iiinova — where your digital self keeps evolving

Reposted by Christian Wolf

hailey @hailey.at · 9h

he cannot be trusted outside so he gets the harness

orange cat in a harness sitting in the rising sun

81334

Reposted by Christian Wolf

Antoine Emond 🇫🇷 🇸🇪 @aemond.wsocial.eu · 10h

Le Chonk is here 🐱 Mistral's Large 4 is a 1 trillion parameter model trained in its own European data centres, with open weights coming on 27 October. Looking at the benchmarks, it's up there with the best open models. It even does a bit better than GPT-6 Astra on some aspects. Bien joué, Mistral !

1133

Christian Wolf @chriswolfvision.bsky.social · 10h

Procedural Core: A Compact Recurrent Initialization for Vision Transformers By @damienteney.bsky.social 's group Idea: train a ViT with parameters tied over layers (="recurrent ViT") on abstract procedurally data. Then initialise models of ANY arch (!) with these weights. arxiv.org/abs/2609.37631

051

Christian Wolf @chriswolfvision.bsky.social · 11h

You mean the ECCV'26 work involving also M. Black? I don't think there is any relationship, neither in application (we do map-free pose estimation and streaming VQA) nor in technique (we distill context transformers into recurrent transformers). Here Blender has only be used for the illustration.

000

Christian Wolf @chriswolfvision.bsky.social · 15h

Having fun with Blender ... Distilling context transformers into recurrent transformers: arxiv.org/abs/2606.21562 @weinzaepfelp.bsky.social @mbsariyildiz.bsky.social

1205

Reposted by Christian Wolf

Noa Garcia @noagarciad.bsky.social · 22h

I am running a study on how computer vision researchers perceive dual-use risks of the field. If you work in CV, and would like to help us, the survey takes less than 3min. Reply or DM me for the link. You will be helping us a lot! Reposts are much appreciated 🙏

31011

Christian Wolf @chriswolfvision.bsky.social · 05/10/2026

I don't agree with this. This is not about being an exam, but science, and science requires clear communication. Contributions must be understandable and reproducible. This is not about gate keeping but keeping up standards.

100

Christian Wolf @chriswolfvision.bsky.social · 05/10/2026

I think lots of the views will simply evolve, since society evolves and AI evolves. Personally I still think the LLMs are incredibly useful as coding agents but (for the moment) horrible as (standalone) writing tools. Also, I currently refuse to read AI generated documents sent to me.

100

Christian Wolf @chriswolfvision.bsky.social · 05/10/2026

I have only 3 but now I receive invites from ACs to receive additional papers.... ?!?

020

Reposted by Christian Wolf

NeurIPS Europe @neuripseurope.bsky.social · 05/10/2026

Our second call for NeurIPS Paris 2026 volunteers will open tomorrow TUE Oct 6th, 2pm CET. We are still seeking for PhD students and postdocs to assist during the conference in exchange for complimentary registration. Application form: forms.gle/bBFUyUNngtCc...

143

Christian Wolf @chriswolfvision.bsky.social · 05/10/2026

I have skimmed over my 3 and ... same but w/o theorems.

110

Christian Wolf @chriswolfvision.bsky.social · 04/10/2026

Just checked the 3 papers I need to review for ICLR 2027. Topics: perfect match. The paper IDs however ... we were warned but it's still a shock when you see them.

240

Christian Wolf @chriswolfvision.bsky.social · 03/10/2026

I read that a certain president asked the Chatbot of a certain billionnaire supporter how certain countries would react to an invasion. I wonder when it will be discovered that it makes sense to have system prompts specific for this user.

030

Reposted by Christian Wolf

worm @rfksbrainworm.bsky.social · 02/10/2026

If you ever wondered why visible light is those particular wavelengths, it's because life evolved to see in the narrow range where water is transparent

Diagram of absorption of emf by water. There's a gap right that dips in the 500 nm range

10260611700

Christian Wolf @chriswolfvision.bsky.social · 03/10/2026

Ok, je vais tester ça, merci 😀 J'adore les champignons.

000

Christian Wolf @chriswolfvision.bsky.social · 03/10/2026

Hmm ... ça ne crame pas les champignons?

100

Christian Wolf @chriswolfvision.bsky.social · 03/10/2026

The pope being Schmidhubered was not on my bingo card.

190

Reposted by Christian Wolf

arXiv @arxiv.bsky.social · 01/10/2026

arXiv has updated our policy on rate limiting for all submitters. This update was made to fairly distribute moderator time & support the arXiv community of staff, volunteers, readers & authors. Please read our announcement to learn more: blog.arxiv.org/2026/10/01/updated-r…

216874

Christian Wolf @chriswolfvision.bsky.social · 01/10/2026

I think we can sometimes escape the English mono culture, yes.

000

Christian Wolf @chriswolfvision.bsky.social · 01/10/2026

Same. I left the house this morning at 6am and cycled in a T-short, on Oct 1st. Interesting. But I do remember a mid-November day in 2006 where I was at 1500m in the Alpes in a T-Shirt...

100

Christian Wolf @chriswolfvision.bsky.social · 01/10/2026

Je sens que Bluesky est pret pour des memes sur la causalité.

5235

Reposted by Christian Wolf

joel @jmoonz.bsky.social · 30/09/2026

LLMs are stochastic parrots in the same way humans are a bag of meat. I.e. a reductive description that has almost no utility in describing the system.

1585

Christian Wolf @chriswolfvision.bsky.social · 01/10/2026

Somebody is wrong on the internet again? 😉

020

Christian Wolf @chriswolfvision.bsky.social · 30/09/2026

Went cycling to escape AI. Crossed 2 cyclists 🚴‍♂️ 🚴‍♀️ and overheard "So with AI this has all now become much easier ..." 😵‍💫

0151

Christian Wolf @chriswolfvision.bsky.social · 30/09/2026

I have ideas for a boys band or a girls band (Posh spicy = Claude, Baby Spice = Claude, Sporty Spice = Claude ...)

020

Reposted by Christian Wolf

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 30/09/2026

Putting together a squad for the heist of a century. There's the hacker (Claude), the financier (Claude), the demolition expert (Claude), the inside man (Claude), the driver (Claude)

810811

Reposted by Christian Wolf

AI Conference DL Countdown @dlcountdown.bsky.social · 30/09/2026

AAMAS'27 (reg): 2 days + 2h. AAMAS'27 (paper): 9 days + 2h. CVPR'27 (reg): 42 days. CVPR'27 (paper): 48 days.

042

Christian Wolf @chriswolfvision.bsky.social · 30/09/2026

There was a repeater installed which ran on a different protocol and had configured itself to have its own double DHCP requests and packets were sent back and forth. Also, the repeater had an older protocol running. Anyway, Claude logged into the router itself and solved all this.

110

Christian Wolf @chriswolfvision.bsky.social · 30/09/2026

I now have 3 to 5 cases a day where I need an answer to a question and normally it would require a lot of manually deep diving into complex stuff, mechanical boring cognitive work, which Claude solves in minutes and autonomously. - Can't log into git - Question on 2y old code - My Wifi is slow

2111

Christian Wolf @chriswolfvision.bsky.social · 29/09/2026

Making space probes autonomous with LLM-based agents ...

020

Christian Wolf @chriswolfvision.bsky.social · 29/09/2026

The list of products that Claude & Co now have made actually usable: ffmpeg matplotlib git (beyond commit/pull/push)

3270

Christian Wolf @chriswolfvision.bsky.social · 29/09/2026

I am actually not against receiving cold emails and I actually read (well, start to read ...) all of them. So, if a very promising candidate writes me, I will reply. What I absolutely don't understand is the two cases in my first post: no relationship to our work, and fully automated processes.

100

Christian Wolf @chriswolfvision.bsky.social · 29/09/2026

Yep. And now in most cases they follow up with a second email a couple of days later. Just in case that would be helpful ...

220

Christian Wolf @chriswolfvision.bsky.social · 29/09/2026

2020: cold emails with applications mentioning unrelated areas like Biology. Candidate didn't even check my/our webpage. 2026: cold emails which relate the candidate's CV with our work and latest papers. 100% LLM written, generic LLM language. Same effect, same probability of getting a reply.

2153

Reposted by Christian Wolf

Phillip Isola @phillipisola.bsky.social · 28/09/2026

Students sometimes ask me if it still makes sense, in this accelerating age, to pursue a PhD in AI. Perhaps counterintuitively, I think it's a great time to do so. I wrote up some thoughts on this here: web.mit.edu/phillipi/www...

web.mit.edu

On the Value of Doing a PhD in the Age of AI

18116

Christian Wolf @chriswolfvision.bsky.social · 27/09/2026

This is one of the craziest color schemes I have ever seen. Pink to blue to yellow red is circular and confusing.

000

Reposted by Christian Wolf

AI Conference DL Countdown @dlcountdown.bsky.social · 27/09/2026

The ICLR DL has passed but 60k registered papers doesn't mean that research stops, the deadlines keep coming 😀 AAMAS'27 (reg): 5 days + 2h. AAMAS'27 (paper): 12 days. CVPR'27 (reg): 45 days. CVPR'27 (paper): 51 days.

151

Christian Wolf @chriswolfvision.bsky.social · 26/09/2026

Enjoy! 😍 Would like to do the same but I am far from my home and I don't have my bike with me ...

020

Christian Wolf @chriswolfvision.bsky.social · 26/09/2026

Go go go! We submitted yesterday evening 😀

010

Christian Wolf @chriswolfvision.bsky.social · 25/09/2026

À great bike, congrats ! Yes bike shorts are almost mandatory 😀

110

Christian Wolf @chriswolfvision.bsky.social · 25/09/2026

I don't know but we need the discussion and the competition. It must be clear that this can't be just the new standard.

020

Christian Wolf @chriswolfvision.bsky.social · 25/09/2026

In the meantime, back on the hellsite 😬

020

Christian Wolf @chriswolfvision.bsky.social · 25/09/2026

This is IMO not a valid argument. First, this is not a free service, the community pays it through conference registrations (+donations, I know). Second, and more importantly, we cannot have leaks, we cannot have repeated leaks, it is not acceptable. If they can't fix it, we need to move away.

240