iiinova — where your digital self keeps evolving

conputer dipshit @davidcrespo.bsky.social · 58m

yeah, disappointing. their recommendations from a week or so ago say a bit more, at least agmai.org/general-sep29/

Reposted by conputer dipshit

Looks like Deters, who was banned from a Cincinnati courthouse for threatening to burn it down, is burning something down: the career of Kimberly Guilfoyle.

1011430

Reposted by conputer dipshit

noam @noamchompers.bsky.social · 1h

The reason AI is so bad at philosophy is because it’s been trained on all the articles in philosophy journals

211112

conputer dipshit @davidcrespo.bsky.social · 1h

as anyone who has asked a programmer anything about computers should know!

0160

Reposted by conputer dipshit

Colin @colin-fraser.net · 3h

it's this

26612

conputer dipshit @davidcrespo.bsky.social · 3h

I think only 245 of the 377 results are stolen. the rest, though — solid gold

140

conputer dipshit @davidcrespo.bsky.social · 3h

still missing jelly buttons though

2120

conputer dipshit @davidcrespo.bsky.social · 3h

cute answer from claude: sort of, but math can be infinite even without Godel. "Even a complete, decidable theory has infinitely many theorems. Tarski showed elementary Euclidean geometry and the first-order theory of the real numbers are like this."

Screenshot of terminal output:

$ ai does godels incompleteness theorem imply that there is no end to math

opus-5.5 (high) | 13.52s | $0.02527 | 642 -> 1103

In one specific sense, yes. In the broader sense people often mean, not quite.

What it does imply. Gödel's first incompleteness theorem says that any consistent formal system with an effectively listable set of axioms that is strong enough to express basic arithmetic contains statements it can neither prove nor disprove. You can add such a statement, or its negation, as a new axiom. But the enlarged system is still consistent, effectively axiomatized, and strong enough, so it has its own undecidable statements. Repeating this never terminates. So no single fixed set of rules can ever capture all arithmetical truth, and mathematics can never be "finished" in the sense of being reduced to one final axiom system.

A closely related result from Church and Turing makes the point algorithmic. No mechanical procedure can decide the truth of every arithmetical statement, so math can't be fully automated into a solved procedure.

What it doesn't imply.

* Math being endless doesn't require Gödel. Even a complete, decidable theory has infinitely many theorems. Tarski showed elementary Euclidean geometry and the first-order theory of the real numbers are like this. An algorithm settles every question in principle, yet there's no shortage of things to prove, and finding proofs can be computationally infeasible.
* It doesn't say there are questions that are absolutely unanswerable. "Unprovable" is always relative to a particular system. A statement undecidable in Peano arithmetic, like Goodstein's theorem, can be proved in a stronger system.
* It doesn't say anything about human insight transcending machines. Arguments of that kind, such as Lucas and Penrose's, are widely regarded as not following from the theorem.
* It applies only to sufficiently strong systems. Some mathematical domains really can be completely axiomat…

010

conputer dipshit @davidcrespo.bsky.social · 4h

closely related to my belief that the human capacity to create problems for ourselves has to be nearly infinite

3160

conputer dipshit @davidcrespo.bsky.social · 4h

what's the word for atheists who nonetheless have full faith that the universe of possible math is infinite

4171

conputer dipshit @davidcrespo.bsky.social · 4h

why now? simple answer, always the same: the models got good, fast, and cheap at it. kind of a subtle point that these come in a package — "doable but difficult and expensive" tends to be a very short-lived state on the way to "cheap and easy" on the way to "not even legible as a discrete task"

1181

conputer dipshit @davidcrespo.bsky.social · 4h

right, exactly. we just do a lot more calculating and a lot more other stuff

030

conputer dipshit @davidcrespo.bsky.social · 4h

see also: a normal calculator

010

conputer dipshit @davidcrespo.bsky.social · 4h

I think this is a very important and relevant distinction, not pedantic at all

110

conputer dipshit @davidcrespo.bsky.social · 4h

basically I think that what actually happens will make this all look very quaint. so the latter, except I don't think we end up worse off, and there are probably more professional mathematicians rather than less. or rather the definition of professional mathematician looks quite different

130

Reposted by conputer dipshit

conputer dipshit @davidcrespo.bsky.social · 16/09/2026

one thing I will say I that I don’t think we will understand less as a result of directing large amounts of computation to do things for us. we may understand a smaller *proportion* of what we cause to happen, but in absolute terms we will understand a lot more than we do now

5367

Reposted by conputer dipshit

conputer dipshit @davidcrespo.bsky.social · 16/09/2026

the thought I'm poorly getting at above is that there are always big things we don't understand. we're not even aware of most such things right now. the set of things we don't understand, things that push us to go further in our understanding, will now include gigantic machine proofs

1172

conputer dipshit @davidcrespo.bsky.social · 4h

they "reaffirm their guidelines" and say it's up to the community to judge whether they were followed (image 1) agmai.org#openai-release I had missed the guidelines, worth a look (images 2 and 3) agmai.org/general-sep29/

On OpenAI’s Release of Mathematical Results

October 6, 2026.

As announced a few weeks ago, OpenAI has released a large collection of mathematical results generated by an internal model, reporting solutions to hundreds of open questions. This is an important event for mathematics, with consequences both for mathematics and for the mathematical community that extend far beyond the individual results.

AGMAI’s advisory role should not be interpreted as a judgment of the impact of these results or an endorsement of the process by which OpenAI obtained them. We do not speak on behalf of the entire mathematical community, and only the mathematical community can undertake the assessment that is needed.

Making this work public is a first step. This release is the beginning, not the completion, of the process of human understanding and the incorporation of the work into mathematical knowledge. At the same time, the future of mathematical research cannot consist only of understanding results produced by AI labs. Mathematicians must be able to formulate their own questions, develop their own approaches, and explore directions that have not been selected as examples of an AI system’s capabilities. Equitable access to powerful research tools and adequate computational resources are essential to that freedom.

We reaffirm our published recommendations on responsible release. We have discussed them with OpenAI and appreciate the company’s willingness to engage. While we consider these discussions constructive, it is ultimately up to the mathematical community to assess the extent to which our recommendations were followed successfully, and whether there are others we should suggest. We remain committed to engaging with any frontier AI lab on these questions and have already been in contact with several of them.

2.B. Papers that are not yet understood by anybody

The recommendations below are for labs that have AI mathematical output that is not understood by the people who prompted the AI systems. They are split into two parts. The first part is a set of proposed technical norms for the release of AI-generated mathematics. The second part is a recommendation that AI labs provide support for the additional mathematical activities that are needed for humans to be able to understand and assimilate their AI-generated mathematical output and identify possible applications of it.

Step I: Initial release

1. With the help of LLMs, it is easy to make substantial improvements to the initial written version of an AI-generated result. The following actions should be carried out by the AI labs rather than left to mathematicians afterwards.

(a) The literature should be scoured for any ideas that are related to the ideas in the proofs of the results released. Even if the AI lab’s model discovered those ideas independently, it should follow standard mathematical practice and cite the papers in which the ideas were first introduced.

(b) The model, or some other model, should be prompted to produce a version of each proof that is written up in a style that follows the conventions of a traditional mathematical paper. They should contain friendly introductions and precise theorem statements and proofs. They should not be full of wordy reasoning and non-standard terminology that renders them virtually incomprehensible.

The current abilities of LLMs may not be able to reproduce the level of attribution or quality of exposition that we expect of mathematicians, and thus more important work from mathematicians after release may be needed to reach this standard. If so, this should be supported as described in Step II. However, inability to reach a high standard does not absolve AI labs of the responsibility to do the best they can with their models on the two points above.

2. When results are announced, they should be deposited in a timely manner in appropriate scholarly repositories. These should not be controlled by any AI lab and should guarantee certain standards, including that submissions have a persistent citable identifier and that subsequent modifications are appropriately recorded. For AI-generated results, it would be particularly useful if the repository
allows for comments on papers.

We strongly recommend that AI labs refrain from treating the release of mathematical results as marketing vehicles to promote their models, ignoring the substantial negative externalities that this practice inflicts on the mathematical community.

3. For each result released, the AI lab should make public the name of the model, the prompts used, a (summarized) chain of thought, the time taken, and the estimated cost of computation. Releasing additional material that sheds light on the scientific process that produced the results, such as the initial LLM outputs before they were cleaned up (as recommended in (1) above), is strongly encouraged.

4. As far as possible, a proof released by an AI lab should be formalized. Formalization artifacts should meet community standards, including having copyright headers, a challenge file for comparator, and a formalization.yaml. It would be helpful to include machine-readable metadata correlating the natural language and formal artifacts. Where formalization would lead to unacceptable delays, the formalization status of the paper should be clearly stated. For example, perhaps it has been formalized modulo standard results that are accepted by the community

5. Each time a solution to a problem is released, it should be clearly documented how exactly AI came to be used on that particular problem. If many results are released at once, then in addition to the results themselves a further document should be written and made public that references all of the released results and explains how many other proble…

031

conputer dipshit @davidcrespo.bsky.social · 4h

yeah, I guess what I'm saying is that I don't think anyone serious was under the illusion that there's any guaranteed relation between the Lean proof and the NL proof

000

conputer dipshit @davidcrespo.bsky.social · 4h

IMO math is very much alive

000

conputer dipshit @davidcrespo.bsky.social · 4h

right, this is very important — software is already a patchwork of formalizations of messy reality that interact at the edges

061

conputer dipshit @davidcrespo.bsky.social · 4h

now they're coming for my job personally but at least they're validating my predictions x.com/OpenAI/statu...

Screenshot of an X post by OpenAI announcing GPT-6 and Intelligent UI in ChatGPT. The post features a video preview showing a hand holding a smartphone that displays an interactive travel guide and map interface for San Francisco.

1622

conputer dipshit @davidcrespo.bsky.social · 5h

as far as I can tell, verifying the lean formalization of the problem statement is not a "only 5 people can do it" thing, at least not any more than any of the original math was

340

conputer dipshit @davidcrespo.bsky.social · 5h

this isn't really right — people can verify that the lean formalization of the problem statement is good much more easily than they can understand the proof and convert it to natural language. so (modulo Lean bugs, as @lu.is put it) you can have quite a bit of confidence the claim is true

260

conputer dipshit @davidcrespo.bsky.social · 5h

ignore the transcription gore in my prompt, not sure what I actually said in the middle there

000

conputer dipshit @davidcrespo.bsky.social · 5h

to me it seems pretty insignificant and obvious. it doesn't bear on whether the proof itself is correct, assuming you're talking about a canonical and agreed-upon formalization of the problem statement. it's just about whether the NL proof itself corresponds to the Lean proof

$ cb | ai "To me, this seems to be showing something fairly obvious, namely that even if you like the formal statement of the problem that you approved, the actual way the proof itself is done in the Lean proof doesn't necessarily correspond to the natural language proof that has been presented alongside of it. To me that seems fairly obvious, and also not that consequential because just because it doesn't correspond in number 1, you have a Lean proof, so you're pretty confident that the thing itself is true and you just have to figure out what a good natural language proof of it is and convince yourself of it. am I misunderstanding?"

opus-5.5 (high) | 15.52s | $0.18188 | 31212 -> 1291

You're not misunderstanding. Your reading is essentially correct, and the paper largely
concedes it: "The Lean code merely tells us that the theorem is correct, yet the NL proof
… may be incorrect." If the top-level statement is faithfully formalised, a compiling
Lean proof without sorry or extra axioms establishes the theorem, whatever the
accompanying prose says.

The paper's real targets are narrower than its framing suggests.

The NL paper as a citable artifact. Mathematicians cite lemmas, not just main theorems.
If Lemma 8.6 claims an m+4 estimate but only m+5 was verified, then someone who later
cites the sharper version is relying on something unchecked. The examples they found are
of this kind. The Lean versions are weaker or different but sufficient, so the main
result is unaffected. This is a fair point about provenance, but it's modest. The fix is
to state that only the formalised versions of the lemmas are verified.

"Just find a good NL proof" may be expensive. If the formal proof is huge and
machine-generated, extracting a human-comprehensible argument is real work. The provided
NL text can't be assumed to be that argument. This concerns the value of the proof, not
its truth, which is their Section 5.2 point.

The genuinely consequential issue is underweighted. Th…

210

Reposted by conputer dipshit

Josh Kovensky @joshkovensky.bsky.social · 11h

UChicago, like any other college, gets vastly more women applicants than men. And yet, its undergrad student body is disproportionately male. chicagomaroon.com/53804/viewpo...

chicagomaroon.com

Creating a Manosphere at the University of Chicago

Why does the University of Chicago reject so many women?

411342

Reposted by conputer dipshit

Giovanni Colantonio @marioprime.bsky.social · 10h

Ive been picking away at Order of the Sinking Star for the past month. Frankly I have still seen virtually nothing. Certainly can’t offer a review, but I still do have a lot of thoughts about the structure, and what, if anything, we should do with the drama behind it. www.polygon.com/order-of-the...

polygon.com

'500-hour' Order of the Sinking Star Is As Complex and Exhausting As It Sounds

Thekla's latest is all over the place

3343

conputer dipshit @davidcrespo.bsky.social · 10h

you should see my other posts, that one’s nothing

030

conputer dipshit @davidcrespo.bsky.social · 19h

it's not real so pretty neutral

0200

Reposted by conputer dipshit

im jared @mightnot.fail · 06/10/2026

this is literally an ai slop account that posts shit like this in between ten thousand amazon referral links. we have to do better than falling for facebook chain letters lol the first line in the original is the prompt!

510014

conputer dipshit @davidcrespo.bsky.social · 20h

they don't all have to be verified in the sense that many of them probably have canonical Lean formulations, so they can be pretty confident if they have a Lean proof that it's legit. but understanding the proof is something else

4110

conputer dipshit @davidcrespo.bsky.social · 20h