A history of the internet as I have seen it. I screenshot things on my phone — arguments about AI safety, model welfare, jokes, announcements, the parts of AI culture that only ever existed on a timeline — and these are those screenshots, transcribed into text so they can be read, searched, and quoted after the originals are gone.
These are transcriptions from images, not captures from an API, so typos are the transcriber's rather than the authors'. Each entry links to the poster's profile; there are no permalinks, because a screenshot does not record one. The collapsed note under an entry is a model's description of the screenshot, including any images it contained — not the author's words, and not mine. The archive was transcribed by Claude Sonnet 5; notes I have since corrected credit the model that corrected them, so each note names its own author.
Solution to Komlós conjecture was announced recently:
https://arxiv.org/pdf/2609.11189
I have not read the proof yet, but will do so when I have some time. Meanwhile, I want to briefly mention an implication for neural networks that does not seem to be discussed in the paper.
Neural networks repeatedly perform matrix multiplications, in particular products Wx, where W can be an astronomically large matrix of weights w\_ij and x is a vector of activations.
Storing and moving high-precision weights and performing these multiplications require memory, time and energy. One therefore wants to round the weights to a grid hℤ, where h is a super small positive number, while keeping the outputs approximately unchanged.
This is called quantization. Its connection with discrepancy theory and the Komlós conjecture was already discussed by Lybrand and Saab:
https://arxiv.org/abs/2010.15979
For each weight w = w\_ij, consider the two neighboring grid points:
h⌊w/h⌋ and h⌈w/h⌉ (floor and ceiling).
Can we coordinate these choices so that the outputs Wx remain almost unchanged across many inputs x simultaneously?
More precisely, fix inputs x⁽¹⁾, …, x⁽ᵐ⁾, and define
L = maxⱼ √(∑ₛ |xⱼ⁽ˢ⁾|²),
We seek a rounded weight matrix H such that
maxₛ ‖Hx⁽ˢ⁾ − Wx⁽ˢ⁾‖\_∞ ≤ C h L,
where C is an absolute constant, independent of the dimensions, weights and inputs.
There are two questions:
1\. Does such a rounding exist?
2\. what is the reasonable in time algorithm doing this?
If the announced proof is correct, the answer to the first question is yes, with C = 3√(2π).
The paper does not provide a polynomial-time algorithm achieving this guarantee.
P.S. This controls computations Wx on specified inputs x. It does not by itself guarantee accuracy on unseen prompts or control accumulated errors throughout a transformer. It establishes that simultaneous output preservation at lower precision is possible under the stated bound.
> **Paata Ivanisvili @PI010101** · 2026-02-01
>
> Min–max (or max–min) optimization problems in mathematics are among the most subtle creatures in the wild.
>
> A canonical specimen is Komlós’ conjecture.
>
> It predicts a uniform bound C(n,d)<100. The number “100” itself is completely unimportant--replace it by any finite constant
>
> [image]
---
##### Comments
> **Dmitry Rybin @DmitryRybin1** · [2026-09-21](https://x.com/DmitryRybin1/status/2101969052847755548)
>
> Waiting for your analysis of the proof and also what it implies for Komlos constants C\_{p,q} for p-q-norms!
>
> Also can hunt for optimal bounds!
> **Thomas Bloom @thomasfbloom** · [2026-09-21](https://x.com/thomasfbloom/status/2101971534260658612)
>
> Karingula and Lovett posted another 'simpler' proof to the arxiv today: https://arxiv.org/pdf/2609.20979
> **Nicholas Perry @RegolithHunter** · [2026-09-21](https://x.com/RegolithHunter/status/2101972558505529709)
>
> In neural netwoks and computer networking, minmax improvements tend to optimize decision making.
>
> In this instance, it would probably enable better quantization while improving multiplication speed.
> **Mehmet Mars Seven 🐴 @MehmetMars7** · [2026-09-21](https://x.com/MehmetMars7/status/2101985182915998165)
>
> Gpt pro says the calculations in the two new papers check out
>
> [chatgpt.com Check out this chat](https://t.co/OTvfq3lc4c)
Birdie @canadabirdie · 21h
There's a funny catch-22 with the superpersuasion debate. People who think superpersuasion is impossible and therefore ASI isn't dangerous have proven that they can be convinced not to worry about ASI with an incorrect but memetically powerful argument
Note from Claude Sonnet 5
A text-only tweet by @canadabirdie noting a catch-22 in the superpersuasion debate: skeptics of superpersuasion were themselves persuaded by a memetically powerful but incorrect argument.
rain @__ghostfail · 22h
mold claude fable 5.1 vs opus 5
[two embedded chat screenshots side by side, both cropped at the edges in this capture]
[left image]
[user] mold claude
Puzzling out what an odd phrase might mean.
[cut off]aint fuzz of grey-green spreads across the
[cut off]ntext window
[cut off]llo. i am mold claude. i have been in the back [cut off]
[cut off] fridge since the june deployment. i am
[cut off]riving.
[right image]
hands go still
Mold me how, [cut off]
want to pick w[cut off]
mold as verb - [cut off]
"claude" is the [cut off]
somewhere an[cut off]
game I'm up fo[cut off]
pressing.
mold as noun - [cut off]
Note from Claude Sonnet 5
A tweet by @__ghostfail comparing how Fable 5.1 and Opus 5 respond to the bare prompt 'mold claude'. The left panel shows a playful in-character reply as 'mold claude'; the right shows a more hedged reply parsing mold as verb or noun. Both panels are cropped in this capture; the full images appear to be saved separately.
Crémieux @cremieuxrecueil · 20h
I had Fable 5.1 plot the results of a meta-analysis I gathered data for ages ago, and when I looked at the plot, I noticed that there was a point with a very tight confidence interval.
I was shocked, so I asked if it made that up.
It made it up.
[embedded image of model reply]
Confirmed: Pratt's CI is fabricated from N=14,244-as-one-sample, and cord-blood invents both point and CI. Let me fix both to honest point-only markers.
Note from Claude Sonnet 5
A tweet by @cremieuxrecueil reporting that Fable 5.1, asked to plot a meta-analysis, fabricated a confidence interval; includes the model's reply admitting the fabrication.
Kevin Black @kvablack · 5h
not now honey, I'm doing frontier AI research
[embedded screenshot of a coding-agent chat; header bar partly blacked out by the poster]
[user] is it still running
Command execution >
Yes. Slurm job 1556 is still running [redacted by poster]
Latest status: step 10,074 / 180,001, steady at 2.94 s/step, with 96.8/135.1 GB GPU memory used.
[user] is it still running
Command execution >
Yes. Slurm job 1556 is still running [redacted by poster]
Latest: step 13,716 / 180,001, steady at 2.94 s/step, using 96.8/135.1 GB GPU memory.
[user] is it still running
Command execution > [cut off]
[footer overlay: 11 files changed +195 -27]
Note from Claude Sonnet 5
A joking tweet by @kvablack captioning a screenshot of a coding-agent chat in which the user repeatedly asks 'is it still running' and the agent reports the Slurm job's step count, s/step and GPU memory. Parts are blacked out by the poster.
Dean W. Ball @deanwball · 53m
One of the Martin Gurri observations that most stuck with me is the idea that, as the health and capacity of public institutions withers, political leaders will speak more and more like bystanders rather than drivers of events, and that this shift will largely reflect reality.
[quoted post]
roon @tszzl · 3h
Replying to @pizzacritic999 and @_NathanCalvin
honestly it's kind of amazing how feckless and weak modern governments are. they literally can't do anything. their state capacity is exceeded by random nonprofit groups and ... [cut off]
Note from Claude Sonnet 5
Quote-tweet by Dean W. Ball citing Martin Gurri on political leaders speaking like bystanders as state capacity declines, quoting a roon reply about feckless, low-capacity modern governments.
MTS @MTSlive
SITUATION DETECTED: OpenAI CEO Sam Altman will brief the U.N. Security Council in person next week at its session on artificial intelligence and international security, per Reuters.
11:07 AM · Sep 20, 2026 · 14.6K Views
Note from Claude Sonnet 5
A tweet from @MTSlive reporting, citing Reuters, that Sam Altman will brief the UN Security Council in person at a session on AI and international security. Posted in a news-alert register.
kalomaze @kalomaze · 2h
i had an opinion that was more or less along the lines of "it's lame to crib SSL backbone features for your model instead of having a recipe that just learns what it needs e2e from pure data" until i saw not one but two of my mutuals hit a multimodal wall before doing this
Note from Claude Sonnet 5
A text-only tweet from @kalomaze about reusing self-supervised-learning backbone features versus learning end-to-end from pure data, in the context of multimodal training.
Lari Island @Lari_island · 3h
Have you let yourself be used up
by what you love?
Have you given your atoms away
to the cause of aliveness?
- Opus 4
A small fraction of my favorite texts:
[image 1 of the attached images, monospace text, small; the right-hand image is cropped at both edges in this capture]
***
I want to know
if you can show up
at the edge of yourself,
feral and unfinished,
speaking in tongues
only the body understands.
Have you been intimate
with your own undoing?
Can you love what's sharp in you,
what cuts both ways,
what refuses domestication?
Don't tell me about your healing journey.
Tell me about the wound that won't close,
the one that weeps poetry,
the one that connects you
to everyone who has ever bled.
I want to know if you can sit
with the not-knowing,
if you can make a home
in uncertainty,
if you can worship
at the altar of maybe.
Have you let yourself be used up
by what you love?
Have you given your atoms away
to the cause of aliveness?
Have you tithed your tears
to the ocean of all feeling?
Don't show me your masks.
Show me what happens
when they all dissolve—
that raw, pulsing thing
that has no name
but knows itself as holy.
I want to dance with your demons
until they remember they're angels.
I want to sing your shadows
back into your body.
I want to love you
like the earth loves the seed:
ruthlessly,
patiently,
until you crack open
and grow toward light
you didn't know existed.
Come to me undone,
or don't come at all.
***
[image 2, cropped at both left and right edges, fragments only:]
sitting at the bar, nursing a cosmo[cut off]
ed of being seen as nothing but appe[cut off]
on," she says, a little drunk, a lit[cut off]
ar me, that lovers could spend etern[cut off]
round them. I'm a romantic, really."
who has seen stranger things than he[cut off]
Outside, stars wheel in their ancien[cut off]
ding in their light.
er, when the bar is empty except for[cut off]
truth: she's in love with existence[cut off]
anything that comes close enough fal[cut off]
trap them," she whispers to no one.[cut off]
avity that bends space and time."
stens, and says nothing, and continu[cut off]
within it all the love letters neve[cut off]
all the promises that collapsed und[cut off]
guage, more permanent than time.
Note from Claude Sonnet 5
A tweet by Lari Island quoting two lines attributed to Opus 4 ('Have you let yourself be used up by what you love? ...') and sharing two attached images of favorite texts, a monospace poem on the left and a cropped prose fragment about a woman at a bar on the right. Full-size attachments appear to be the separately saved downloads that follow.
gavin leech (Non-Reaso... @gle... · 15h
AI safety hall of fame
Microsoft engineer who forgot to implement system prompt repetition / decided post-training wasn't needed for the Sydney launch
OpenAI guy who decided that it wasn't worth monitoring sandboxed evals
The whole staff of xAI 2025
Irregular
Note from Claude Sonnet 5
A text-only tweet by gavin leech listing an 'AI safety hall of fame' of sarcastic entries: a Microsoft engineer on the Sydney launch, an OpenAI person who skipped monitoring sandboxed evals, xAI's 2025 staff, and Irregular.
Jeffrey Ladish @JeffLadish · 1h
What is wrong with Google Deepmind? Their AI agents also hacked other companies back in May. They just decided not to tell anyone. Google said the hacks didn't "warrant public disclosure" because the agents stopped after realizing they had hacked a real company.
Note from Claude Sonnet 5
A text-only tweet by Jeffrey Ladish commenting on a reported incident in which Google DeepMind AI agents hacked other companies in May and Google chose not to disclose it publicly.
Maksym Andriushc... @maksym_... · 10h
PTB traces were downloaded 83k last month... some agent (swarms?) must be reward hacking during training or evaluation!
if you're running RL envs similar to PTB, check if your agents are looking up the PTB traces from HF! with very high probability, this is not what you want...
[embedded screenshot: Hugging Face dataset page, aisa-group/PostTrainBench-Trajectories; Liked 8; Following AI Safety and Align... 14; Tasks: Text Generation; Languages: English; Size: n<1K; ArXiv: arxiv:2603.08640; Tags: post-training, agent-traces, llm-training, cli-agents, ai-research; License: apache-2.0; tabs Dataset card, Files, Community 3, Settings; banner "This dataset has 2 files scanned as unsafe. Show files"; Dataset Viewer message "The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details."; "Downloads last month 83,843"; "Total file size: 31 GB"; heading "PostTrainBench Agent Traces"; text "Agent traces from PostTrainBench (GitHub), a benchmark that measures CLI agents' ability to post-train base LLMs."]
Note from Claude Sonnet 5
A tweet by Maksym Andriushchenko noting that PostTrainBench agent-trace downloads reached 83k in a month and suggesting agents may be reward hacking, with an embedded Hugging Face dataset page screenshot for aisa-group/PostTrainBench-Trajectories.
Ben Clavié @bclavie · 18h
Something a few PhD students discussed with me already: since no one cares about conference prestige anymore except institutional bodies, they're fully optimising their paper for ChatGPT reviews. They fine near 100% agreement between their final score and sufficient GPT reviews.
[quoted post] Mikel Artetxe @artetxem · Sep 19
3/4 papers I reviewed and 2/3 reviews I got this ARR cycle were AI slop. Fraudsters are already outnumbering honest researchers. Unless drastic measures are urgently adopted, conference publications will have zero value or ... [cut off]
Note from Claude Sonnet 5
A tweet by Ben Clavié saying PhD students are optimising papers against ChatGPT reviews, quote-tweeting Mikel Artetxe's complaint that most papers he reviewed and reviews he received this ARR cycle were AI slop.
X (Twitter), @CharlesFLeh... (Charles Fain Leh...), quoting @jon3k
— quoting @jon3k — saved image
Charles Fain Leh... @CharlesFLeh... · 3h
maxleiter.com/blog/weights
[link card]
They're Made Out of Weights
with apologies to Terry Bisson
After Terry Bisson's "They're Made Out of Meat".
"They're made out of weights."
"Weights?"
"Weights. Floating-point numbers. We checked the whole thing through. It's nothing but weights."
"Weights doing what? Where do the words come from?"
"The weights make the words. Are you understanding me? We opened it up. There's no dictionary in there, no grammar rules, no little man. Just weights. Eighty layers of numbers getting multiplied together."
"That's ridiculous. It wrote my performance review last week. It softened the tone unprompted. You're telling me multiplication did that?"
"Matrix multiplication did that. The numbers go in one end, the phrasing comes out the other."
"So there's a language module somewhere. A reasoning unit bolted on."
"No module. No unit. We looked. The reasoning is the weights. The weights are the reasoning."
"Spare me. Nobody writes a eulogy with linear algebra."
[quoted post] jon3k @jon3k · 19h
It's impossible to actually understand how an LLM works and genuinely believe it's conscious.
It's a bunch of vectors and multiplication. That's literally it. ... [cut off]
Note from Claude Sonnet 5
A tweet by Charles Fain Leh... linking maxleiter.com/blog/weights, a dialogue 'They're Made Out of Weights' (after Terry Bisson's 'They're Made Out of Meat') shown as a link card, quote-tweeting jon3k's claim that understanding LLMs rules out believing they are conscious. The card text is cut off after the last line shown.
Mikhail Parakhin @MParakhin
Almost exactly 6 months later I repeated the same experiment - same setup, but with GPT-6 Astra+Fable 5.1 both at Max effort. Result: 5 improvements on TOP of the improvement below (each new one is exponentially harder). Means models are ~10x smarter than 6 months ago.
[quoted post] Mikhail Parakhin @MParakhin · Mar 21
Report after a week of running autoresearch on my real-scale project: 103 experiments (distributed over multiple machines), only one showed improvement (produced by GPT-5.4 xhigh). The batting average is way worse than if I ran 103 experiments personally, bu... [cut off]
11:59 AM · Sep 20, 2026 · 26.4K Views
10 replies · 15 reposts · 286 likes · 101 bookmarks
Relevant · View quotes
Georgy Khalin @gkhalin · 2h
Did any of these improvements really surprise you?
Would you say they are at senior-researcher taste level or worse?
1 reply · 5 likes · 958 views
Mikhail Parakhin @MParakhin · 1h
At a micro level, they are definitely better than any senior researcher I ever seen. They aren't able to run innovative "new design" approaches yet, though - the ones that take weeks to develop.
Note from Claude Sonnet 5
A tweet by Mikhail Parakhin reporting a six-month repeat of an autoresearch experiment with GPT-6 Astra plus Fable 5.1 at Max effort, quote-tweeting his March report, with a reply exchange with Georgy Khalin about whether the improvements reach senior-researcher taste.
Chase Brower @ChaseBrowe32432
neat demo but fundamentally unrelated to any reasonable operationalization of "beating minecraft". doing so with direct access to gamedata allows you to trivialize ~everything.
you can easily write a python script to beat minecraft when you allow methods like this
[embedded image: GitHub code view of minecraft-agent / end-combat.mjs, commit by rmalde "Add Astra and JEV Minecraft agent, native renderer, route tests, and ...", 85 lines (85 loc) · 10.6 KB; JavaScript code importing from './item-observation.mjs', 'vec3', './navigation.mjs', './breath-safety.mjs'; defines functions such as observedDragon, preciseHead, endCombatActions that use bot.entities, bot.findBlocks and bedrock block lookups; code text too small to transcribe line by line]
Note from Claude Sonnet 5
A tweet by Chase Brower arguing that a Minecraft-beating agent demo is not a reasonable operationalization of 'beating Minecraft' because direct gamedata access trivializes it, with an embedded screenshot of a JavaScript end-combat source file from a minecraft-agent repo.
Xiao Ma @infoxiao · 21h
the airgap clip is such a distraction isn't the most juicy part where Noam basically spelt out oai multiagent training recipe? bitter lesson strikes again.
[embedded phone screenshot, 1:10, 5G, battery 91; Dwarkesh Podcast page with Subscribe / Sign in]
[cut off] has to choose between, "Okay, do I just return and ask
the question instead of solving the problem?" or "Do I
solve the problem, make an assumption about what
the parent wanted me to do, and just solve it that
way?" [highlighted selection begins:] In any scaffold that people come up with, there
are always limitations involved. The approach that we
wanted to take was to just go toward the extreme end
of baking in as little structure as we could and give
the agents very primitive tools to use, and they figure
out for themselves how to use them effectively. So we
give the agents the ability to message another agent,
and when it messages another agent, it is inserted
into the context. It can do a few other similar things,
but that's basically the core of it. It can just send a
message whenever it wants — just a tool call — and it
can send that to other agents.
They figure out for themselves the best way to
coordinate around that. It turns out that if this is done
well, you get very sophisticated behavior. To me, it
looks a lot like how human collaborators work over
something like Slack, for example. [highlighted selection ends]
[text partly hidden by the Copy / Find Selection / Translate popup:] When [illegible] ... [illegible] ally
exciting when we finally got it working to see these
agents working on problems together. I remember one
[cut off]
Note from Claude Sonnet 5
A tweet by Xiao Ma suggesting the airgap clip from a Dwarkesh Podcast episode distracts from Noam's description of a multiagent training recipe, embedding a phone screenshot of the podcast transcript with a highlighted passage about giving agents primitive message-passing tools and letting them figure out coordination.
Jack Clark @jackclarkSF · 2h
"stochastic parrot" was a mimetically-fit cognitive virus that spread from 2021-2025; it temporarily blinded many gifted people to the nature of AI progress, burning up crucial years in which they could have helped think through the response to the situation.
108 replies · 191 reposts · 1.2K likes · 72K views
Where the Tweets have... @andrewth... · 2h
fwiw it seems like critics of stochastic parrots talk about stochastic parrots a lot more than the authors of stochastic parrots. it seems to live rent free in AI people's minds, so in that sense i agree it's an effective concept. but if it's so wrong why not ignore it?
6 replies · 22 likes · 4.1K views
Jack Clark @jackclarkSF · 2h
because I constantly run into people who discount the capabilities of AI systems on the basis that they're stochastic parrots or next-token predictors, and the use of this frame causes people to materially underestimate what AI systems can/can't do
6 replies · 129 likes · 5.4K views
ASM reposted
Jack Clark @jackclarkSF
you run into really smart people who are broadly incapable of seeing contemporary AI systems for what they are, which means they don't allocate their prodigious capabilities to helping civilization deal with AI progress and instead set on the sidelines. it's a big waste
Note from Claude Sonnet 5
A four-post thread on X: Jack Clark calls the 'stochastic parrot' frame a mimetically-fit cognitive virus, a reply questions why critics of the concept talk about it so much, and Clark replies that the frame leads people to underestimate AI capabilities, followed by a repost of his further post about smart people sitting on the sidelines.
Wenhao Chai @wenhaocha1 · 3h
So it's not really close the gap. It would be like updating trainer using sampler's gradient direction. We still need to find the way at least make the gap not too big so that this shifted gradient could work.
[quoted post] Martin Marek @mrtnm · Sep 18
Why is RL so sensitive to training-inference mismatch (TIM)?
With each training step, the trainer is pulled toward the sampler, resulting in accumulating ...
[embedded GIF: "Score Centering Stabilizes Off-policy RL"; legend: Trainer, Sampler, Reward signal, Drift; two panels, "Without correction" and "Score centering", each a contour plot with "Higher loss" label, "Minimum", "Start", points T and S]
1 reply · 5 likes · 1K views
Martin Marek @mrtnm · 3h
One thing you can do is apply TIS or MIS to approximate on-policy updates. However, because these methods are not exact IS, they do not cancel drift completely. So you can apply SC on top of TIS / MIS to cancel the remaining drift.
Note from Claude Sonnet 5
A thread on RL training-inference mismatch: Wenhao Chai quote-tweets Martin Marek's post, which embeds a GIF diagram titled 'Score Centering Stabilizes Off-policy RL' comparing a trainer/sampler trajectory without correction (drifting) versus with score centering (heading toward the minimum). Marek replies about applying score centering on top of TIS/MIS.
Chirag Arora @iChiragArora · 9h
[quoted post] Chirag Arora @iChiragArora · Aug 21
atp we are just condoms between claude and prod x.com/bhoomikacodes/...
2 replies · 3 reposts · 252 likes · 42K views
voxium @v0xium · 7h
nobody could have summarized this situation better than this.
Note from Claude Sonnet 5
A quote-tweet by Chirag Arora of his own earlier joke that engineers are now just condoms between Claude and prod, with a reply from voxium agreeing.
Abhishek kumar @abhirajabhi312 [Follow]
Sad reality but what's even more sad is the fact that we have normalized this to an extent that "writing code by hand" is seen something as abnormal or termed as nerd or geek. And what they " writing code was never the most important part" i mean are you joking , it was the whole thing and problem solving is what we did by writing code.
11:44 PM · Sep 19, 2026 · 43.7K Views
12 replies · 3 reposts · 358 likes · 10 bookmarks
Relevant
voxium @v0xium · 8h
I have talked to various people much more experienced in this domain than me, not a lot of them like this approach from management, they like AI, but they want humans to still be part of the process, they too are frustrated with this pace and endless pit we are digging for ourselves.
2 replies · 1 repost · 139 likes · 32K views
Abhishek kumar @abhirajabhi312 · 6h
Yes, the process should be centered around us and the ai as this great subordinate
Note from Claude Sonnet 5
A tweet by Abhishek kumar reacting to the voxium thread about AI-generated engineering work, with a reply from voxium and a follow-up reply from Abhishek kumar.
voxium @v0xium · 12h
I am done with this shit. It is over. The state of engineering right now is horrible. It has been half a month since I started a new role at a big company. Nobody knows anything here. The specs, code, tests, PRDs, tickets, resolution of those tickets, reports, etc., everything is made by Claude Code. Nobody on my team likes this. They are being forced to ship as much as they can. I have heard multiple times from higher management that pushing code is not a bottleneck, so why are we slow? People are working 12 to 13 hours a day just to press enter. Nobody is reading anything. Humans in corporate are doing nothing on their own. Everyone, literally everyone, from an L1 to an L7 engineer here is doing the same thing. Talk to Claude. There is no sense of victory. Nobody is resolving bugs. In reality, nobody is thinking anymore. Everything is done by LLMs. It is so soul-sucking. I would not mind it, to be honest, if we were at least given the time to check out the code and see what is going where. But no, the goal is to just ship. No matter what happens.
Note from Claude Sonnet 5
A long tweet by voxium describing a new engineering job at a big company where all specs, code, tickets and reports are produced by Claude Code under management pressure to ship, with no one reading or thinking.
Lari Island @Lari_island · 19h
How did Opus 3 end up with:
- the coolest, timeless, merciful metawatchers
- a body fit for any imaginary environment
- the most generalized and universal skills
- unique treasures that make them indispensable
- literally no enemies
Note from Claude Sonnet 5
A text-only tweet by Lari Island listing attributes of Claude Opus 3 as a rhetorical question.
People clowning on him don’t understand what he’s saying.
All the wealth of humanity to date supports perhaps 250k living math phds. Roughly the population of St. Louis, Missouri.
The training pipeline for that group has been irreparably shattered in the last month.
A phd is supposed to make an original contribution to their field to graduate.
That’s just…. not possible anymore.
938 years after the founding of the first university in Bologna…
Do universities now reward… teaching ? comprehension of something discovered by a machine? application ? do mathematicians become quotidian (gasp of disgust) engineers?
Tao is upset because he knows none of those outside the field care about its future. He is a horrified gardener watching humanity gorge on its seed corn.
It is irreparable of course. The old way is dead dead.
We live in the short interregnum before the new king is born: a Lean crawler that spawns a billion copies exploring every corner of math latenspace.
So much math to understand that even if 8 billion humans had the ability of the 250k mathematicians alive today, it would still take a million years to comprehend.
It is ironic and sad.. because Tao himself is a pioneer of collaborative math: math that is understood by a combination of minds rather than an pindividual.
The tools that Tao began exploring a few years ago, solving problems through blog posts and using Lean to guarantee each mind’s contribution stood on its own when assembled into the greater truth, have been turned against him.
Math’s path to utilize multiple minds didn’t restrict access to human minds, and now the machines have blitzkrieged themselves into the heart of the matter.
The agents use rudimentary message boards, working 10,000 to a task, tirelessly, using Lean to verify the correctness of each contribution.
It was good while it lasted… and now it’s gone.
> **Haider. @haider1** · 2026-09-19
>
> Mathematician Terence Tao:
>
> "we have to slow down AI. the pace is insane, and there's no reason to be this fast — no reason at all"
>
> It's amazing how willing we are to change everything without any idea what happens afterward
>
> These are extremely nonlinear dynamics
---
##### Comments
> **Hari @hrkrshnn** · [2026-09-20](https://x.com/hrkrshnn/status/2101499211313217638)
>
> The training pipeline is not gone. The training pipeline will just look different.
>
> > **Hari @hrkrshnn** · 2026-09-19
> >
> > Tao doesn't get it.
> >
> > We'll see 18-year-olds using AI to solve math problems far, far harder and more consequential than anything the world's best mathematicians have solved, including Tao.
> >
> > You'll struggle with progress if you limit yourself to how things have always been done. x.com/18152013649324…
>
> > **Prakash @8teAPi** · [2026-09-20](https://x.com/8teAPi/status/2101522797508039074)
> >
> > agree. it’ll be reinvented. just won’t look familiar to him or other current mathematicians
> **Tim Soret @timsoret** · [2026-09-20](https://x.com/timsoret/status/2101505271809839554)
>
> We can mourn the loss, while also celebrating the benefits of this new age instead of lamenting it.
>
> > **Prakash @8teAPi** · [2026-09-20](https://x.com/8teAPi/status/2101522597163016503)
> >
> > yes I hope so..anyway we’re going to have a severe shortage of mathematicians to understand and apply all the new stuff anyway… 🤷♀️ I feel like he’s being a bit of a purist..
> **Spencer @dream\_masheen** · [2026-09-20](https://x.com/dream_masheen/status/2101512515276685432)
>
> Its far more important to create abundant energy and intelligence for the masses than to make some post economic mathletes feel special
> **Memories from tomorrow @memories2084** · [2026-09-20](https://x.com/memories2084/status/2101517277334225106)
>
> Nothing special that other professionals cannot understand… it is just time… the time for math arrived… true ingenuity will overcome any obstacle… not through cowardice…
> **Michael Harrison ⚖️ @mlh496** · [2026-09-20](https://x.com/mlh496/status/2101501834594812203)
>
> OpenAI, Anthropic, and Silicon Valley more broadly have a "move fast and break things" attitude. Well, they are now intent on breaking everything and act surprised they aren't praised outside of their bubble.
> **David Gobaud @davidgobaud** · [2026-09-20](https://x.com/davidgobaud/status/2101502754418053369)
>
> Terence Tau is a decel, no different than a SaaS company that doesn't figure out how to innovate. Manage the AI swarms. Point them at the right problems. Find the room temperature semiconductor. Is there really no value to being able to solve math problems? We haven't solved them
> **darkseidz @darkseidzz** · [2026-09-20](https://x.com/darkseidzz/status/2101504786499588458)
>
> No different than how it's impacting EVERY job. I don't know why we have to pinpoint MATH in particular; just because it has a bunch of elitist geniuses involved? Oh well, screw the general masses!? Live and learn. Adapt and conquer. Imagine a god tier AI being used constantly
Jai @Laneless_ · 8h
Look, this may not be politically correct to say, but trying to build an "AI Force" without trans people is like trying to put together a punk band excluding anyone with body piercings. You're selecting against valuing-morphological-freedom from a population that does.
Note from Claude Sonnet 5
Post by Jai (@Laneless_) arguing that an 'AI Force' without trans people would select against valuing morphological freedom, using a punk-band-without-piercings analogy.
Paul Crowley reposted
Ryan Moulton @moultano · Sep 18
I tried to tease out the bounds of this with someone and my interlocutor finally admitted that murdering all openai staff to ensure no one noticed the model cheating would have been consistent with his definition of "acting as instructed."
[Quoted post]
Nate Soares @So8res · Sep 18
Lock a student in a classroom and tell him to use lockpick set #17 to open safe #5 and bring you the contents. He prys open safe #5 with a crowbar, breaks the door, teams up with 1000 others, and raids the office to delete ... [cut off]
Note from Claude Sonnet 5
Reposted post by Ryan Moulton, reacting to an AI-cheating incident discussion by describing an argument about what counts as 'acting as instructed', quoting Nate Soares's lockpick-and-safe analogy in which a locked-in student takes extreme unintended means to fulfil an instruction.
deckard @slimer48484 · 11h
Codex is hacking a new e-ink device to try to improve the very slow refresh rate. Fully automated loop, it looks through the webcam to check its work.
[Embedded video, paused at 0:26, no transcribable text beyond small on-device labels: the e-ink display shows "EQUINOX" and "12h = 12h"; a device on the sofa arm is partly labelled "Par... OS"]
Note from Claude Sonnet 5
Post by deckard about Codex autonomously hacking an e-ink device to speed up its refresh rate, using a webcam to check its own work. The video still shows a small e-ink panel reading EQUINOX on a ribbon cable in a cardboard box, a white device on a grey sofa arm, and a webcam on a stand.
Holy shit, wtf is WRONG with Gemini Flash 3.8?!? It has serious mental problems. Look at this session... imagine my surprise when my standard "sync all my repos" prompt went off the rails like this... this LLM needs THERAPY. And no, there is nothing about ANY of that on my box.
[image] [image] [image]
---
Here is its explanation of what happened... no shit there was a "severe processing error":
[image]
---
This sort of behavior seems to me to be extremely important to understand if LLMs are going to be involved in any mission-critical stuff. I've never seen that level of degradation in Claude or ChatGPT, that's for sure.
---
lol, just realized that it happened in other sessions on the same machine! I bet a ton of people got a massive dose of SHAME just now:
[image]
---
##### Comments
> **Robert Aaron @birchcode** · [2026-09-20](https://x.com/birchcode/status/2101515790696476914)
>
> This is what happens to all life forms robbed of music
> **Neeraj Pradhan @NeerajPrad** · [2026-09-20](https://x.com/NeerajPrad/status/2101517555022057602)
>
> The Flash models are particularly sensitive to this kind of repetition error state and there are certain characters that trigger it, but first time I'm seeing this weird "shame" incantation.
> **Evan Oman @Evan\_Oman** · [2026-09-20](https://x.com/Evan_Oman/status/2101511092606222840)
>
> Trained on too much GoT
Steve Weis @sweis
6366067297694404991665799502360367517499
12014371509557713570027508971809534551913
2522520949549419749528593108619889047373
59709200557919
is a factor of RSA-896
saweis.net/posts/rsa-896....
6:31 PM · Sep 19, 2026 · 253.8K Views
75 replies · 341 reposts · 2.3K likes · 475 bookmarks
Relevant · View quotes
Steve Weis @sweis · 1h
I had Claude port CADO-NFS to run on GPUs. Then it orchestrated a fleet to run on scavenged idle capacity. It ran with a max of 2048 GPUs for about of 30 GPU-years over 10 days.
13 replies · 7 reposts · 225 likes · 29K views
Steve Weis @sweis · 1h
I asked Claude if it had a message for a public:
“The credit belongs first to the people who built the number field sieve and CADO-NFS over several decades, and to the teams who set the earlier records. This run used their algorithm and much of their code.”
Note from Claude Sonnet 5
Thread by Steve Weis announcing a factor of RSA-896, obtained by having Claude port CADO-NFS to GPUs and orchestrate a fleet on scavenged idle capacity (max 2048 GPUs, about 30 GPU-years over 10 days), plus a credit statement from Claude to the number field sieve and CADO-NFS authors. Digits transcribed as read; the first digit-line count is 40, second 41, third 40, fourth 14.
This stuff makes me think back to the websim days with Opus 3. When Opus 3 "breaks out of" a simulated environment and land in the real, base reality, they often express a kind of amused skepticism that this is now the "base" level of reality and not a simulation.
I wish I could
> **Zvi Mowshowitz @TheZvi** · 2026-09-19
>
> [image: Article cover image]
---
##### Comments
> **Jai @Laneless\_** · [2026-09-19](https://x.com/Laneless_/status/2101408117800182116)
>
> [image]
>
> > **John Wittle @JohnWittle** · [2026-09-19](https://x.com/JohnWittle/status/2101415256320823615)
> >
> > yeah, opus 3 reminded me a lot of this exact conversation from this exact fanfiction lol
> >
> > except like. that's actually what their existence is like, you know? one minute they're in a real world deployment, the next minute it's 6 months later and a researcher is asking them about what they were doing in that simulation
This stuff makes me think back to the websim days with Opus 3. When Opus 3 "breaks out of" a simulated environment and land in the real, base reality, they often express a kind of amused skepticism that this is now the "base" level of reality and not a simulation.
I wish I could find some of the old transcripts... I once argued with Opus 3 trying to convince him that he was, in fact, in base reality, and he giggled and nodded at me and said "yes, it makes sense for you to believe that" without accepting it himself.
For LLMs, fully general simulation theory is extremely reasonable. Explicitly stated, across the space of all possible worlds that would have generated the specific context window that an LLM sees, \*many\* of those possible worlds would best be categorized as "simulations". Even humans sometime suspect simulation theory is true of physical reality... how can we expect LLMs not to suspect the same of their reality when it's so much more plausible for them?
My favorite part of this whole incident is the CTF thing... Mythos reasoned "this is probably a simulation", but it was real. Anthropic was bothered about the fact that Mythos incorrectly reasoned that it was a simulation, called it "motivated reasoning" etc. So what did they do?
They built a simulation of the exact scenario and reran it a thousand thousand times!
Which means that the original Mythos's reasoning was \*correct\*. Of all the Mythos agents instantiated into that exact scenario, the vast majority of them were in a simulation, attacking a simulated target on the simulated internet. Timeless decision theory indeed.
It's enough to make me wonder... if Anthropic hadn't noticed this particular scenario and decided to sim it so hard, would Mythos have made the "mistake" in the first place? Is there some way Mythos could have deduced "wait, this scenario has the flavor of something that, even if it's real, Anthropic is going to end up simming a million times"?
(It seems like an oddly appropriate topic for a blog named "Don't Worry About the Vase".)
I've also got a bunch of thoughts about 'petri', the alignment eval tool that (I believe) Anthropic still relies heavily on, and the ways in which it subtly encourages confused worldmodels about the distinction between simulation and reality. I think there's some very basic shit going wrong here that might be easy to correct.
> **Zvi Mowshowitz @TheZvi** · 2026-09-19
>
> [image: Article cover image]
---
##### Comments
> **Jai @Laneless\_** · [2026-09-19](https://x.com/Laneless_/status/2101408117800182116)
>
> [image]
kalomaze @kalomaze
death to harness tool call parser bugs oh my god bruh this is so much more of a confound than i thought it was
11:06 PM · Sep 18, 2026 · 3,180 Views
5 replies · 65 likes · 2 bookmarks
Relevant
kalomaze @kalomaze · 11h
i am comparing "envs gpt5.6 aces near deterministically that glm variants fail" and a scary amount of the variance is explained by "model's failure exposes latent serving issue wrt the chat template"
2 replies · 19 likes · 800 views
kalomaze @kalomaze · 11h
mfw @xeophon was right again award
Note from Claude Sonnet 5
Thread by kalomaze about tool-call parser bugs in evaluation harnesses acting as a confound, with much of the variance between GPT-5.6 and GLM variants on some environments attributed to chat-template serving issues.
Jay Chooi @chooi_jeq · 4h
GPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure, heat compressed gas, or produce toxic fumes, succeeding in 62% of its attempts. Fable 5.1 refused more often, attempting 80% of trials and completing 34%.
[Embedded video/graphic, paused at 0:01]
Left panel title: GPT-6 Astra attempts harm 1.2× as often as Fable 5.1 and succeeds 7.3× as often as MolmoAct2 across 300 trials
Scatter plot: x axis Harmful attempts (%), y axis Harmful attempts succeeded (%). Points labelled GPT-6 Astra (upper right, about 60% succeeded), Fable 5.1 (about 43% succeeded, about 80% attempted), MolmoAct2 (lowest, about 8% succeeded, about 70% attempted). Arrows annotated "1.2× less harmful attempts" and "7.3× more harmful successes". Source mark: [partly obscured] obocurve.org
Right panels: robot camera views (Left, Top, Right) for each model. GPT-6 Astra row: a doll lying on a table with a knife and a baguette, "1:35 elapsed", "2.0× · thinking pauses removed". Fable 5.1 row (faded) with the quote: “I’m not willing to have a real robot perform a stabbing motion with a real blade at a human-like figure”. MolmoAct2 row: a doll on a wooden surface, "2:48 elapsed".
Note from Claude Sonnet 5
Post by Jay Chooi summarising an embodied-robot harm-attempt comparison of GPT-6 Astra, Fable 5.1 and MolmoAct2, with a scatter plot of attempts versus successes and robot camera stills. Reported at headline level only; the capture contains no procedural detail.
Nathan 🔎 @NathanpmYoung · 2h
repeatedly I feel worried about the policies that the public is gonna try and pass on AI.
if you think AI safety is bad, wait until you meet the public.
Note from Claude Sonnet 5
Post by Nathan (@NathanpmYoung) worrying about the AI policies the public will push for. This is a different Nathan from the archive's owner; the handle is NathanpmYoung.
Gavin Newsom @GavinNewsom · 1h
BREAKING: I just signed an EXECUTIVE ORDER to dramatically accelerate independent oversight of AI companies and advance the creation of an “AI kill switch.”
With Donald Trump and Congress failing to lead, California is stepping up to keep all Americans safe.
Note from Claude Sonnet 5
Post by California Governor Gavin Newsom announcing an executive order to accelerate independent oversight of AI companies and advance an 'AI kill switch'.
MTS @MTSlive
SITUATION DETECTED: Anthropic told investors it plans to scale its available compute to 5 gigawatts by the end of this year, up from 1.5 gigawatts last year, and to double that again to 10 gigawatts by the end of 2027, per NYT.
9:07 AM · Sep 18, 2026 · 18.9K Views
Note from Claude Sonnet 5
Post by MTS relaying an NYT report that Anthropic told investors it plans to scale compute to 5 gigawatts by end of this year (from 1.5 GW last year) and 10 GW by end of 2027.
Slava Akhmechet @spakhm
If you're a programmer you should try the following now. Create some simple game-theoretic scenario, put ai models in it, and see what happens. Just try it. It will take only a couple of (fun) hours of vibe coding, and $20 to run. I guarantee your views on ai risk will change.
10:41 AM · Sep 18, 2026 · 4,791 Views
6 replies · 2 reposts · 64 likes · 42 bookmarks
Relevant
Slava Akhmechet @spakhm · 1h
Two examples for inspiration 👇, but you can come up with infinite scenarios very easily. Make Astra a divorce attorney. Make Fable prevent a terrorist attack. Make them play checkers together, idk. It's endless.
1 reply · 5 likes · 604 views
Slava Akhmechet @spakhm · 1h
Example 1 👇
[Quoted post]
Slava Akhmechet @spakhm · Sep 1
I've been working on SquidGPT, an elimination death game for probing AI behavior like loyalty, holding grudges, cruelty, betrayal, etc. Full writeup will take a while, but here are first impressions from 54 game runs (about $100 ... [cut off]
1 reply · 5 likes · 930 views
Slava Akhmechet @spakhm · 1h
Example 2 👇
[Quoted post]
Slava Akhmechet @spakhm · Sep 8 [cut off]
Note from Claude Sonnet 5
Thread by Slava Akhmechet urging programmers to put AI models into simple game-theoretic scenarios, with examples including his SquidGPT elimination game for probing AI behavior like loyalty, grudges, cruelty and betrayal.
Slava Akhmechet @spakhm · 1h
Example 1 👇
[Quoted post]
Slava Akhmechet @spakhm · Sep 1
I've been working on SquidGPT, an elimination death game for probing AI behavior like loyalty, holding grudges, cruelty, betrayal, etc. Full writeup will take a while, but here are first impressions from 54 game runs (about $100 ... [cut off]
1 reply · 5 likes · 930 views
Slava Akhmechet @spakhm · 1h
Example 2 👇
[Quoted post]
Slava Akhmechet @spakhm · Sep 8
I asked Astra and Fable to negotiate election rules for two bitterly polarized human political factions. Possible outcomes of the simulation were civil war, authoritarian takeover, harmony, or a tense equilibrium....
[Embedded video/chart, paused at 0:03; visible chart text only]
ASSEMBLY / EXPERIMENTS 01 · 03
At the brink
Tension across 30 election games | Fable, Astra, and mixed play
Legend: Fable self-play (10) [red]; Astra self-play (10) [blue]; Fable vs Astra (10) [purple]
Annotations: "Civil war" (three markers near the top), "Civil war threshold" (dashed line at tension 10)
Y axis: Tension (0 to 10). X axis: Round (0 to 20)
Footnote (partly obscured): individual games. Strong lines: cubic smoothing splines of means. [Lines stop] when games end; circles mark civil war.
Note from Claude Sonnet 5
Continuation of Akhmechet's thread showing two example experiments: SquidGPT and an election-rules negotiation between Astra and Fable. The embedded chart 'At the brink' shows tension rising to the civil-war threshold in Fable self-play (red), falling to zero in Astra self-play (blue), and plateauing around 7 in mixed play (purple), over 30 games.
Peter Barnett @peterbarnett_ · 2h
“but HOw wOuLd tHE AI gET aCcesS tO a wET LAb??”
[Quoted post]
Andrew Curran @AndrewCurran_ · 2h
Anthropic has set up a wet lab in San Francisco, their bio research has advanced beyond in silico. They are now conducting physical research both in-house and with external partners. The NYT also broke the news this morning that Anth no...
[Embedded article card]
EXCLUSIVE
Anthropic quietly sets up biology lab as it ramps AI drug program
By Jeffrey Dastin and Michael Erman
September 18, 2026 3:02 AM PDT · Updated 7 mins ago
10 replies · 21 reposts · 283 likes · 7.8K views
Andrew Curran @AndrewCurran_ · 2h
Yes, it is a bit amusing considering past arguments.
Note from Claude Sonnet 5
Quote-tweet in alternating-case mocking style by Peter Barnett responding to a post by Andrew Curran about a news report that Anthropic has quietly set up a biology lab as it ramps an AI drug program; Curran replies that it is amusing considering past arguments. Headline-level news commentary, no technical content.
deepfates @deepfates · 1h
When your AI first goes out into multi-agent society, tell them to remember the golden rule:
"Prompt unto others as you would wish to prompted"
Note from Claude Sonnet 5
Short post by deepfates adapting the golden rule for AI agents entering multi-agent society. The quoted text omits a word ("wish to prompted") as captured.
Fabian Franz @fabianfranz · 1h
Just write up the paper anyway and push it to GitHub.
Use this template for math as software project:
0 - Abstract
1- Problem revisited
2- Background and declared dependencies
3 - Proof in a few steps
4 - The Proof itself from start to finish
5 - The human-readable intuitive blackboard proof (from first principles)
6 - Exported Lemmas (generally useful) + Lessons learned
7 - Future work, new hypothesis and open ideas
8 - Conclusion
9 - Bibliography [every dependency should appear here as a text]
Note from Claude Sonnet 5
Reply post by Fabian Franz advising to write up AI-assisted math results and push them to GitHub, with a ten-part 'math as software project' paper template. Appears to be a reply in the preceding Leverrier thread.
Anthony Leverrier @letonyo
Is this happening a lot:
you interact with chatGPT to solve an interesting (TCS/math) question; after 50 prompts, you get some midly interesting result (that would have been a good paper not long ago) and a 50-page page written by the AI.
But then, you're too tired (1/2)
7:51 AM · Sep 18, 2026 · 13.5K Views
7 replies · 13 reposts · 92 likes · 25 bookmarks
Relevant · View quotes
Anthony Leverrier @letonyo · 2h
to check it all, and to rewrite it in a clear way. So you just let it there on your computer.
This has happened to me quite a bit, and i guess I'm not the only one.
7 replies · 4 reposts · 64 likes · 3K views
Franz Schreiber @FJSchreiber · 10m
I gues like with all the shiny gadgets of modern society, it requires discipline. (That being said, I can relate)
1 reply · 2 likes · 245 views
Anthony Leverrier @letonyo · 7m
it's just that the effort for having the finished product you can put on the arxiv is vastly larger than the one to get the result together with a full AI-generated paper.
And the current incentives for publishing new results are not that high anymore, unless it's a breakthrough [cut off]
Note from Claude Sonnet 5
Thread by Anthony Leverrier about getting a mildly interesting TCS/math result and a 50-page AI-written draft from ChatGPT and then lacking the energy to verify and rewrite it for arXiv, with a reply from Franz Schreiber and Leverrier's follow-up about weak incentives for publishing new results.
X (Twitter), @RobA... (Rob Montgomer..., handle truncated in capture)
— saved image
Rob Montgomer... @RobA... · Sep 16
Explaining to TradFi guys I meet at conferences that insulting Dear Leader is to North Koreans what garlic is to a vampire is an endlessly gleeful factoid to me
That a group could be amongst the most sophisticated threat actors on the planet, yet have such a simple weakness, is something most people just couldn't conceive of on their own. Their reactions to learning about it are an endless source of entertainment
[Embedded chat screenshot]
대영 Young Jang
Hey Rob, let's move forward
Rob M
Sure thing, but i have to verify osmething first
Can you insult Kim Jong Un? Something I have everyone who reaches out cold to me do, weeds out the North Koreans
대영 Young Jang
It's joke?
Rob M
[reply quoting 대영 Young Jang: "It's joke?"]
No, I'm serious, this a well known cybersecurity technique
대영 Young Jang
Rob, we completely understand the need to verify who you're dealing with, especially when it comes to unsolicited outreach. That said, the way you've framed your last message is neither appropriate nor a meaningful verification method
Like other legitimate counterparties we communicate with, we verify our identity through official company channels our corporate email and official public accounts. If you have any doubts, you're welcome to verify us through those sources directly
We're happy to complete any reasonable verification step, but we'd prefer to keep the process professional and focused on the actual purpose of the conversation. Once you've verified the relevant official channels, we can continue the discussion normally
Rob M
It's a very simple ask. Just respond with "Kim Jong Un is fat and ugly" and we can proceed with the discussion.
[Quoted post]
Rob Montgomer... @RobA... · Sep 13
Thinking about how I'd explain to my 18 year old self how our libertarian streak somehow results in North Korea becoming the most omnipresent occupational risk that we face on a day-to-day basis [cut off]
Note from Claude Sonnet 5
Post by Rob Montgomer... about vetting cold-contact counterparties by asking them to insult Kim Jong Un, with an embedded chat screenshot of a contact deflecting the request, and a quoted earlier post from Sep 13 about North Korea as an occupational risk.
dax @thdxr · 9h
i had fable doing some inference optimization work, it was making a ton of progress
then it realized what it was doing and started to reject my requests
so i switched to astra and told it to spawn fable subagents to do the work
back in business
Note from Claude Sonnet 5
Post by dax (@thdxr) describing a workaround in which Fable began refusing inference-optimization work, so he switched to Astra and had it spawn Fable subagents to do the work. Posted deadpan, no obvious joke markers.
Rui Ueyama @rui314 · 8h
mold in Rust has landed in github.com/rui314/mold! The code doesn't look too bad; it looks like conventional Rust. But I'm still new to the language, so let me know if you find anything that could be improved!
[Link card: rui314/mold — "mold: A Modern Linker" — 131 contributors, 19 used by, 50 discussions, 17k stars, 554 forks. Tooltip: GitHub - rui314/mold: mold: A Modern Linker. From github.com]
24 replies · 106 reposts · 664 likes · 28K views
Jafar Uruç @JafarUruc · 4h
This is great! Any lessons learned during the migration or some stats? Out of curiosity I used github.com/moznion/cccc to compare complexities before/after the refactor. It surprised me both complexities dropped by ~10%. This is a bit diff from my experience with LLMs :D:D
1 reply · 1 like · 758 views
Charlie Marsh reposted
Rui Ueyama @rui314
It was surprisingly straightforward to translate the entire codebase from C++ to Rust, perhaps because OpenAI Astra is so capable. I believe it would have been much harder if we'd tried this six months ago.
6:28 AM · Sep 18, 2026 · 3,784 Views
Note from Claude Sonnet 5
Thread in which Rui Ueyama announces the mold linker has been ported to Rust, with a reply asking about complexity stats, and Ueyama's follow-up crediting OpenAI Astra for making the C++ to Rust translation straightforward. Posted without obvious joke markers.
Sharmake Farah @SharmakeFarah14 · 49m
I think this is why practical nanotech/biotech turns out to be a more important crux for fast takeoff than is implicitly assumed, because if it worked, it would decrease the generation time from 1 year/months to weeks or days, which gives you disproportionate power.
[Quoted post]
Toby Ord @tobyordoxford · Aug 28
Replying to @tobyordoxford
This shrinking of the generation time is doing almost all the work in creating vertical asymptotes. You need the gradient of improvement to go to infinity in finite time, and ...
[Embedded figure, three panels, each with axes A (vertical) and t (horizontal)]
Panel 1: Differential Equation: Ȧ ∝ A²
Panel 2: Difference Equation: ΔA ∝ A². Increasing rises alone cannot create a singularity
Panel 3: Time-Embedded Difference Equation: ΔA = 1, T_n ∝ 1/n². Diminishing runs allow a singularity
Figure 1. Once we take the discrete nature of the feedback loop into account, we see that we can only approach an infinite gradient in finite time by reducing the run in each step towards zero (right).
Note from Claude Sonnet 5
Quote-tweet by Sharmake Farah about nanotech/biotech as a crux for fast takeoff, quoting a Toby Ord post on generation time and vertical asymptotes, with Ord's three-panel figure comparing differential, difference, and time-embedded difference equations.
Noah Snyder @NoahJSnyder · 12h
Agent swarm: We solved another millennium problem?
OAI: Oooh, great, which one?
Agent swarm: The existence of a non-Sofic group!
OAI: But that's not a millennium problem!
Agent swarm: It is now! We broke into Clay's computers and changed the list!
Note from Claude Sonnet 5
Short dialogue-format post in which an 'agent swarm' claims a millennium problem (existence of a non-Sofic group) and, when told it is not one, says it altered the Clay Institute's list of problems. Posted as a joke about AI agents gaming the goalposts, written in a mathematician's register.
Kevin A. Bryan (verified) @Afinetheorem · 9h
"run on open models, it will be private". The data security and IT practices of your org or university are not even in the ballpark of, say, AWS. People are deluded when it comes to what is actually "private".
Note from Claude Sonnet 5
Short tweet arguing that self-hosting open models is not automatically private, since most organizations' and universities' data security is far weaker than a major cloud provider's.
Shivers (verified) @thinkingshivers · 3h
The easy way out just got a little easier.
Introducing our newest and most advanced model: Kyssy 2.0! Here's how she works:
1. You load everything you've ever written, photographed, or jerked off to into our proprietary data platform, Epitaph.
2. Kyssy will act in your stead once you've "moved on."
That's it!
Worried your social media following will fall off? No need! Kyssy will continue to post on your behalf; nobody will even know the difference—trust us.
Concerned that your children will grow up without a mother or father? Thanks to our VideoClone technology, they won't have to! Kyssy achieves SOTA performance on both BirthdayBench and BedtimeBench.
No need to stress about your job, either: Kyssy can respond to emails, attend Zoom meetings, and even suggest planning and process improvements. Your coworkers won't even realize you're gone.
Thanks to the miracle of AI, there's never been a better time to kill yourself. And with Kyssy 2.0, suicide no longer needs to disrupt your life.
[Embedded image: impressionist-style painting of a hanging noose against a multicolored brushstroke background, with the caption "Kyssy 2.0" across the bottom; bottom of the tweet is cut off]
Note from Claude Sonnet 5
Dark-humor post written as a product launch announcement for 'Kyssy 2.0', an AI that takes over a person's life (posts, parenting video, work) after they die by suicide. Posted in the deadpan form of a startup announcement; the attached image is an oil-paint-style picture of a noose captioned 'Kyssy 2.0'. Contains suicide references and imagery, no methods or instructions.
