fastball

My workspace account is "Expanded AI Access" and can't access any Gemini 4 models. Disingenuous to ignore the fact that you can't even sign up for Plus/Pro/Ultra on a non-personal account when talking about "what plan you are on".

Mythos has never been publicly available.

Anthropic does not make money if everyone is like you, spending $20/m on their SOTA model and are happy. That sounds very commoditized to me.

"Much better provisioned" might be overselling it. Mainly they are just serving very few customers per Nvidia GPU, which is an expensive solution to the problem.

No, Cerebras is serving GPT-5.6-Sol (released in July).

OpenAI is back to using Nvidia chips (with small batch sizes) for the ultrafast mode of their latest models.

That takes the speed down from from 750tok/s on Cerebras to 300tok/s, and presumably is much more expensive than the Cerebras solution, but Cerebras has supply constraints. I believe they are planning aggressive data center expansion in this quarter but much harder to scale up than Nvidia chips.

This isn't really surprising. LLMs are intended to be generally intelligent. They are spiky on the things frontier labs tend to benchmark against (software/math) but are not intended or expected to be great at everything. The general intelligence unlocks new workflows, it does not perfect existing ones.

That being said, I'm sure a frontier LLM could make some improvements to your CNN.

You are definitely doing something wrong. All of my sessions constantly use subagents and are able to poll with background tasks.

Was the prompt "Re-implement Call of Duty 4, make no mistakes"?

What was preventing this in the era of "everyone has a smartphone"?

The original roadster was Tesla's first car.

On the one hand, I absolutely agree that we should try to measure and reduce verbosity that LLMs are slopping into our codebases.

On the other hand, the fact that Astra is "winning" this bench is not great signal for me, since imo Astra produces basically unreadable code. Astra plays code golf. My own revealed preference is that I'd rather have the excessive comments and tests of Claude rather than the inscrutability of Astra.

How is that comforting? Humans pose almost infinite danger to themselves. We can (and have) done pretty much everything you can do. We've had the capability to glass the entire earth for 70 years. We've had bioweapons, chemical weapons, mass murder, you name it.

Accelerating the ability for any given person to have access to some amount of these capabilities is terrifying, is it not?

And there are dozens of open-weight LLMs, but OpenAI and Anthropic are both immensely valuable because their models are actually intelligent.

Seems like a bit of a nothing burger.

Not always?

They are still gunning for ASI.

If someone considered Peter Thiel an ultra fascist, wouldn't a great thing to do be to take his money and spend it on... not fascist things?

I know Bun is kinda not the fan favorite right now with the Anthropic acquisition and the Rust re-write drama and such, but on every one of the dimensions listed in this article, I like Bun better than either Node or Deno.

How many datacenters are in the "arid west"?

Evaporated water stays in the water cycle though. It comes back down as rain, and returns to the water sources that were depleted (assuming these data centers are generally not using underground aquifers, which I think is true).

Is there much impact of a marginal gallon? This is evaporative cooling, no? It's gonna come back down as rain and refill wherever datacenter took it from.

Are you sure? People think water for data centers is "consumed", like the output is sewage or something. Giving those people concrete numbers is not gonna change such misconceptions (and that's assuming they look at such numbers in reasonable relative terms in the first place).

Fitter Than Linux?

What if you view it like a parent instilling morality into a child instead?

tbh saying all these dumb decision models are similar to Jev is like saying markov chains weren't far from GPT-2.

The value isn't really in the I/O shape, it is in the intelligence combined with the output shape. Every extra ounce of intelligence in these models unlocks additional use-cases. But the converse is also true: a dumb decision model is going to be less useful than using a more intelligent standard LLM.

That is the appeal of Jev: for certain usage it has more intelligence than some small SOTA LLMs. It is the first decision model that actually feels intelligent (to me).

Small tangential note, but I think you might mean "not for the faint of heart".

It hasn't just fixed it, it has introduced a new paradigm in anti-obscurity.

Amazon is not usually very good anymore, quality has suffered massively.

Hallucinate tools that don't exist.

> a bit cheaper

Gemini 2.5 Flash Lite is $500/Gt, Jev is $42/Gt. AKA an order of magnitude cheaper.

> BERT with more data

It is specifically not just that, in the same way that models which have been chat/task-optimized via RLHF (which made these models much more useful for a huge variety of tasks) are not just "the base transformer model with more data".