Mastodon Feed: Posts

ireneista@irenes.space ("Irenes (many)") wrote:

(after that it'll be time to make it work on proper Forth programs)

ireneista@irenes.space ("Irenes (many)") wrote:

now we just need to add a couple other types of literal, and one more comment formatting feature, and that may be all we need to do for running the hex transform on Evocation-assembly programs

ireneista@irenes.space ("Irenes (many)") wrote:

yessssssss we have Evocation's hex transform including integer literals for values measured at the compiler's "runtime" in appropriate places in the disassembly comments

https://code.irenes.space/evocation/commit/?id=e03216ba8bbc7547b1451ed99497c0693446983b

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

let's do some more back of the envelope estimates.

no company publishes the energy usage of their models, so as before we just have to do the best we can by making conservative extrapolations.

altman claimed in 2025 that ChatGPT uses 0.34Wh per prompt (this is lower than independent estimates, but let's go with it to be conservative).

Previously i had made the very conservative estimate that considering the "plain info" and retail use of the product bringing it down, the average prompt might be something like 3 paragraphs, or 300 tokens. Later models that use reasoning tokens use orders of magnitude more tokens (prompting claude code to estimate the average number of output tokens without writing code just now used ~2k). At the time of that blog post, the newest general purpose model would have been GPT-4.5, but the consumer model that would have made up the bulk of the average for the power estimate would have been the infamous GPT-4o, a ~200B parameter model - neither of these are "reasoning" models, so lets keep our 300 token estimate for pre-"reasoning" models.

So that would be 0.34 / 300 = 0.00113 Wh/token.

According to ~ rumors and speculation ~, people believe that GPT-6 astra is a ~5T model, and that this internal model might be ~10T parameters. Parameter count is a very imprecise estimate of energy usage, because energy usage depends on a ton of things like hardware efficiency, the attention mechanism selecting how many of the params are active, and so on. But because we don't have any better information, we assume that energy usage is quasi-linear with number of parameters.

So then that would be (10 trillion / 200 billion) * (0.34 Wh / 300 tokens) = 0.0566Wh/token for the giant 10T internal model.

Then multiplying that through for 300 billion tokens yields us 17 GWh. That's roughly a third of LA county's daily residential energy use, a county of 10 million people.

Boosted by neatnik@social.lol ("Neatnik :prami:"):
numist@xoxo.zone ("Scott Perry") wrote:

RE: https://social.lol/@neatnik/117227874076649666

"I have almost reached the regrettable conclusion that the Negro's great stumbling block in his stride toward freedom is not the White Citizen's Counciler or the Ku Klux Klanner, but the white moderate, who is more devoted to "order" than to justice; who prefers a negative peace which is the absence of tension to a positive peace which is the presence of justice; who constantly says: "I agree with you in the goal you seek, but I cannot agree with your methods of direct action"; who …"

Boosted by soatok@furry.engineer ("Soatok Dreamseeker"):
Intaglio_Dragon@furry.engineer ("Intaglio Whitegraven") wrote:

@MeiLin @soatok I'm reminded of that "Mom, can you give me money" meme:

• Boss, can you give me money to go to Berlin?
• To attend the IT conference?
• Yes, to attend the IT conference.
• Actually attends Eurofurence like a boss 😎

Boosted by soatok@furry.engineer ("Soatok Dreamseeker"):
Gyroplast@furry.engineer wrote:

@soatok
To paraphrase Loriot:

A life without being a pug is possible, but pointless.

You likely have succumbed to furry mindrot. You might experience other symptoms:

• excessive tolerance and respect towards all creatures
• sudden urges to hug, or be hugged
• phantom tail
• being a good boy
• awoos

Do not consult anyone or seek medical help. The condition is irreversible, but spreads only through a mildly contagious pathOwOgen through prolonged and close exposure to infected groups.

Boosted by glyph ("Glyph"):
aphyr@woof.group ("École des Bro-Arts") wrote:

My new drag king persona is an orchestral percussionist with an anxiety disorder: Tim Panic.

Boosted by glyph ("Glyph"):
rmondello@hachyderm.io ("Ricky Mondello") wrote:

Here’s the blog post I promised y’all on how I recommend switching password managers in 2026! It covers:
- why your iPhone or iPad, and not your Mac, might be the place to start your switch
- when to make the switch and how to treat your “old” app going forward
- tips and tricks for switching to Apple Passwords, if that’s where you’re going (can be skipped!)
- the recent controversy in password management and some thoughts on values

https://rmondello.com/2026/09/07/switching-password-managers-2026/

neatnik@social.lol ("Neatnik :prami:") wrote:

For anyone seeking an unbiased account of what this is about, I highly recommend this piece: https://coyotetracks.org/blog/pressure-to-stay-genteel/

To this day, Manton still hasn't denounced the transphobic statement made by his employee, or the barrage of harassing emails that his employee sent to an omg.lol member who was offended by that statement. He clearly knew that it was a bad look, though, since he removed the employee's name from the Micro.blog website. He also still has not made a clear and unequivocal statement of support (or even acceptance) of queer people. The closest he's ever come was "Micro.blog is a good place for LGBTQ+ people."

He probably thinks this post is a "personal attack", because any time a privileged white man's integrity is questioned, they feel attacked. But this post is just me sharing facts. They may be inconvenient for Manton, but they're 100% true.

Boosted by glyph ("Glyph"):
fionafibration@mathstodon.xyz ("barbie's dream house of leaves") wrote:

RE: https://mathstodon.xyz/@tao/117230983938174038

I think it's extremely damning that even Tao, who is very involved with OpenAI, recognises this for what it is: a false promise, a contribution to mathematics which leads to dead-ends in our theories and understanding, a company using unsolved conjectures as a resource to fuel their stock prices while providing little to no value back to the field of mathematics.

I hope that in the next couple years we as mathematicians seriously reconsider our current system where theorems proven/disproven are the primary social currency of mathematics, because these LLMs are hijacking these incentives in ways that are very damaging to us.

As an early-career researcher it's hard to not want to jump ship right now. I started doing mathematics very young and I have never loved something like I love maths, but all of the recent news makes me wonder whether it's actually a good idea to continue doing math.

Almost every single PhD student I have talked to is doubting whether they have a future with the current situation, and I hope we can figure things out before the current generation of mathematicians PhDs arrive into a world of maths that has been mined out and left barren by AI companies pumping more speculative value into their bubble.

(1/3)

glyph ("Glyph") wrote:

sometimes you find a person with a blog who is _absolutely_ just living their best life and working on their special interest and it is a beautiful thing https://www.benjoffe.com/fast-date

Boosted by glyph ("Glyph"):
Ussjoin@cloudisland.nz ("Brendan Francis O'Connor") wrote:

@glyph Do you know anyplace to post jobs specifically of the form “we are *not* demanding you use LLMs?” My company is trying to find a new Infra lead and all the candidates are… not actually able to do the work, many because they appear to be mostly ChatGPT by volume. It just occurred to me you might have a suggestion.

neatnik@social.lol ("Neatnik :prami:") wrote:

Earlier today, the chief protocol guy at Bluesky wrote something nice about omg.lol. https://bsky.app/profile/bnewbold.net/post/3muzgzcgzyc2l

Then Manton Reece showed up to complain about "the owner's long-running personal attacks on me." I guess he couldn't pass up the opportunity to take a swing.

I spun up a Bluesky account to respond, and swapped a few replies with Manton. He responded to everything I said with "that didn't happen" or "that's a total distortion of the truth".

You can't have a productive conversation with someone who refuses to acknowledge real harm. And you definitely can't have a productive conversation with someone who chooses gaslighting over earnest discussion.

It was a waste of time and energy. I deleted my replies and my Bluesky account. If Manton ever wants to have a real conversation about this, I would welcome that. If he just wants to endlessly whine about my past requests to see him denounce bigotry in all of its forms (especially bigotry by another member of the Micro.blog team), which is what he's labeling "personal attacks", then that's his choice. If he wants to continue to play the victim instead of demonstrate accountability, that's his prerogative.

Boosted by cstanhope@social.coop ("Your weary 'net denizen"):
wizard@xyzzy.link ("[XYZ] wlo") wrote:

violating the wtfpl by not doing whatever the fuck i want

Boosted by pixxl:
VileLasagna@mastodon.gamedev.place ("Vile Lasagna") wrote:

@tante it screams against the insecurity of sloplickers that a lot of them take great offence at others' refusal to engage. They know about a myriad of ways in which these things are, uuuummm, extremely bad, actually. An important function of the narrative of inevitability is that it removes agency from themselves: it's an overwhelming force that happens to them, it's silly to even imagine not engaging.

But when you do, the illusion breaks and they take that as a personal, moral attack

Boosted by pixxl:
tante@tldr.nettime.org wrote:

RE: https://tldr.nettime.org/@tante/117235476116975345

The most surprising realization I had in the last years was how important it is to be public about your refusal to participate in harmful acts therefore giving others "permission" to do so as well. It really matters.

jscalzi@threads.net ("John Scalzi") wrote:

Happy birthday to objectively one of our best, and subjectively one of my favorite, songwriters: @realaimeemann!

ireneista@irenes.space ("Irenes (many)") wrote:

dunno. go read our other thread. it's better :)

ireneista@irenes.space ("Irenes (many)") wrote:

sigh are we being too harsh? the amount of protection it gives isn't zero, and it's true that evil people and organizations are trying it. we still think the style of thinking is worth interrogating though.

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

When companies that throw around billions of dollars like its nothing claim they have done something no scientist or mathematician ever could, remember that NIH's total annual budget is $48b, NSF's is $8.75b ..... And OpenAI had $34b of costs and expenses in 2025, including a $19b R&D budget.

NIH and NSF pay for most basic research in the US across all disciplines. OpenAI makes consumer language models and language model accessories. If any single problem had a $19b R&D budget like LLMs do, no fucking shit progress would go up like a hockeystick

Boosted by neatnik@social.lol ("Neatnik :prami:"):
jasonm@social.lol ("Jason :neobread_this_is_fine:") wrote:

white guys with bad takes

ireneista@irenes.space ("Irenes (many)") wrote:

bull fucking shit.

ireneista@irenes.space ("Irenes (many)") wrote:

it doesn't matter how absurd the claim is

it doesn't matter that everything you care about would already be gone before the claim even becomes relevant

somebody else might do it first, if we don't, so we have no choice! it's not our fault, it's game theory!

ireneista@irenes.space ("Irenes (many)") wrote:

"we cannot allow there to be a mine shaft gap!"

we kind of think Failsafe is a better movie than Dr. Strangelove, but this particular line just does a very good job of illustrating a type of thinking that's endemic to statists and anyone selling the cop/military version of "security"

in particular, it's the same thinking behind claims that it's necessary to use LLMs to find security issues

Boosted by glyph ("Glyph"):
glyph ("Glyph") wrote:

Is using a local LLM any worse than playing a video game? No clickbait here, the answer is "yes, much worse", but perhaps you will be interested in an explanation of why: https://blog.glyph.im/2026/09/but-what-about-video-games.html

soatok@furry.engineer ("Soatok Dreamseeker") wrote:

Today I was reminded that it's possible to, when imagining an avatar for oneself in a fictional setting, not assume a creature.

Maybe I've been a furry for so long that it's normal?

Looks friend-shaped. Why not friend?

jonny@neuromatch.social ("jonny (nonvenomous)") wrote:

OpenAI claims that they used 300 billion output tokens from a swarm of ~10,000 concurrent instances of an internal model that is almost certainly larger and more expensive to run than GPT-6. That's the total token output, including exploring related problems, which, along with spying on other researchers work via codex, "led them to the solution," so that is the minimal accounting of what led to the solution.

All of OpenAI's API costs are presumed to be subsidized, lower than the actual cost of running the models. If one were to pay the API costs for 300 billion tokens for GPT-6, at $75 per million output tokens with long context, that would be 22.5 million dollars of tokens (the minimum, assuming 100% cached output with short context is still 3.75 million, but that's an unrealistic minimum). OpenAI apparently also had a full team of people working on this, so 22.5 million doesn't include their salaries, etc.

Tristan Buckmaster only has one grant listed from NSF, a CAREER grant for $450k from 2022 through 2028, averaging $70k per year the last three years.

If i search the NSF grants database for any award from the Division of Mathematical Sciences with "navier-stokes" in the abstract (surely, imprecise, and most likely a sizeable overcount because this is "all work that mentions the equation" not "work spent specifically on the millenium prize proof"), then i get $135m worth of awards since 1978, including large training grants for centers, etc. so $22.5m would be ~16% (or, roughly, 1/6th) of all funding that NSF has allocated to research that mentions the problem in their grants for almost 50 years.

If i then use NSF's API to find all the publications that have come from these grants, I find 4,928 papers. I can match 4264 to entries in OpenAlex (85%), yielding 142,606 citations (1/6 of which is 710, and 23,767, respectively).

So, assuming we believe OpenAI's accounting, the headline is basically "OpenAI researchers seem to have stolen the work of mathematicians that were using Codex, and then spent 1/6 of all the funding NSF has given with abstracts mentioning navier-stokes over the past 50 years, which amounts to an entire subdiscipline of work, feeding, housing, and training generations of mathematicians, in order to generate a proof two of the statements in a millenium prize problem"

I dont know how NSF funds math, but i sort of doubt they give grants for "try to solve millenium problems full stop." idk, maybe an alternative strategy would be to just "fund basic research," because even in this specific domain of an axiomatic universe with an unambiguously evaluable solution to a problem where brute force is possible, "AI" doesn't seem like it's really that much of a bargain, and "paying people to be experts at things" has lots of known good side effects like "someone actually understands the solution"

Boosted by glyph ("Glyph"):
xgranade@wandering.shop ("Cassandra has had an adventure") wrote:

Reader, I literally went back to bed. Then I woke up, cuddled the doggo, drank some coffee, and watched the Zelda direct. GrapheneOS still is slopware, the CTO of Bluesky (Old Twitter, now for TERFs and boosters!) is still saying inane shit about stats, there's still slop in the Linux kernel and a distro that's upstream from like two thirds of other distros, there's still an international movement to exterminate people like me, and more.

But it did, dear Reader, help quite a bit.

Boosted by ireneista@irenes.space ("Irenes (many)"):
zanzi@mathstodon.xyz ("Zanzi @ Monoidal Cafe") wrote:

look, all i'm saying is that mathematics is the art of turning coffee into theorems, and no machine will ever drink coffee