÆšØā (see āIn vs. Out of Distribution analyses are not that useful for understanding LLM reasoning capabilitiesā https://lnkd.in/gZ2wBM_F ).
At the very least, this episode further argues for increased vigilance/skepticism on the part of AI research community in how they parse the benchmark claims put out commercial entities."
Big stupid snake oil strikes again.
Every time they go āthis wasnt in the dataā it turns out it was. A while back they did the same with translating rareish languages. Turns out it was trained on it. Fucked up. But also, wtf how are they expecting this to stay secret and there being no backlash? This world needs a better class of criminals.
But also, wtf how are they expecting this to stay secret and there being no backlash?
No, they bet on it not mattering and theyāve been completely right thus far.
itās enough if it ends up not mattering long enough for them to cash out, then they donāt care
Ah right yes.
The conspiracy theorist who lives in my brain wants to say its intentional to make us more open to blatant cheating as something thatās just a ācost of doing business.ā (I swear I saw this phrase a half dozen times in the orange site thread about this)
The earnest part of me tells me no, these guys are just clowns, but I dunno, they canāt all be this dumb right?
holy shit, thatās the excuse theyāre going for? they cheated on a benchmark so hard the results are totally meaningless, sold their most expensive new models yet on the back of that cheated benchmark, further eroded the scientific process both with their cheating and by selling those models as better for scientific researchā¦ and these weird fucks want that to be fine and normal? fuck them
Yeah we would like to stop lying and cheating, but the number you see.
they canāt even sell o3 really - in o3 high mode, needed to do this level of query, itās about $1000 per query lol
do you figure itās $1000/query because the algorithms they wrote with their insider knowledge to cheat the benchmark are very expensive to run, or is it $1000/query because theyāre grifters and all high mode does is use the model trained on frontiermath and allocate more resources to the query? and like any good grifter, theyāre targeting whales and institutional marks who are so invested that throwing away $1000 on horseshit feels like a bargain
so, for an extremely unscientific demonstration, here (warning: AWS may try hard to get you to engage with Explainer[0]) is an instance of an aws pricing estimate for big handwave āsome gpu computeā
and when I say āextremely unscientificā, I mean āI largely pulled the numbers out of my assā. even so, theyāre not entirely baseless, nor just picking absolute maxvals and laughing
parametersassumptions made:- āsomewhat beefyā gpu instances (g4dn.4xlarge, selected through the tried and tested āsquint until it looks rightā method)
- 6-day traffic pattern, excluding sunday[1]
- daily ā4h peakā total peak load profile[2]
- 50 instances mininum, 150 maximum (letās pretend weāre not openai but are instead some random fuckwit flybynight modelfuckery startup)
- us west coast
- spot instances, convertible spot reserves, 3y full prepay commit (yeah I know full vs partial is a big diff; once again, snore)
(and before we get any fucking ruleslawyering dumb motherfuckers rolling in here about accuracy or whatever: get fucked kthx. this is just a very loosely demonstrative example)
so youād have a variable buffer of 50ā¦150 instances, featuring 3.2ā¦9.6TiB of RAM for working set size, 800ā¦2400 vCPU, 50ā¦150 nvidia t4 cores, and 800ā¦2400GiB gpu vram
letās presume a perfectly spherical ops team of uniform capability[3] and imagine that we have some lovely and capable active instance prewarming and correct host caching and whatnot. yāknow, things to reduce user latency. letās pretend weāre fully dynamic[4]
so, by the numbers, then
1y times 4h daily gives us 1460h (in seconds, thatās 5256000). this extremely inaccurate full-of-presumptions number gives us āservice-capable life timeā. the times your concierge is at the desk, the times you can get pizza delivered.
x3 to get to lifetime matching our spot commit, x50ā¦x150 to get to ātotal possible instance hoursā. which is the top end of our sunshine and rainbows pretend compute budget. which, of course, we still have exactly no idea how to spend. because we donāt know the real cost of servicing a query!
but letās work backwards from some made-up shit, using numbers The Poor Public gets (vs numbers Free Microsoft Credits will imbue unto you), and see where we end up!
so that means our baseline:
- upfront cost: $4,527,400.00
- monthly: $1460.00 (x3 x12 = $52560)
- whatever the hell else is incurred (s3, bandwidth, ā¦)
=200k/yper ops/whatever person we have
3y of 4h-daily at 50 instances = 788400000 seconds. at 150 instances, 2365200000 seconds.
so we can say that, for our deeply Whiffs Ever So Slightly values, a secondās compute on the low instance-count end is $0.01722755 and $0.00574252 at the higher instance-count end! which gives us a bit of a handle!
this, of course, entirely ignores parallelism, n-instance job/load/whatever distribution, database lookups, network traffic, allllllll kinds of shit. which we canāt really have good information on without some insider infrastructure leaks anyway. if we pretend to look at the compute alone.
so what does $1000/query mean, in the sense of our very ridiculous and fantastical numbers? since the units are now The Same, we can simply divide things!
at the 50 instance mark, weād need to hypothetically spend 174139.68 instance-seconds. thatās 2.0154 days of linear compute!
at the 150 instance mark, 522419.05 instance-seconds! 6.070 days of linear compute!
so! what have we learned? well, weāve learned that we couldnāt deliver responses to prompts in Reasonable Time at these hardware presumptions! which, again, are linear presumptions. and thereās gonna be a fair chunk of parallelism and other parts involved here. but even so, turns out itād be a bit of a sizable chunk of compute allocated. to even a single prompt response.
[0] - a product/service whose very existence I find hilarious; the entire suite of aws products is designed to extract as much money from every possible function whatsoever, leading to complexity, which they then respond to byā¦ producing a chatbot to āguide usersā
[1] - yes yes I know, the world is not uniform and the fucking promptfans come from everywhere. Iām presuming amerocentric design thinking (which imo is probably not wrong)
[2] - letās pretend that the calculatorsā presumption of 4h persistent peak load and our presumption of short-duration load approaching 4h cumulative are the same
[3] - oh, who am I kidding, you know itās gonna be some dumb motherfuckers with ansible and k8s and terraform and chucklefuckery
when digging around I happened to find this thread which has some benchmarks for a diff model
itās apples to square fenceposts, of course, since one llm is not another. but it gives something to presume from. if g4dn.2xl gave them 214 tok/s, and if we make the extremely generous presumption that tok==word (which, well, no; cf.
strawberry), then any Use Deserving Of o3 (letās say 5~15k words) would mean you need a tok-rate of 1000~3000 tok/s for a āreasonableā response latency (ā5-ish secondsā)so youād need something like 5x g4dn.2xl just to shit out 5000 words with dolphin-llama3 in āquickā time. which, again, isnāt even whatever the fuck people are doing with openaiās garbage.
utter, complete, comprehensive clownery. era-redefining clownery.
but some dumb motherfucker in a bar will keep telling me itās the future. and I get to not boop 'em on the nose. le sigh.
They understand that all of the major model providers is doing it, but since the major model providers are richer than they are, they canāt possibly ask OpenAI and friends to stop, so in their heads, it is what it is and therefore must be allowed to continue.
Or at least, thatās my face value read of it, I certainly hope Iām simplifying things too much.
also they are rationalists and hence the most gullible mfs on any of this stuff
Maybe this is common knowledge, but I had no idea before. What an absolutely horrible decision from google to allow this. What are they thinking?? This is great for phishing and malware, but I donāt know what else. (Yeah ok, the reason has probably something to do with āline must go upā.)
I recall seeing something of this sort happening on goog for about 12~18mo - every so often a researcher post does the rounds where someone finds Yet Another way goog is fucking it up
the advertising dept has completely captured all mindshare and it is (demonstrably) the only part that goog-the-business cares about
Good news, everyone: critihype is canceled until the next tweet.
Hat tip to this reply showcasing a beautiful example of image gen. Very Ć©thiccal(athiccal?) technology.
Amazing how that all looks like one of those sites from the geocities era, with sparkles/stars butterflies and unicorns and dolphins all over it. All it needs now is a under construction sign.
it must be ethiccal, it has two dolphins
And a Quarian who went to business school
Imagine going on the Pilgrimage and all you bring back is an MBA and some motivational quotes, instant exile
āZeroā, so true
5ERO, even!
Hmm, surely there is no downside to doing all of oneās marketing, both personal* and professional, through the false certainty and low signal of short-form social media. The leopard has only licked Samās face, it will never bite and begin chewing!
*You and I may find the concept of a āpersonal brandā to be horrifying, but these guys clearly want to become brands more fervently than Bruce Wayne wanted to become a bat
