I was thinking of sharing them publicly on my blog but then realized it could be problematic / misunderstood / misused, so for now I refrained from it... but yes I can send them privately. I'll email you later.

We regularly use LLMs to review our own papers before submission. I agree that in principle it's hard to "prove" that a review is entirely LLM-generated vs. LLM-written... but in practice it's quite easy especially when you see multiple "LLM-smelling" points in a review.

AISTATS adopted the iron fist against "LLM-generated" reviews and I fully agree with it.

Hopefully we can figure out a way to allow responsible, "LLM-assisted" reviews where one genuinely uses the LLM to assist (not replace) the reviewer, which can be a force multiplier.

Even just the fact that we have oral tests (throughout school and then university) shocks a lot of people.

It's easy to fall into the fallacy "I went through this system so it is obviously the best system" and surely it has downsides but tbh it does prepare you for a lot of things in life...

I fully agree. I have been thinking a lot about this and my current advice is use the standard error of the median (bootstrapped), with a separate metric for reliability if needed. The latter is less standard, we have been reporting e.g. 80% performance quantile, and std err of that (bootstrapped)

All correct. A steelman argument can be made that if you are comparing methods you also may care about the actual dispersion which is somewhat related to reliability. But even then you‘d probably best showcase other metrics focusing on that. And yes ML would benefit from more stats.

THATS TEH SPIRIT!!!

Kudos to fellow co-organizers! @desirivanova.bsky.social
@conorhassan.bsky.social Cen-You (Scott) Li

Short abstract submission is open (DL Oct 16, 2025 AoE)
Accepted abstracts will be presented as posters.

🌍 Location: Copenhagen (co-located with #EurIPS)
📅 Date: 2 Dec, 2025
🔗 Website: sites.google.com/view/amortiz...
📝Call for abstracts: sites.google.com/view/amortiz...

We'll discuss tricky questions such as:
• When does amortization fail badly even though it's rarely talked about?
• Which applications can these neat methods actually work for?
• Can we really deploy these methods in the real world?
• ... and more!

Very interesting, looking forward to reading it in detail! For broader context, I recommend looking into the woefully underappreciated literature on (autoregressive) (transformer) neural processes. (Conditional) NPs are probabilistic models predicting data based on arbitrary sets of data.

For example for surrogate-based Bayesian inference, our VBMC works with GPs in the low-likelihood evaluation regime:
github.com/acerbilab/py...

Then for many likelihood evaluations we have other methods, like this one with normalizing flows:
arxiv.org/abs/2504.11554

Yep!

We are not quite doing it yet in a single existing method -- mostly because it'd be hard to publish --, but definitely thought about it many times.

I assume it was referring to different authors of multi-author papers. I also read it as boilerplate “we are not competing with NeurIPS, also go to NeurIPS” by making it clear that this is just a satellite and not a full-blown schism (plus there likely isn’t enough space for everyone).