hackpert

We found evidence of specific layer-localized "reasoning" circuits in a few models last year too! A very much work-in-progress paper is here: https://openreview.net/forum?id=mTjGBrkdtz

Huh there is so much limonene in Coca Cola?! Limonene works as a very good…pesticide and herbicide! I did a research project on limonene like 10 years ago with my mentor and it outperformed most commercial pesticides in controlled settings. It really can't be that great to ingest.

These metaphorical database analogies bug me, and from what it seems like, a lot of other people in comments! So far some of the most reasonable explanations I have found that take training dynamics into account are from Lenka Zdeborova's lab (albeit in toy, linear attention settings but it's easy to see why they generalize to practical ones). For instance, this is a lovely paper: https://arxiv.org/abs/2509.24914

Hi! Did you ever end up running this reproduction? If yes, could you also check if the Putnam/IMO problems are in the training data perhaps by trying to have it complete the problems n times? I would totally do this myself if I weren’t GPU poor!

O

If anyone else is curious about which ARC-AGI public eval puzzles o3 got right vs wrong (and its attempts at the ones it did get right), here's a quick visualization: https://arcagi-o3-viz.netlify.app

Sorry never mind! I wasn't thinking at all when I wrote that thought, but that obviously doesn't make sense :)

That's fair but what if you could estimate the direction of incoming light with other sensors? Using inverse diffraction etc. Just a thought

Hi! I've been working on theorem proving systems for some time now. I would love to help out with an AlphaProof reproduction, but I can't reach you on discord for some reason!

Thank you, those insights are invaluable! This is a specific and potentially dumb question and I completely understand if you can't answer it!

The practical motivation for MoEs is very clear but I do worry about loss of compositional abilities (that I think just emerge from superposed representations?) that some tasks may require, especially with the many experts phenomenon we're seeing. This is an observation from smaller MoE models (with like top-k gating etc.) that may or may not scale, that denser models trained to the same loss tend to perform complex tasks "better".

Intuitively, do you think MoEs are just another stopgap trick we're using while we figure out more compute, better optimizers or could there be enough theoretical motivation to justify their continued use? If there isn't, perhaps we need to at least figure out "expert scaling laws" :)

I'm not sure how to quantify how quickly or well humans learn in-context (if you know of any work on this I'd love to read it!)

In general, there is too much fluff and confusion floating around about what these models are and are not capable of (regardless of the training mechanism.) I think more people need to read Song Mei's lovely slides[1] and related work by others. These slides are the best exposition I've found of neat ideas around ICL that researchers have been aware of for a while.

[1] https://www.stat.berkeley.edu/~songmei/Presentation/Algorith...

There has been some interesting work when it comes to distributed training. For example DiLoCo (https://arxiv.org/abs/2311.08105). I also know that Bittensor and nousresearch collaborated on some kind of competitive distributed model frankensteining-training thingy that seems to be going well. https://bittensor.org/bittensor-and-nous-research/

Of course it gets harder as models get larger but distributed training doesn't seem totally infeasible. For example if we were to talk about MoE transformer models, perhaps separate slices of the model can be trained in an asynchronous manner and then combined with some retraining. You can have minimal regular communication about say, mean and variance for each layer and a new loss term dependent on these statistics to keep the "expertise" for each contributor distinct.

I tried to do this ages ago in 2018 by adapting OpenAI's flow architecture and it sort of seemed to work (was at least promising). With today's models with a significantly more disentangled latent space it should be much easier to do! I saw a transformer trained on the UK Biobank recently, excited for this space!

Wow I had hoped for a more productive discussion than these 1-1 comparisons of Bard vs ChatGPT that I'm seeing everywhere. The model deployed with this version of Bard is clearly a smaller model than the biggest LaMDA/PaLM models Google has been working on for ages. Which, according to their publications, show unprecedented results on _proof writing_ of all things (see Minerva). While their strategic decisions may be questionable (or they're just trying to quantize the model for mass deployment without burning billions per month in compute costs), its almost silly to question Google's ability to build useful LLMs.

I've been using Metaphor for a few weeks now and have almost entirely switched from Google and other search engines. Keyword based search simply doesn't come close when it comes to getting the _right_ results. While I have to sift through a few pages of results on Google and then maybe find what I'm looking for, on metaphor, there's almost no SEO spam or Wikipedia-style links dominating the top results. It directs you to sources that are relevant to your search query. I don't know how they did this (probably a lot of very specific and targeted tricks), but Alex and team have created a marvelous product and I'm excited to see where this goes! Congrats on the launch!

Right, but doesn't that mean that it could potentially be used for designing algorithms that have componentwise numerical stability over some kind of floating point standard, but this, by definition being a result over finite fields, should be numerically stable?

(apologies if I misunderstood, I wasn't calling you out specifically but a generalized misconception I've noticed in a lot of other discussions so far)

While your point about numerical stability is correct in general, there are no numerical stability issues here and I think this conception, which I've seen in more than one place now, stems from a fundamental misunderstanding of the paper's results. While they _did_ come up with a faster TPU/GPU algorithm too, the primary result is not a fast matmul approximation, it is an exact algorithm comprising of stepwise addition/multiplication operations, and hence is numerically stable and should work for any ring (https://ncatlab.org/nlab/show/ring). AlphaTensor itself does not do the matrix multiplication, it was used to perform an (efficiently pruned) tree search over the space of operations to find an efficient, stable algorithm.

UPI's penetration in (urban and semi-urban anyway) India is honestly incredible. I worked with/on the tech when it was very nascent in 2016 and in 2022, on visiting India after a long time I was stunned to see _everyone_ has a little PayTM QR code card. Vegetable vendors, taxi drivers, roadside hawkers, small business owners. It's brilliant because the system is banking the traditionally unbanked, and generating tremendous amounts of data that can hopefully be put to good use by economists, honestly unlike any other system in the world. Even MPesa in Africa has stupidly high withdrawal and transaction fees, which UPI doesn't need at all.

That is actually pretty freaking cool. Not hard to do by any means, just cool.

x2go works quite well actually. Its definitely not like local but it does the job.

A similar thing happened to me and my partner in the US, in a fairly liberal part, so I'm not entirely surprised that it's happening in SEA. People wouldn't rent to us because we told them we cook at home and these white folks were like "but Indian food smells!"

Honestly! This is such a good initiative that it bothers me that this is flagged. Housing issues for queer people is one the biggest issues across the world. I mean the first things we consider before even booking a hotel across most of the world is "am I going to be thrown out of this?"

Ooh good recommendation, lobste.rs is cool. I've been looking for someone with a potential invite for a while too if anyone here is offering :).

About the bit regarding you possibly commuting from Paris to London for work, I'm curious - how would that even work? Would you go through customs every morning/evening? Has anyone here done this?

*her mentor :)

Source: https://ewintang.com/

Yes yes yes please somebody look into latency with these fancy ML methods! You can quite literally have most ML models approximated to a very good degree as very fast DSP using very few processor cycles given modern-day CPU optimizations. Or heck use an ISA simulator plugged into another fancy ML model, and have it also optimize to minimize instruction count while recreating the same signal! (having a model optimize on its own inference is a neat trick, but I digress.) I’m sure ML is just one bottleneck among many (looking at you, Chromium) but I so desperately wish people started caring about latency again.

Last month, I was working with some people on building a decentralized health data store for vaccine waitlists (a la Mailbox) and certificates and we were looking at Solid as one potential backbone. I really admire what they’re trying to do. However, as some other people have observed, their example applications are quite awful and not very useful(?) and they seem to be marketing it to end users, not developers. In my experience, most end users do not care about privacy and security until 1) they either lose a lot of money or other valuables 2) some hacker comes knocking on their door for extortion and they lose a lot of money or other valuables.

It would be so much better if they just made it so easy and simple for developers to use that it becomes their default, and not rely on hope that end users will suddenly start using decentralized apps in the name of privacy.

While this certainly seems to be redefining ASD as other commenters seem to suggest and we can debate if that is a good thing or not, it seems like it would be relatively straightforward to test the validity of this theory (perhaps not easy but straightforward.) Simply look at autism occurrence rates in countries where partner choice is still socially determined such as India/Pakistan/parts of the Arab World(?) etc. and see if there is any correlation between some arbitrarily defined scale of partner choice freedom and ASD rate, after factoring out individual genetics :S.

PS. I haven’t read the actual paper yet so I don’t know if they have already done this.

I disagree heavily. The focus of this conversation seems misdirected, whether intentionally or not. It’s not about the fragility of morality as a system at all, which it of course is because it’s based on consensus. It’s about fundamental principles and upholding them, and how you prioritize those principles.

You seem to believe in upholding right to free speech, which, surely is crucial to the country. However, this other group you speak of is protesting their rights to life and liberty, which they are being actively deprived of by an oppressive regime. Isn’t it slightly unfair? What rights is the right being deprived of? Entitlement to a certain election result or a certain office? That opposes the very definition of this democracy. They have means for their voices to be heard and their issues to be heard, still, by the right’s representatives in Congress. Black people didn’t/don’t because law enforcement was/is being racist and oppressive in unfairly killing them.

Honestly I just sometimes want a clamp with a flexible sturdy arm, like one of those workbench lamps, to suspend my laptop above my bed. Pretty cool work nonetheless!