nbardy

This is a weird number.

And reeks of the same sort of reallocation fallacy that makes people think the rich making too much makes them poor.

If we really just say 4xd everyone’s salary in America. Prices are gonna rapidly rise in everything.

Things don’t get materially better unless we build the material things we need. We need to come up with a way to build more houses. Lower the cost of healthcare. Not just increase everyone’s money supply.

These are true and it does make theses fields the first to fall, but also the hype comes form the fact that it can escape these conditions as well: - generalize to non verifiable domains (https://arxiv.org/abs/2507.17746)[This is on going work but has had steady progress in many angles of attack) - Visual Reasoning is alive and well in video diffusion and image models(see recent works around using diffusion model priors as world model features for physics reasoning) - Prior art exists for anything humans can do online?(Is this one even a hold back?)

We’re definitely going to need a lot of Gpu’s

Weird flex. There is cheaper better and faster models you could of moved to with an hour effort

The first bit was interesting and then you flipped right to generic cynicism.

They would be impressed with our technology even if it has downsides. Wisdom is knowing humans and technology and imperfect tools.

Signing this sounds like a good way to get fired. Executive in corporations gets to make the decisions. Employment is at will, if you don’t like it you get to leave otherwise you’re not fulfilling your contract

No, look a Composoer 2, it stands out starkly on its own in the pareto frontier on low cast and fast models.

Composer 2.5 was a huge leap with minimal compute from xAI.

They can compete with OpenAI and anthropic with xAI scale compute. They have a top notch model team and incredible training data and huge enterprise costumer contracts.

It’s a bit misleading to say nothing special, as they are doing more than just increasing parameter count. Progress has been steady in all the sub components of training from data filtering and weighting to sparse attention, optimizers to up and down the stack various efficiency in training computing.

They’re using more compute, a bigger model and tons of training quality improvements to get more out of an equivalent model.

This does feel like the perfect setup for Claude though.

Much easier to create a vm testing swarm of 100 disitributions with llms

This has been my thought for a long time. I think all that matters from attention is that there is crosswise comparison going on.

You need some amount of parallel compute and some amount of global comparison.

And the rest is basically a ways to parameters and scale.

(This is in theory, in practice you can get a lot of small % stability and efficiency improvements that really compound in algorithmic details of model architecture)

In general this is the way I see open source going.

We won't reuse open source libraries as libraries we import, but as design inspiration for the bespoke tools we make.

It's too cheap to make your own stuff and too expensive to be stuck with someone else primitives.

But grounding AI Coding in existing tools is incredibly powerful.

Confidently yes. OpenAI for sure has been training larger models internally and distilling.

Pre-training scaling laws all support larger models being more cost effeceint to train then smaller models. And distillation is comparably cheap. So you can get the most juice by training the biggest model you can and distilling it.

There is endless returns to frontier intelligence, just because most people can't make use of it doesn't mean someone can't make a ton of money off of it.

Most software engineers will just need cheap tokens.

But things like physics and drug discovery have no foreseeable upper bound.

There is endless returns to frontier intelligence, just because most people can't make use of it doesn't mean someone can't make a ton of money off of it.

Most software engineers will just need cheap tokens.

But things like physics and drug discovery have no forseeable upper bound.

Seems like wrapping async await functions with CSP was a better way to handle this . Clojure already had a nicer pattern for this

People keep saying this and they don't understand how businesses work.

Cursor has 1B in enterprise revenue. It doesn't matter if people can clone their product, those deals don't move slowly

There is step changes that actually merit this though. And a zero day machine IS one of those. It went from 4% zero day success rate to 85% on firefox.

Can you not see the significance of that?

You can estimate on tok/second

The Trillions of parameters claim is about the pretraining.

It’s most efficient in pre training to train the biggest models possible. You get sample efficiency increase for each parameter increase.

However those models end up very sparse and incredibly distillable.

And it’s way too expensive and slow to serve models that size so they are distilled down a lot.

How much of your RAM does that use including kv cache. Is there enough left to run real dev workloads AND the llm?

Also can you run batchwise effectively like vllm on cuda?

Enough to run multiple agents at the same time with throughput?

Why does apple want to make this hardware hard to access?

What actual benefits do they get?

I guess they can have their own models run faster than the competition on their hardware? But they don't even really have anything that consumers use on the ANE as far as I can tell and local LLMs are taking off on macs and could really benefit from this

They are far behind. Go check re-swe bench to see the overfitting measured

Or just try to use them. They don’t generalize as well.

They are benchmaxxed.

They should probably fund their military first.

It’s petulant the way the EU is throwing a hissy fit after we’ve had lop-sided trade deals for years and funding the entire NATO alliance ourselves.

They act like we’re going to war with them when we’re asking for parity and for their self reliance to increase.

No it’s not. I have written cuda kernels and 8bit optimizers with this.

They’re actually very good at speed optimization and can iterate very quickly taking notes on trials and failures and benchmarks. I’ve had it write 10 different attempts in around an hour and benchmark them all then merge and beat very strong baselines in torch

Claude Code officially added native support for the Language Server Protocol (LSP) in version 2.0.74, released in December 2025.

I think from training it's still biased towards simple tooling.

But also, there is real power to simple tools, a small set of general purpose tools beats a bunch of narrow specific use case tools. It's easier for humans to use high level tools, but for LLM's they can instantly compose the low level tools for their use case and learn to generalize, it's like writing insane perl one liners is second nature for them compared to us.

If you watch the tool calls you'll see they write a ton of one off small python programs to test, validate explore, etc...

If you think about it any time you use a tool there is probably a 20 line python program that is more fit to your use case, it's just that it would take you too long to write it, but for an LLM that's 0.5 seconds

Your way off, this reads more like anti capitalist political rhetoric than real reasoning.

Look at Nvidia nemotron series. They hav become a leading open source training lab themselves and they’re releasing the best training data, training tooling, and models at this point.

When are people going to drop the immigration is good at all costs assumption.

We need a well managed set of immigration polices or country WILL take advantage of US. These are our military rivals and we sell our most advanced math, physics and engineering seats to the highest bidder. It’s a self districting disaster and it’s not just on us to treat people better.

Look at the rate of Indian asylum seekers in Canada to see the most extreme case. It happens anywhere you extend naivety and boundless good will.

Those arc agi 2 improvements are insane.

Thats especially encouraging to me because those are all about generalization.

5 and 5.1 both felt overfit and would break down and be stubborn when you got them outside their lane. As opposed to Opus 4.5 which is lovely at self correcting.

It’s one of those things you really feel in the model rather than whether it can tackle a harder problem or not, but rather can I go back and forth with this thing learning and correcting together.

This whole releases is insanely optimistic for me. If they can push this much improvement WITHOUT the new huge data centers and without a new scaled base model. Thats incredibly encouraging for what comes next.

Remember the next big data center are 20-30x the chip count and 6-8x the efficiency on the new chip.

I expect they can saturate the benchmarks WITHOUT and novel research and algorithmic gains. But at this point it’s clear they’re capable of pushing research qualitatively as well.

You haven’t actually looked at their fundamentals. They’re profitable serving current models including training costs and are only losing money on future RD training, but if you project future revenue growth on future generations of models you get a clear path to profitability.

They charge higher costs than OpenAI and have faster growing API demand. They have great margins compared to the rest of the industry on inference.

Sure the revenue growth could stop but it hasn’t and there is no reason to think it will.

Why are you assuming Anthropic is for sale? They have a clear path to profitability, booming growth, and a massive and mission driven founding team.

They could make more money keeping control of the company and have control.

This is misleading. They had 4.5 which was a new scaled up training run. It was a huge model and only served to pro users, but the biggest models are always used as teacher models for smaller models. Thats how you do distillation. It would be stupid to not use the biggest model you have in distillation and a waste since they have the weights.

The would have taken some time to calculate the efficiency gains of pretraining vs RL. Resumed the GPT-4.5 for whatever budget made sense and then spent the rest on RL.

Sure they chose to not serve the large base models anymore for cost reasons.

But I’d guess Google is doing the same. Gemini 2.5 samples very fast and seems way to small to be their base pre train. The efficiency gains in pertaining scale with model scale so it makes sense to train the largest model possible. But then the models end up super sparse and oversized and make little sense to serve in inference without distillation.

In RL the efficiency is very different because you have to inference sample the model to draw online samples. So small models start to make more sense to scale.

Big model => distill => RL

Makes the most theoretical sense for training now days for efficient spending.

So they already did train a big model 4.5. Not using it would have been absurd and they have a known recipe they could return scaling on if the returns were justified.