I understand the sentiment behind this, but I'm just extremely bearish on rankings that do fancy mathematical tricks or use black-box algorithms because the more of that you do, the more you can bias it towards a specific outcome.

The closer the metric is to the raw data instead, the better.

Two issues: first, this is not transparent. There's NO way to tell *which* papers were counted, nor how 'most important papers to this paper' is computed. The papers contributing to the ranking should be listed.

Second, CSRankings is CC BY-NC-ND 4.0 so I'm pretty sure this is copyright infringement

I can't believe I spent the evening skimming through every Visual Reasoning paper at ICLR instead of finishing my SoP.

It may be in bad taste to call out a reviewer like this, but I don't believe anyone who gives weaknesses like this, as if it isn't empirical common sense for anyone working with MLLMs that larger models give better outcomes when inference speed isn't relevant, is acting in good faith.

Tbh, it's better to go through Yann LeCun's lectures on JEPA first before trying to comprehend the mathematical foundations of LeJEPA, since this is a specific optimization on top of the original JEPA theory.

The lecture notes were easier to work out initially.

Google DeepMind also dropped something extremely similar (SIMA 2), trained on different games rather than Genshin, but they didn't release their technical report yet so I'm valuing this higher

Ahh, I'd forgotten that every case of GenAI "deepfakes" on celebrities in the last couple years have ended up with nothingburger consquences or been tied in up court over vague legislation 😖

Hopefully at least one reviewer at CHI (going off the timing) notes this while reading it.

Even if they didn't care about privacy, they should've censored the images on the basis of the fact that if the real person requests a DMCA takedown of their photos the paper itself will also be hit.

I found a typo in my master's thesis three months after it went on Proquest 😭

Lowkey it's so unfair that VIZ folks get to make beautiful websites as a side effects of their work 🥺

I saw soo many amazing websites in the last few months looking at accessible visualization folks. I didn't even keep track, but I should go back and find my favourites again.

(There is One Impostor* among these)

* a conference I'll bet money on being predatory. The rest are completely legitimate but niche or local.

Don't forget being a conference ranking organization on the side!

Would a position paper accepted at conferences like PRICAI, ICCIT, CSITA, EICC or CHR be accepted?

C'mon, this is a 14 year old asking about a major event that took place when he was 9 years old that his parents sheltered him heavily from.

Considering he's likely interacting with a lot of 15-20 year olds who were more cognizant during COVID and keep referring back to it, it's a legit question.

And, glancing at the list of tasks, most of the ones GPT-5 missed, even if not good *enough* today, are just... pretty much solveable via engineering?

The funniest thing about this is Gary Marcus put his name on it even though we all know if an LLM-based approach, already at 60%, ever manages to reach 100% he'll still refuse to accept its AGI-hood.

Yes, yes, you're all very much welcome for the em dashes.

you should live vicariously through playing Subway Builder instead