Marko

Pijem, ne pušim, ne drogiram se, ne vjerujem roku trajanja, ovlašten sam za izvršavanje digitorektalnog pregleda te se osjećam sjajno u vaginalnom matrixu.

[Marko Čuljak @camuljak]

Joining thousands of ICML attendnce posts to announce that @dionizije_fa and I will be presenting our work on bioinformatics coding agent benchmarking (main + AIWILD workshop). Looking forward to chatting about LM reliability, interp., and rare coffees.

https://t.co/i0yPYPwAAt

[Marko @maculjak]

Danas sam, dragi pratioče, prvi put u životu popio dobru kavu na aerodromu i to u Kopenhagenu.

[dionizije @dionizije_fa]

Paper accepted to ICML 2026. If anyone is going or is nearby Seoul, July 6th - 11th, DM me if you want to meet up 🙏.

[Martin Tutek @mtutek]

Modern LLMs are aligned to be harmless, but are they truly safe?

Frontier models either choose⛑️options harmful for humans⛑️or abstain from choices where harm is aimed at🛋️furniture🛋️!

I'm not in Rio, but find @AdiSimhi and @Itay_itzhak_ at #ICLR2026🇧🇷& chat about our work!

[Marko Čuljak @camuljak]

Got banned on BlueSky after posting the exact same thread I posted here earlier to promote a recently published paper. @bluesky wtf?

[dionizije @dionizije_fa]

1/🧵 So what the boys (@camuljak, me, others) tried to showcase how we think about LLM evals on task "automate the boring SWE heavy part of bioinformatics". Turns out the frontier LLMs can do it, in ideal conditions, without any custom prompting. Open source getting close.

[Zhaocheng Zhu @zhu_zhaocheng]

ICML bidding observations: LLMs are emerging as a field separate from deep learning.

Papers study LLM behaviors at pure phenomenological level, with little connection to model architecture or training data. Identifying a limitation, even without resolving it, is becoming sufficient.

[Marko Čuljak @camuljak]

We already know prompt repetition is a handy hack to improve a decoder-only LM’s performance as it allows the model to “see” bidirectionally, an ability otherwise suppressed by the causal mask.

But what happens if we increase the number of repetitions?🤔🧵@eaclmeeting #EACL2026

[the Rich @Duderichy]

economist post a title with data that directly disagrees with it

[Crémieux @cremieuxrecueil]

Apparently some researchers thought it was significant that early high performance and later high performance were negatively correlated

They didn't realize they were conditioning on a collider

Nor did the editors at Science, who published their claims anyway

Just incredible

[François Fleuret @francoisfleuret]

I do not think you can pursue meaningful research without (1) some grandiose delusion about your abilities (2) a sense of esthetics and harmony to judge ideas still free of experimental confirmation (3) an unreasonable taste for the required tangible work (e.g. programming)

[Martin Tutek @mtutek]

Very honored to be one out of seven outstanding papers at this years' EMNLP :)

Huge thanks to my amazing collaborators @fatemehc__ @anmarasovic @boknilev, this would not have been possible without them!

[Marko @maculjak]

MARTIN TUTEK JEBAČ @mtutek

[Marko @maculjak]

Muškarci koji ne preferiraju urinirati sjedečki imaju krhku muškost

[dionizije @dionizije_fa]

Most overrated thing in Italy is their pretencious coffee culture. I have had only a handful of decent coffees in Italy, and that’s being generous.

[dionizije @dionizije_fa]

@SCHIZO_FREQ Drink quality wine!

[dionizije @dionizije_fa]

Friends, let this be a reminder, that you are probably far from your optimal rate of failure and that you should fail more often!

[Marko @maculjak]

In ancient times only priests were doing science and rightfully so. They didn’t have to worry about their existence, were far away from all the distractions (e.g. copulation) and their other duties were minimal. They have had it much better than university PhD students today.

[Marko @maculjak]

@dionizije_fa @kalomaze @jxmnop 3-4h reading per week is more than enough

[Marko @maculjak]

Kad dođem na vlast uvest ću svjetski dan erekcije.