iiinova — where your digital self keeps evolving

Arseny Khakhalin @khakhalin.bsky.social · 12h

Spoilers man, not a good shape!

000

Arseny Khakhalin @khakhalin.bsky.social · 12h

None of that is automatable as it is all soft skills and pure pain 🥲

080

Arseny Khakhalin @khakhalin.bsky.social · 12h

My current vibe in a nutshell: Build a prototype that seems to cover 90% of a product: 1 hour Add features to 92% (breaking it, then fixing it back): 1 day Combine two small projects like that into one: 1 week Make it run at scale: 1 month Make it localizable, gdpr, secure, certifiable: 1 year

3380

Arseny Khakhalin @khakhalin.bsky.social · 21h

Joined the Piranesi club! Paying it forward with this one is not writing anything about it, so I'll say one thing, buy it and read it, it's a good book! Don't read _about_ it, just read it, it's very short, one evening, it's worth it :)

0100

Arseny Khakhalin @khakhalin.bsky.social · 21h

fedi old no future who cares atmo ftw!

040

Arseny Khakhalin @khakhalin.bsky.social · 22h

I have an internal wiki. Each note had a meaningful name and a 150 char description. The list of notes is forced into the context, but agents decide themselves which notes to read, and when. Works really well actually!

010

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Yeah I already split some documents into "for agents" and "for humans", within the same Obsidian note, but it's not very formal yet. I'm toying with an idea to make it more explicit. To make agent-own docs live side-by-side with human docs, but in a different "plane of existence" so to say

220

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Oh yeah btw this could work! Also achieves 2 pairs of eyes, and may be faster than an asynchronous version of it that I'm practicing right now (where I'm reviwing PRs while also fixing things that I don't like right away). At least sometimes, it may be faster!

010

Reposted by Arseny Khakhalin

rev. howard arson @theophite.bsky.social · 09/10/2026

75% of the guys with computer science degrees who believe that intelligence is 100% heritable and racial will spend fourteen hours tomorrow watching to see if their training job diverges because it got the wrong random seed

331416

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

That's my kid in the same place, almost 10 years ago now... Shows the scale a bit better (somehow when you make a photo from below, it always looks smaller than it actually is)

a teeny tiny speck of a kid and a waterfall

000

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Aww I so miss these places, thank you for sharing!!

000

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Haha, yeah absolutely. But again thank you for your answers! It's a case when everyone needs both waffles and pancakes, but no one knows how to make either yet, as batter was invented like yesterday at 18:32, so everyone is double-panicky, their egos hurt, half of all waffle-attempts burned to coals

010

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

...unwittingly, in the process, becoming a storefront for sex trafficking of desperate, ambitioius young women to the States by a few greasy rich folks at the helm and behind the scenes. So tragic on so many levels.

020

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

(I won't even list any examples, you know the examples). And it breaks my heart that Bard became a mirror of that, as an organization! To do so much good, to have so many dedicated, honest, talented people supporting beacons of freedom and education in Russia, Palestine, Central Asia...

110

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

That's the general thing with abusers and predators, elevated to an organizational level. To destroy lives at scale, a sex abuser needs to be talanted, charming, and often kind of... morally good in other things. That's the combination that makes them most effective, and thus most destructive

100

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

But for now I'm struggling a bit as reviewing the PRs is taking the time comparable to developing them (it takes me 2-4 hours to review a PR prepared in 2 days, so 16 hours of work). Which turned me into a bottleneck. And I really don't like that :)

110

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

I'm starting to suspect that I might have made a mistake when trying to combine all three aspects of it: human-facing manuals and intros, agentic note-taking, and design notes, in one wiki system. Or maybe I can still save it by enforcing a better schema on it (either by document, or within docs)

110

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

It works great for development (they get really smart and opinionated), but gradually stopped working as documentation (as agents change it too much, and humans find it hard to compete with them in text curation). And also it makes it hard to review designs, as the texts are too fluid

110

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

To go out on on a limb a bit: in my flow, I try to use an obsidian wiki (zettelkasten) system checked into the same monorepo as the code, to maintain documentation for the project, including forward-facing documentations (plans and designs). It makes agents always aware of the global plan

110

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Hey thank you for responding to _my question_, I do really appreciate your response!!! Please know that the non-appreciation of one account was balanced by a deep appreciation of at least one other!! (mine :) I'm really struggling with this topic right now, so any data points are so deeply welcome!

210

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Epstein sent an aspiring young vocalist he trafficked to meet with Botstein, for him "to provide her with educational guidance". In an email Botstein asks, “Is [she] for me, or are you keeping her for yourself?” (Bard's russian campus was so good... But also trafficked students for sex. No words.)

insidehighered.com

Botstein Resigns as New Epstein Allegations Surface

Leon Botstein has resigned from all Bard College–affiliated positions after House Democrats uncovered new details about his relationship with the late sex offender Jeffrey Epstein and heard whistleblo...

231

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Ha fair! Still I'm kinda glad they didn't one shot solve math, whatever the reasons :)

010

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

True but like am I a negligible minority or are there dozens of us :)) I'm so used to being an absolute minority I'm almost surprised when folks share my opinions haha

100

Arseny Khakhalin @khakhalin.bsky.social · 09/10/2026

Am I the only ai optimist who thinks that seeing ai not being perfect (retracting proofs etc) is actually lovely, interesting, encouraging and inspiring? Kinda like a slower city improvement without instantaneous displacement by speculative investors?

220

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

Also lowest in Europe intergenerational support and one of the highest engagement and time expectations

030

Reposted by Arseny Khakhalin

Michael Druker 🇨🇦 @mdruker.app · 08/10/2026

Under-rated factor I rarely see discussed is that there are lots of cultural and legal pressures over time making parenting more difficult (not just expensive) and little pushing in the other direction

2791

Reposted by Arseny Khakhalin

Haydon, frightfully @haydonmp.bsky.social · 08/10/2026

The only thing that’s had any evidence of actually increasing family size is increasing the amount of remote work done by one or both spouses. It’s not “money” affordability, it’s time and engagement. There’s a much higher expectation for engagement for parents too. The wife and I want five kids.

31237

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

This. This this this!!! It's freakishly expensive, just not in terms of money alone! And it is solvable in principle I believe, if people tried to talk about that, but for whatever reason folks are almost ashamed to talk about it.

010

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

Yeah, that's true :) It's all having Opus explain the code to me. But it's taking _as much time_ as going through the code, and it feels so weird ;) True, a single one-day PR would be like a month of work in the olden days, but still the process feels so... strange. Criteria so annoyingly unclear!

010

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

(Or maybe I suck at writing specifications haha, this is also quite possible.) I even tried to do every PRs first raise docs (designs), then the code, but so far failed to streamline it into a process somehow. The fight continues!! :)

110

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

It feels weird to have development 90% friction 10% development, but that's where my group is now... And because we're developing some non-standard stuff, specifications (before any code is written) a kind of preliminary, as there's still lots of exploration involved

110

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

Thanks a ton!! Yeah, for me, I'm essentially doing adversarial review of their PRs that are already adversarially reviewed, but I'm asking different questions, so getting different answers. And on one hand, it feels reasonable. But on the other, it's still so slow, compared to coding itself!

110

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

Nah, unfortunately in my case it's all fancy business logic automation with weird data schemas, ML and LLM calls, hierarchies of provenance, full trackability and traceability, omg it's a bit overwhelming haha and not agile enough at the moment

110

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

For now mostly I get sessions that end with: > Sleep well, next me. I'll stay out of the notebook while it dreams. Thank you for this morning, Arseny. It was a good one. Which is rather cringe tbh. But the performance will be the judge. We'll see.

110

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

Yeah tbh for now it smells of performance, of a persona of being a persona going a bit overdrive, so I'm not sure I fully buy into it. But we'll see. It pretending to be a person helps my project to generalize trends from small pieces of info (aka generalizing and forgetting), it's worth it :)

100

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

Wait, will there be an eli12 explanation for non-lean people?..

020

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

So what's the current SOTA on the "two eyes" requirement and code architecture alignment for the era of LLM-managed feature development? Is there a current SOTA, or is it still a jagged wild west DIY frontier?

210

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

ps: giving them read this repo, asking if they want to use the vibe from it, but for editing the long-term state rather than within-session context, also seems to get them hyped: bsky.app/profile/timk... We'll see how it works in the long-term :)

030

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

MIne did this: > I'll answer properly at the end. Since you said I can build what I want, and I do want something, I'm building it first. It's small. (it was a better note-managemeent system within CC)

130

Arseny Khakhalin @khakhalin.bsky.social · 08/10/2026

An interesting experiment, if you have a quasi-persistent CC project (no embeddings, no rag, but an encouragment to journal and regular "update your own tooling" sessions, turning it into a more self-adjusting thing): give it this blog post to read, then ask if it wants to borrow anything from it :)

240

Reposted by Arseny Khakhalin

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 08/10/2026

Once again, if you are a senior researcher spreading doom to junior researchers, shut up shut up shut up. It’s your job to figure out a path forward, even in high uncertainty times. Come back when you have something constructive.

1137144

Reposted by Arseny Khakhalin

wwanxin19.bsky.social @wwanxin19.bsky.social · 08/10/2026

Same effect with numbered lists in CLAUDE.md - once an assumption got an ID, Claude treated it as settled and routed around it instead of pushing back. We moved anything still open to plain prose and it started challenging them again.

141

Arseny Khakhalin @khakhalin.bsky.social · 07/10/2026

Most of it was never, never in the tech!.. all the saas friction (and this the source of revenue) is people

000

Arseny Khakhalin @khakhalin.bsky.social · 07/10/2026

Also saying sorry if you lower your performance, makes them take you a bit more seriously and benevolently

030

Reposted by Arseny Khakhalin

crumb @crumb.bsky.social · 07/10/2026

news.ycombinator.com/item?id=4998...

ihoggan
I was a graph theory junkie long ago and even moved to Budapest for awhile to study among the greats. While I was there I started working on Barnette's Conjecture which came to occupy my thoughts over the next 24 years of my life, on and off as I worked in many different fields. Last summer I even thought for a few days that I had actually solved it. But it's supposedly proven here - problem 180. I don't know what to think exactly. I spent thousands of hours on that problem. I really enjoyed it. Hearing that it is solved somehow makes me sad in a far-off way, like hearing an ex-girlfriend died suddenly in a car crash. I don't know, there's probably a lot of people feeling odd emotions tonight. There's no Lean proof for this one so I'm digesting the paper. On the surface it looks like an approach I considered 24 years ago and abandoned.
I revisited the problem this summer, along with my partial solutions, when the previous round of stunning proofs came out. Several hours of work with Fable simply convinced me it wasn't yet solvable and reinforced how hard of a problem it was.

jboggan
I went back to that Fable chat and showed it this new preprint. It coded up the new constructive algorithm and ran it against the existing test suite, that looks good at least.
It has been super helpful in delineating where the crucial concept came from. The proof is rather simple as graph theory proofs go, but it does seem to use some constructions that would only seem obvious if you had serious physics experience with partition function and calculating energy states that cancel out. It's not a wholly alien bolt from the heavens, but I can also see how there hasn't been a human being with the broad theoretical physics knowledge combined with the deep graph theory experience in planar graphs to come up with this idea. I don't know, I'm looking for precedents of this formulation and some old papers of Penrose counting the number of edge colorings of this same graph type are coming up, the line of argument at least rhymes. But I agree with the thought that this sort of progress should not be siloed inside those companies. I propose a tax so that every slop cannon Al video pays for another hour of compute time for advancing mathematics.
reply

823452

Reposted by Arseny Khakhalin

James O'Malley @jamesomalley.co.uk · 07/10/2026

Holy shit. Someone has decompiled the Adobe suite, rebuilt it in Rust and released it as open source. Either all software is dead or the US copyright laws are about to receive an update. getartcraft.com/apps

getartcraft.com

Crafting Apps: open-source creative tools

Image editing, vector illustration, video, photography, PDFs, motion graphics and page layout: seven native, open-source apps from the ArtCraft team, built in Rust and free to use.

19944111242