iiinova — where your digital self keeps evolving

Ferenc Huszár @inference.vc · 09/01/2026

NeoLabs are the new AltCoins, seed rounds the new ICO. DeepMind is BitCoin OpenAI Ethereum Anthropic Solana xAI is DogeCoin There are some that will find specialised niches, others who are actual businesses. Most can't be bothered and are probably mainly Ponzi. Won't stop people investing..

010

Ferenc Huszár @inference.vc · 21/12/2025

Adjustment for living in Britain? :)

000

Ferenc Huszár @inference.vc · 17/12/2025

me: Yay! I'm running my own company now, I am no longer forced to use Microsoft Teams for meetings. also me: *waiting to join the third Teams call of the day, this time with our law firm*

030

Ferenc Huszár @inference.vc · 29/11/2025

I have plenty of colleagues from pre deep learning days who turned “missing out” into a personal goal and core component of their research taste.

040

Ferenc Huszár @inference.vc · 29/11/2025

To be fair, If the missing out part in FOMO is interpreted as continuing to work on something which now looks like very limited impact relative to new areas that opened up where your expertise is useful than pe heaps FOMO is a good thing.

140

Ferenc Huszár @inference.vc · 29/11/2025

Don’t forget the Gary Marcus/Yann Lecun subculture

170

Ferenc Huszár @inference.vc · 20/11/2025

I think GPT now do some new kind of personalization. I once asked it to roleplay as a pirate ghost for a kid’s escape room game, now it writes me Deep Research essays with pirate ghost lingo thrown in. It’s the “Spotify mixing in nursery rhymes in my gym playlist” problem all over again.

010

Ferenc Huszár @inference.vc · 20/11/2025

In the distant future, our European children may rediscover a usable internet after all.

020

Ferenc Huszár @inference.vc · 19/11/2025

Cursor 2.0 is great, but desperately needs background music.

020

Ferenc Huszár @inference.vc · 08/11/2025

Mathematical breakthrough of the year: Financial Times finally calculating the normalizing constant of the uniform distribution over ALL outcomes.

0120

Ferenc Huszár @inference.vc · 08/11/2025

There are pockets of other stuff. ALT/COLT crowd seem to be content continuing to prove/improve bounds. Causal inference with ML is a bit of a thing (although likely pretending/promising to be AI for science for the money).

050

Ferenc Huszár @inference.vc · 08/11/2025

Getting the zucchini right seems to be a nontrivial dependency of this

120

Ferenc Huszár @inference.vc · 04/11/2025

How is this not against the code of conduct?

110

Ferenc Huszár @inference.vc · 01/11/2025

I’m very bad at getting things done but pretty good at making things happen. 🤷🏽‍♂️ How does this work?

030

Ferenc Huszár @inference.vc · 01/11/2025

My favourite vegan cream substitute is pasta water. Not only does it work better for pasta sauces, it reduces risk of death by assassination after your Italian friends learn that you put cream in carbonara…

010

Ferenc Huszár @inference.vc · 26/10/2025

We explain in the supplementary material the process we ran internally (no IRB) There is a nice independent reaction article published in the same issue of PNAS about the ethics of these experiments: www.pnas.org/doi/10.1073/...

pnas.org

Twitter manipulates your feed: Ethical considerations | PNAS

Twitter manipulates your feed: Ethical considerations

130

Ferenc Huszár @inference.vc · 23/10/2025

Take for example the “time to first head” (TTFH) function whose input is a pseudorandom coin flip sequence. This function’s output (pseudogeometric) will be super sensitive to the random seed. Many machine learning experiments behave a lot more like TTFH than empirical averages.

020

Ferenc Huszár @inference.vc · 23/10/2025

Problem is, empirical averages is not how pseudorandom are used. For example: Pseudorandom sequence is used to generate a permutation for stochastic optimisation, or to simulate random moves in a sequential game like chess. The whole trajectory depends critically on what happens at the beginning.

100

Ferenc Huszár @inference.vc · 22/10/2025

Thinking Machines co-founder Andrew Tulloch reported/rumoured to receive £3.5bn pay package over 5 years to rejoin Meta. To put this into perspective, this would approximately cover Armenia's entire education budget (over the next 5 years). Not accounting for income/capital gains taxes of course.

030

Ferenc Huszár @inference.vc · 22/10/2025

Just realised how hopeless and inadequate it feels when - accidentally - I google for information instead of asking ChatGPT (equivalent). Context: I was looking for a bedtime reading book that is complex, for preteens but structured as a collection of stand-alone episodes rather than one long novel

000

Ferenc Huszár @inference.vc · 14/10/2025

I can certainly imagine causing harm with all of these. But then by this logic Microsoft Teams and JavaScript should also be on the list🤷🏽

0120

Ferenc Huszár @inference.vc · 11/10/2025

Is Pope Leo giving the NeurIPS keynote this year?

081

Ferenc Huszár @inference.vc · 10/10/2025

When I applied to PhDs, I could derive the E-M algo after 5 yrs of uni and I felt I stood out. Many applicants didn’t do much ML before. This year, Miranda, barely out of high school starts her undergraduate with a NeurIPS spotlight, two AI olympiad medals, having bern to EEML 2x plus AI retreat 😨

0141

Ferenc Huszár @inference.vc · 08/10/2025

I’m curious what should/could they have done which in hindsight would have worked?

100

Ferenc Huszár @inference.vc · 08/10/2025

This seems to assume NGOs in those countries actually developed any productive counterstrategies or actionable insights. My impression in Hungary is that they pretty much barely hold on. I’m not sure what there is to learn other than getting a preview of what’s coming.

100

Ferenc Huszár @inference.vc · 03/10/2025

@nsaphra.bsky.social thank you f or your service, keep up the good work

210

Ferenc Huszár @inference.vc · 27/09/2025

I like the sound of this common EU corporate entity idea! Except quite clearly the virtual state shouldn’t be called “regime 28” but “regime 0” or failing that, “regime -1”. Open for consultation until 30 Sept so take a look! www.eu-inc.org

eu-inc.org

EU–INC — One Europe. One Standard. — Pan-European legal entity.

EU–INC is a proposal for a pan-European standardized legal entity to unlock pan-European startup scaling.

051

Ferenc Huszár @inference.vc · 26/09/2025

Like a phoenix rising from the ashes, like a turd that wouldn’t flush, I re-emerge again in the geriatric Millennial basketball scene. Wish me good health.

050

Ferenc Huszár @inference.vc · 18/09/2025

Mom, can we watch Star Wars? Mom: We have Star Wars at home Star wars at home:

000

Ferenc Huszár @inference.vc · 10/09/2025

We will have GPT6 before GTA6

0122

Reposted by Ferenc Huszár

Terence Tao @teorth.bsky.social · 18/08/2025

I wrote an op-ed on the world-class STEM research ecosystem in the United States, and how this ecosystem is now under attack on multiple fronts by the current administration: newsletter.ofthebrave.org/p/im-an-awar...

newsletter.ofthebrave.org

I’m an award-winning mathematician. Trump just cut my funding.

The “Mozart of Math” tried to stay out of politics. Then it came for his research.

18791324

Ferenc Huszár @inference.vc · 08/08/2025

OpenAI automated HR assistant

040

Ferenc Huszár @inference.vc · 21/07/2025

Generalised second price auction for JavaScript imports?

020

Ferenc Huszár @inference.vc · 21/07/2025

Arrived in Sarajevo for @eemlcommunity.bsky.social Located in a narrow valley, it’s an ideal setting to get the momentum going in my deep learning talk later. I’m keen to explore beyond the hotel coffee though which, one assumes, is carefully curated to highlight floral notes of a morning cigarette

060

Reposted by Ferenc Huszár

Annabelle Michael Carrell @ab-carrell.bsky.social · 19/07/2025

So you want to skip our thinning proofs—but you’d still like our out-of-the-box attention speedups? I’ll be presenting the Thinformer at two ICML workshop posters tomorrow! Catch me at Es-FoMo (1-2:30, East hall A) and at LCFM (10:45-11:30 & 3:30-4:30, West 202-204)

054

Ferenc Huszár @inference.vc · 16/07/2025

BREAKING: OpenAI point guard Jason Wei joins Meta in shocking trade news. OpenAI receives two ConvNet researchers and a 2027 first-round draft pick. Veteran Yann LeCun rumored to be missing another season due to slow Achilles recovery. Apple “cautiously optimistic” about G-league chances.

2411

Reposted by Ferenc Huszár

Annabelle Michael Carrell @ab-carrell.bsky.social · 14/07/2025

Your data is low-rank, so stop wasting compute! In our new paper on low-rank thinning, we share one weird trick to speed up Transformer inference, SGD training, and hypothesis testing at scale. Come by ICML poster W-1012 Tuesday at 4:30!

2257

Ferenc Huszár @inference.vc · 14/07/2025

Low-Rank Thinning is presented tomorrow at ICML ⏰ Tue 15 Jul 4:30 - 7 p.m. PDT icml.cc/virtual/2025...

icml.cc

ICML Poster Low-Rank ThinningICML 2025

020

Ferenc Huszár @inference.vc · 12/07/2025

Off to ICML next week? Check out my student Annabelle’s paper in collaboration with @lestermackey.bsky.social and colleagues on low-rank thinning! New theory, dataset compression, efficient attention and more: arxiv.org/abs/2502.12063

arxiv.org

Low-Rank Thinning

The goal in thinning is to summarize a dataset using a small set of representative points. Remarkably, sub-Gaussian thinning algorithms like Kernel Halving and Compress can match the quality of unifor...

0115

Ferenc Huszár @inference.vc · 11/07/2025

They solved the alignment problem

1141

Ferenc Huszár @inference.vc · 10/07/2025

Ah, what I meant is restricting inputs to A!=B, so say I don’t care about the diff between > and >=. But in any case I will read the symmetry argument properly.

010

Ferenc Huszár @inference.vc · 09/07/2025

Career highlight for sure.

010

Ferenc Huszár @inference.vc · 09/07/2025

Yes, and he didn't mind. I read it somewhere that he doesn't like men except Obama, so I felt pretty proud.

110

Ferenc Huszár @inference.vc · 09/07/2025

There once was a good man called Frederik He landed a sweet job in Limerick For fifty-six grand He would teach on command "Entschuldigung", "Tschüss" and "Ich liebe dich"

0100

Ferenc Huszár @inference.vc · 09/07/2025

I don't think we disagree. You're saying there is no guarantee, and that it is not hard to come up with examples demonstrating it fails. I say there's no guarantee, which is why it's very surprising and intriguing that there are non-trivial examples where it does not fail.

100