71–80 of 83 posts
They trained it in 33 days for ~20m (that includes apparently not only the infrastructure but also the salaries over a 6 month period). And the model is coming close to QWEN and Deepseek. Pretty impressive
Given that it's a 400B-parameter model, but it's a sparse MoE model with 13B active parameters per token, would it run well on an NVIDIA DGX Spark with 128 GB of unified RAM, or do you practically need to hold the full model in RAM even with sparse MoE?
Even with MoE, holding the model in RAM while individual experts are evaluated in VRAM is a bit of a compromise. Experts can be swapped in and out of VRAM for each token. So RAM VRAM bandwidth becomes important. With a model larger than RAM, that bandwidth bottleneck gets pushed to the SSD interface. At least it's read-only, and not read-write, but even the fastest of SSDs will be significantly slower than RAM. That…
Earlier quoted context omitted.
> Experts can be swapped in and out of VRAM for each token. I've often wondered how much it happens in practice. What does the per-token distribution of expert selection actually look like during inference? For example does it act like uniform random variable, or does it stick with the same 2 or 3 experts for 10 tokens in a row? I haven't been able to find much info on this. Obviously it depends on what model you are…
The model is explicitly trained to produce as uniform a distribution as possible, because it's designed for batched inference with a batch size much larger than the expert count, so that all experts are constantly activated and latency is determined by the highest-loaded expert, so you want to distribute the load evenly to maximize utilization. Prompt ingestion is still fairly similar to that setting, so you can firs…
Earlier quoted context omitted.
> but I lack some confidence in models trained on LLM output, so I hope it wasn't that. That's misguided. Models have been trained on synthetic data for ~2+ years already. The "model collapse" myth is based on a very poor paper that got waaaay more attention than it deserved (because negativity sells, I guess). In practice every lab out there is doing this, because it works.
When ChatGPT first released and jailbreaks were pretty easy, I was able to easily get some extremely good/detailed output from it, with very little errors or weirdness. Now even when I can get jailbreaks to work with their newer models, it's just not the same, and no open-source model or even commercial model has seem to come close to the quality of that very first release. They're all just weird, dumb, random or inc…
Interesting statement. But wouldn’t that mean that Google is in an even better position in regard to primary, or at least pristine data?
They trained it in 33 days for ~20m (that includes apparently not only the infrastructure but also the salaries over a 6 month period). And the model is coming close to QWEN and Deepseek. Pretty impressive
I'll straight up accuse them of on purpose muddying the waters. To get to the point of executing a successful training run like that, you have to count every failed experiment and experiment that gets you to the final training run. They spent well over 100 Million to train this model by that definition, and all definitions which don't include the failed runs up to the successful one at the end are at best disingenuou…
What did they do to make the loss drop so much in phase 3? Also, why are they comparing with Llama 4 Maverick? Wasn’t it a flop?
considering this is a preview of the instruct and it's spitting distance from maverick, it's likely to showcase "look what we can do with limited funds, imagine what we can do with more"
Earlier quoted context omitted.
I've always found LiveBench a bit confusing to try to compare over time as the dataset isn't meant to be compared over time. It also currently claims GPT-5 Mini High from last summer is within ~15% of Claude 4.5 Opus Thinking High Effort in the average, but I'll wait with bated breath for the millions of amazing apps which couldn't be coded before to start showing up (or, more likely, be told in 6 months how these 2…
> I'll wait with bated breath for the millions of amazing apps which couldn't be coded before to start showing up Appstore releases were roughly linear until July 25 and are up 60% since then: https://www.coatue.com/c/takes/chart-of-the-day-2026-01-22
I can guarantee if you restricted yourself to just that 60% you wouldn't be responding to me doubting AI apps are already amazing things people are actually supposed to be so excited about using though.
Earlier quoted context omitted.
I've always found LiveBench a bit confusing to try to compare over time as the dataset isn't meant to be compared over time. It also currently claims GPT-5 Mini High from last summer is within ~15% of Claude 4.5 Opus Thinking High Effort in the average, but I'll wait with bated breath for the millions of amazing apps which couldn't be coded before to start showing up (or, more likely, be told in 6 months how these 2…
See peer reply re: yes, your self-chosen benchmark has been reached. Generally, I've learned to warn myself off of a take when I start writing emotionally charged stuff like [1]. Without any prompting (who mentioned apps? and why would you without checking?), also, when reading minds, and assigning weak arguments, now and in my imagination of the future. [2] At the very least, [2] is a signal to let the keyboard have…
I had a very different read of Tao's post last month. To me, he opens that there have been many claims of novel solutions which turn out to be known solutions from publications buried for years, but nothing about rapid increase in the rates or even claims mathematicians using LLMs are having most of the work done by them yet.
He speculates, and I also assume correctly as well, that that contaminations are not the only reason. Indeed, we've seen at least 1 novel solution which couldn't have come from a low interest publication being in the training data alone. How many of the 3 examples at the top end up actually falling that way is not really something anyone can know, but I agree it should be safe to assume the answer will not be 0, or even if it was it would seem unreasonable to think it stayed that way. These solutions are coming out of systems of which the LLM is a part, and very often a mathematician still actually orchestrating.
None of these are just popping in a prompt and hoping for the answer, nor will you get an unknown solution to an LLM by going to ChatGPT 5.2 Pro and asking it without the rest of the story (and even then, you still will not get such a solution regularly, consistently, or at a massively higher rate than several months ago). They are multishot from experts with tools. Tao makes a very balanced note of this in reply to his main message:
> The nature of these contributions is rather nuanced; individually and collectively, they do not meet the hyped up goal of AI autonomously solving major mathematical open problems, but they also cannot all be dismissed as inconsequential trickery.
It's exciting, and helpful, but it's slow and he doesn't even think we're truly actually at "AI solves some Erdos problems" yet, let alone "AI solves Erdos problems regularly and at a rapidly increasing rate".
They trained it in 33 days for ~20m (that includes apparently not only the infrastructure but also the salaries over a 6 month period). And the model is coming close to QWEN and Deepseek. Pretty impressive
> 2048 Nvidia B300 GPU With average price of $6/hour that is $12,288/hour for whole cluster. Times 33 days times 24 hours it comes out to be $9.7MM , assuming no discounts. That leaves $10.3MM/6 months for salaries, which is 103 employees at $200k/year or 51 employee at $400k/year.
Earlier quoted context omitted.
See peer reply re: yes, your self-chosen benchmark has been reached. Generally, I've learned to warn myself off of a take when I start writing emotionally charged stuff like [1]. Without any prompting (who mentioned apps? and why would you without checking?), also, when reading minds, and assigning weak arguments, now and in my imagination of the future. [2] At the very least, [2] is a signal to let the keyboard have…
I'm going to stick to the stuff around Tao, as even well tempered discussion about the rest would be against the guidelines anyways. I had a very different read of Tao's post last month. To me, he opens that there have been many claims of novel solutions which turn out to be known solutions from publications buried for years, but nothing about rapid increase in the rates or even claims mathematicians using LLMs are h…
Didn't bother reading after that. I deeply respect you have the self-awareness to notice and spare us, that's rare. But it also means we all have to have conversations purely on your terms, and because its async, the rules constantly change post-hoc.
And that's on top of the post-hoc motte / bailey instances, of which we have multiple. I was stunned (stunned!!) by the attempted retcon of the app claim once there were numbers.
Anyways, all your bete noirs aside, all your Red Team vs. Blue Team signalling aside, using LMArena alone as a benchmark is a bad idea.