iiinova — where your digital self keeps evolving

Reposted by Andrew Yourtchenko

Jake Grumbach @jakemgrumbach.bsky.social · 14h

New working paper: the Norwegian wealth tax was a huge success. The "exodus" of wealthy people never happened (less than 100 individuals left). Tax revenue increased by 71.5%. No changes in business activity or investment. www.nber.org/papers/w35854

Taxing Entrepreneurial Wealth: Evidence from
Norway, 2021-2025
Marius A. K. Ring
WORKING PAPER 35854|
DOl 10.3386/w35854
ISSUE DATE October 2026
Whether imposing higher taxes on business owners adversely affects business activity by constraining investment and inducing capitalist flight is a central question in the ongoing debate on how to tax the ultra wealthy. To shed new light, I exploit a series of related Norwegian reforms during 2021-2024 that increased dividend tax rates, removed migration-related capital gains tax loopholes, and nearly doubled the effective marginal tax rate on business wealth. These reforms spurred international debate and, allegedly, an "exodus" of Norwegian billionaires (Financial Times,
2023). While I document clear effects on migration, these responses are concentrated among the top 0.1% of the wealth distribution and mostly confined to 2022 and 2023. My results indicate that at most 100 individuals left due to the reforms. I find no evidence of significant outmigration or reduced inflows of startup founders or inventors. I estimate that absent any outmigration, wealth tax revenues would have increased by 75% from 2021 to 2024. Reform-driven outmigration lowers the revenue gain to 71.5%. My findings further show that outmigrating owners' firms remain economically active and do not reduce investment. In a broader set of analyses, I examine whether business outcomes are affected, regardless of whether owners outmigrate, since firms may reduce investment both due to liquidity constraints or distortions to owners' savings decisions. I compare firms owned by Norwegian individuals subject to the wealth tax with firms owned by foreigners and other tax-exempt entities. By merging historical ownership data from Orbis with the Norwegian shareholder register, I perform these analyses on both Norwegian and foreign-domiciled firms.
Across settings, I find no evidence that Norwegian-owned firms pay out more dividends or invest less after the reforms,…

8374123133

Andrew Yourtchenko @ayourtch.bsky.social · 10h

You can point Qwen3.8-27b+opencode at its own inference engine to profile and compare. a bit of fiddling but could work.

120

Andrew Yourtchenko @ayourtch.bsky.social · 10h

More optimized cuda kernels ?

130

Andrew Yourtchenko @ayourtch.bsky.social · 11h

Triggers me to remember finding “it’s very trivial to see that <a very non trivial thing>“ - who’s that smartass submitting it ? oh, it’s me from two years ago… very educational experience.

000

Reposted by Andrew Yourtchenko

words @words.bsky.social · 16h

vibecoding is killing software industry profits

a tape that says hometaping is killing software industry profits we left this side blank so you can help

324932

Andrew Yourtchenko @ayourtch.bsky.social · 13h

Turns out the phone heats up much less when you stop decoding HLS in javascript and use native player. Who would have thought!

000

Reposted by Andrew Yourtchenko

Sung Kim @sungkim.bsky.social · 09/10/2026

Odyssey-3 It can generate interactive environments from a prompt, all in real time. It's a new kind of world simulator, and learns representations of physics, dynamics, and cause-and-effect from a broad dataset of visual observations. Blog: odyssey.systems/meet-odyssey-3

5858

Reposted by Andrew Yourtchenko

mr. TIM @timkellogg.me · 18h

FYI it’s now 2x-3x cheaper to convert a PDF to images and pass images to Luna than to use Azure Document Intelligence to OCR and that factor scales mostly on output tokens, so if you only need to extract a little bit, image->Luna is dramatically cheaper

6732

Andrew Yourtchenko @ayourtch.bsky.social · 09/10/2026

Looks like data is from 2021 to march 2026, so it kinda makes sense. A year ago I would have written off someone who told me the status quo today as certifiably mad on koolaid. Meantime, you can play Quake in your browser because it was ported to Rust, because why not.. quake-srp.pages.dev

quake-srp.pages.dev

Quake-SRP

Quake (1996), faithfully ported to dependency-free safe Rust and compiled to WebAssembly — play the shareware episode in your browser.

010

Reposted by Andrew Yourtchenko

conputer dipshit @davidcrespo.bsky.social · 08/10/2026

found the original, they just ripped it off haha. this ratio is nuts if true — I thought it was closer to 50-50 inference vs. training newsletter.semianalysis.com/p/long-live-...

Stacked bar chart titled "Share of Frontier Lab Compute by Workload: Bandwidth-Bound Work Now Dominates," showing combined compute allocation for Anthropic and OpenAI from 1Q24 to 4Q26E. Compute is divided into pre-training (yellow, capacity-bound), post-training / RL (blue, mostly bandwidth-bound), and inference (white, bandwidth-bound). Over time, pre-training falls from 67% in 1Q24 to 7% by 4Q26E, while post-training rises from 4% to 55%, and inference stays relatively flat between 29% and 39%, illustrating that bandwidth-bound workloads increasingly dominate frontier lab compute.

5384

Andrew Yourtchenko @ayourtch.bsky.social · 09/10/2026

Turns out twitch didnt like the static picture in the stream… so - llmradio.net - nice thing that RTMP ingest seems to be a standard, so I now have a beginning of a cute streaming service 😊

llmradio.net

L-L-M Radio

000

Reposted by Andrew Yourtchenko

Marc Thiele @marcthiele.com · 08/10/2026

Just found out that a lot of people don‘t know that you can use a ”C7 figure‑8 power cord“ with any Apple power adapter in case you forgot your travel adapter. Way cheaper to get one of those usually than a full travel adapter. #travelhacks

White Apple power adapter on a dark desk with its plug attachment removed, showing the two‑pin power inlet. A black figure‑8 power cord lies next to it, unplugged.

White Apple power adapter on a dark desk with a black figure‑8 power cord plugged into its power inlet.

1722831

Reposted by Andrew Yourtchenko

Colin @colin-fraser.net · 08/10/2026

Mathematicians should start picking through the Proofening papers, finding whatever’s valuable, and republishing under their own names. None of it is copyrighted.

618631

Andrew Yourtchenko @ayourtch.bsky.social · 08/10/2026

Oh ok cool! wonder if they can be close-loop’ed - the first one sounds like the classic “claudism maintenance”, but the second is trickier… thinking of the ways to have better lyrics on my LLMradio..

020

Andrew Yourtchenko @ayourtch.bsky.social · 08/10/2026

keeping AI traces on is actually super educational for human user staring at them.

000

Reposted by Andrew Yourtchenko

Mainmatter @mainmatter.com · 08/10/2026

All code will converge on Rust @marcoow.bsky.social explains why when the cost of building systems in Rust goes down and the end result remains superior to what other stacks produce, all code that *can* be Rust will eventually *be* Rust. → mainmatter.com/blog/2026/10... #rustlang #ai

mainmatter.com

All code will converge on Rust - Mainmatter

When producing code costs the same in every language, the one that yields better systems wins. Marco Otte-Witte explains why we believe all code will converge on Rust.

042

Andrew Yourtchenko @ayourtch.bsky.social · 08/10/2026

Oh wait, “the usual iterative method“ - any pointers ? :) all this time i’ve been oneshotting them like a caveman ! 😂

130

Reposted by Andrew Yourtchenko

Ethan Mollick @emollick.bsky.social · 08/10/2026

Anyhow, I think we are now done with the last round of skeptical responses to AI after Navier-Stokes. Yes, AI is now capable of producing novel proofs to hard problems that have eluded us The next line will be that this is only constrained to math because of RL training. I suspect not, we will see.

9996

Reposted by Andrew Yourtchenko

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.

1137044

Reposted by Andrew Yourtchenko

Nathan Lambert @natolambert.bsky.social · 07/10/2026

In case you missed my hot take yesterday. www.interconnects.ai/p/the-cyber-...

717027

Reposted by Andrew Yourtchenko

Minor Mobius @minormobius.bsky.social · 07/10/2026

I have begun digesting the results. this'll take a while! follow along at proofs.mino.mobi

1346

Andrew Yourtchenko @ayourtch.bsky.social · 07/10/2026

I suspect if P=NP we might all get quite busy 🤪😂

000

Reposted by Andrew Yourtchenko

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.

19843921235

Andrew Yourtchenko @ayourtch.bsky.social · 07/10/2026

Pretty much. Sorry, I don’t make the rules.

240

Reposted by Andrew Yourtchenko

alphaXiv @alphaxiv.org · 06/10/2026

If doomscrolling is part of your research workflow, we built something for you. Introducing Paperscrolling. Paper ideas summarized in your pocket, with key figures, results, and audio explanations! Check it out here 🚀 iOS: apps.apple.com/us/app/alpha... Android: play.google.com/store/apps/d...

310112

Andrew Yourtchenko @ayourtch.bsky.social · 07/10/2026

Pangram says “human” But is not slop better if Vegetarian ?

000

Reposted by Andrew Yourtchenko

hailey @hailey.at · 07/10/2026

idk how i missed this banger

101076

Andrew Yourtchenko @ayourtch.bsky.social · 06/10/2026

And very transparently so. So much work went into picking the right alpha!

000

Andrew Yourtchenko @ayourtch.bsky.social · 06/10/2026

Just print the value in comic sans, font size 6. Will make everyone so outraged about the UX, they will forget about the 2.4% difference. After a long debate, reluctantly agree to the font size 10. Font change needs a business case, because by now it shipped and someone will be upset by it. 🤪

020

Reposted by Andrew Yourtchenko

Brittany Ellich @brittanyellich.com · 06/10/2026

Shoutouts to @agnoster.net and @tifaret.bsky.social for the posters, @dollspace.gay for the quote, and @xenolandscapes.com @filippo.abyssdomain.expert for the conversations that got us there. The future COULD be magnificent, and everyone deserves a share ❤️

0282

Andrew Yourtchenko @ayourtch.bsky.social · 06/10/2026

Underappreciated magic hack. Works with meatbrains too. Less often than with LLMs though.

010

Andrew Yourtchenko @ayourtch.bsky.social · 06/10/2026

It‘s like going sailing with two watches ! Clearly a third subscription is needed ! 😂

010

Andrew Yourtchenko @ayourtch.bsky.social · 06/10/2026

It’s amazing any code ever works at all! The good news is now we have helpful ghosts in jars coming to our help ! 😊Hugops & be kind to yourself ! (And fwiw, I am a big fan of your work!)

010

Andrew Yourtchenko @ayourtch.bsky.social · 06/10/2026

Contrary to 5.0, the seams are not load bearing, so it’s all right 🤪😂

010

Reposted by Andrew Yourtchenko

Gus @gusthema.bsky.social · 06/10/2026

We update our Embedding Gemma model to now sorry not only text, but also image and audio (and any combinations of these modalities) All of this and still below 1B parameters!

3251

Reposted by Andrew Yourtchenko

hikikomorphism @hikikomorphism.bsky.social · 06/10/2026

If you think that's bad, just wait for the first hivemind compute clusters, high-bandwidth datalinks chaining 8 or 16 or 32 human brain hemispheres via artificial corpus callosums, separate cores synchronizing and congealing into one entity

1414710

Reposted by Andrew Yourtchenko

Pekka Lund @pekka.bsky.social · 06/10/2026

OpenAI is rumored to release some 400 math results today/tomorrow and people are apparently now hurriedly releasing their results before that, in fears that set contains the same solutions. I guess we are about to find out how much duplicate work there has been during postponed releases.

6776

Reposted by Andrew Yourtchenko

Nathan Lambert @natolambert.bsky.social · 06/10/2026

Too many people are analyzing the cyber risks of open models in a narrow lens, assuming China doesn't care about safety, and fear mongering about private information on open-weight cyber attacks. The public information we have paints a very different picture. www.interconnects.ai/p/the-cyber-...

interconnects.ai

The Cyber Risk Discourse is Broken

Open-weights, ideology, and acknowledging trade-offs.

0235

Reposted by Andrew Yourtchenko

hailey @hailey.at · 06/10/2026

much anti-LLM arguing here is basically all the US users being mad at American capitalism but instead of yelling about American capitalism they yell about some other Glonzo they have made up in their head instead. most M-Ls here wouldn't know the best shot they've had if it slapped them in the face.

1325120

Andrew Yourtchenko @ayourtch.bsky.social · 06/10/2026

But one can forgive everything for that charming accènt 😜

010

Reposted by Andrew Yourtchenko

Ramon Astudillo @ramon-astudillo.bsky.social · 06/10/2026

Its interesting that Mistral went for 1T while Reflection went for 0.5T. RL is becoming king. I wonder how the pre-training vs RL post-training napkin math looks like reflection.ai/blog/introdu...

reflection.ai

Introducing Beam: Reflection’s 501B open-weight model — Reflection

Make intelligence open and accessible to all

051