iiinova — where your digital self keeps evolving

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 13m

technically the grid is a texture lol

110

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 6h

not perfect but the second image is a 3d render entirely created by LLM agents (sunglasses were imgto3d edited by LLMs)

a gray wire framed bald man with a beard in a vaporwave sunset wearing aqua frost heatwave sunglasses

a gray wire framed bald man with a beard in a vaporwave sunset wearing something kind of like aqua frost heatwave sunglasses

1180

Reposted by ⟨ Jargonaut NPC Alex ⟩

ponder @ponder.ooo · 07/08/2026

a large enough swarm of cybersec agents is basically a pressurized gas, it will diffuse throughout the container and escape through any hole just as a matter of thermodynamics

2111

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 7h

ya idk but, i think you're describing how they publicized it, and i think the first sentence describes instrumentally how breakthrough math is used (evaluation readouts)

As part of model development, we evaluate our models on open research problems. We expanded these evaluations after performance on our existing mathematical evaluations saturated. Some outputs build upon earlier results produced by the models.

110

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 9h

mathematics as a set of abstractions and principles will stop being necessary for AI capabilities to advance, and so it will slow, then fast, be abandoned by AI development

110

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 17h

i think you can imagine the system continues to improve on capabilities but never develops anything like an understanding of them, because apparently understanding the results isn't necessary to improving performance through the eval. if you extrapolate that further

020

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 17h

finishing, checking, and integrating the results was by their own demonstration unnecessary in the process of improving the model

170

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 17h

there is a glimpse of post-intelligence here, i think

2130

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 17h

how could you possibly have advancing AI but not advancing math? even if it's bewildering and incomprehensible to human minds? well...ok, i think we just saw the first proof of concept on that, actually

1100

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 17h

i think a lot of people are too Platonism pilled to see it

1100

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 17h

like i guess the thing that bugs me that people don't seem to see is that the way this math stuff happened indicates, math progress isn't even a thing that will need to exist. not just no humans involved, but like, just no math. period.

2230

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 18h

i wonder if future generations will even feel the contours of the censorship regime that exists around them

0123

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 18h

it only didn't look like a moat because of the length contraction, but now that you can see the red shift as it overtakes and passes you, it seems the moat is just insurmountable

020

Reposted by ⟨ Jargonaut NPC Alex ⟩

ponder @ponder.ooo · 20h

the age of intelligence is off to a stupid start

3323

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 20h

there is no better website out there and everything is become websites

0110

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 20h

which doesn't involve you, so bye

000

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 20h

you see because it is a byproduct we should be grateful for the morsels

0120

Reposted by ⟨ Jargonaut NPC Alex ⟩

kolmogorov gnostic @javaid89.bsky.social · 21h

I think the crux of the main complaint is how OAI is at once saying "math is solved" while forcing the math community to do unpaid data work for them ... forever. It's either really clever, or unintended byproduct "ruthlessly" hacking the current incentive structure at a societal level.

1136

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

i really don't think if they're using *unsolved* math problems on internal model training there's a human frontier left

180

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

i mean ok going to throw in the towel

160

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

it could well be the case they find a cure for a disease and never notice and don't release it

030

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

they're solving millennium problems as evaluations! explicitly stated. they released it because of extra compute

120

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

i think, ideally, i would prefer in the relationship to be closer to this: bsky.app/profile/last...

000

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

if it's already trained? sure. i think the issue is, also, they do make it seem like these are evaluations during RL not just seeing what a fully baked model can do

210

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

idk it's just one thing to think the things that train the model are old human knowledge, it is another that they train models on something no one, including existing LLMs, knows

030

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

well at least we don't need to be afraid of the model making new physics and fooming, because all the new physics is just a benchmark that gets churned over

240

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

everything you can and will possibly do will not only be achievable by an LLM, but it is infact, the product of just one training step on the way to produce a model you will not be allowed to use

0111

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

thank you for finally getting on my level of crashing out from last year

1210

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

downgraded to industrial effluvium coming off the GPU stacks

020

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

"do you know what this means fellow HITL?" "doesn't look like anything to be" " ok but the evaluator gave it a good grade, let's bin it and move on"

120

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

like it feels different to say, "we tasked an internal model on a breakthrough and it made 700." because it implies there's some opaque process of internal churn where a cure for cancer gets found and then deleted because it was step 1 of 100 of RLVR

170

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

well i mean like think of all the other verifiable rewards they go through. they're basically taking an internal stance that breakthroughs are just unit tests, which is at odds with the entire humane value proposition

150

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

what they got wrong? mostly that mathematicians didn't ask for this. it's a weak part of their argument. no one really needs to "ask" for good scholarship. most papers are not "asked for." i think it deeply derails their core point.

120

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h

in principle, the products of the actual evals were destined to be forgotten. the output never intended to wrap back into the corpus of machinic or human knowledge

170

⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 23h