kristjansson

> The phrase "frontier model" is starting to mean two things. One is a checkpoint. The other is a system boundary. LLM-isms aside, I don't think we want this to be the case? An LLM, …

deprecated last month

Be sure you compare inputs tokens to pre-fill rates and output tokens to generation rates.

other way around. it's trained to generate long CoT to reason through problems (and does it well!) but has ~no tool calling capability, and ~no ability to manage more than 1-2 messages. see the …

> Nvidia's raster stack is more-or-less complete That was my point. It's not even a pivot! They're still making consumer cards! They've even product-differentiated enough that…

Controversially, Nvidia employs more than one person and so is in fact capable of producing more than one thing at a time.

Maybe once the model can administer and operate the service too. For now building the 10% of the SaaS that you need still leaves you operating 100% of a new service/process

> no data hallucinations Dangerous thing to assert. It’ll happily run SQL that works, but doesn’t necessarily correspond to intentions or unstated assumptions about the data.

SAN vs local. Local NVME (“instance storage“) on AWS is wicked fast too, but live and dies with the instance

Emacs is not great software because it has good features. It’s great software because it conforms itself and those features to the user’s needs and preferences. If one of those preferences is short, …

My .emacs.d has been carefully handcrafted over years. My .vim/.vimrc accreted over the same period, bits copied in, commented out, … Claude did my neovim in about 45 seconds, with the instructio…

Those are just the sounds that animal makes. Live with the animal long enough, you learn how the sounds correspond to its internal states, even if you don’t really know what they mean. I’d be a bit wo…

It’s insane enough to consider undertaking the risk of a moon or mars colony. But doing so with him as the sole load bearing link in your supply chain? What if mars turns out to be woke and you’re fed…

It's easy[1] to promise, it's hard to deliver. I hope the best for them. [1]: https://magic.dev/blog/ltm-1 (note the date)…

>A full breakdown of the mechanism and how it compares to FlashAttention, DeepSeek sparse attention, and recurrent architectures is in the Technical Report. Oh they did publish details lets read th…

Seems like a lot of machinery over-and-above running the same review n times + aggregating the result. What led you to the design?

Learning that those shortcuts are arbitrary, application-defined conventions is the first step toward enlightenment.

Why does emacs have to be for everyone? Its attraction, to this user at least, is the lifelong process of adaptation and discovery that molds it to the user, and the user to it. It’s like a pet. I lov…

C-u 0 M-x asi-optimize-everything (The prefix arg disables paperclips, for obvious reasons )

or that intelligence is merely the composition of many redundant, lossy, ~random components

https://thickets.mit.edu

It’s expensive, but not unfathomably, esp in an open source setting where capable people might contribute high quality data for post training (worked problems, code reviews, feedback, …) gratis instea…

It’s actually ok to be more critical of a government that’s capricious than one that merely advances polices one disagrees with

Dual use is definitely not a new thing; sacrificing the ostensible primary use for the secondary seems to raise the author's ire (and mine).

> the shelves were being cleared to make space—not for more books but for space itself. ... The new library has four floors. Two of them feature books Despite the frenzy of building at most America…

> 42 kW/kg Is that right? One could get ~200hp from something the size and mass of a small dog?

Update: not any more!

Obscuring core logic is the most egregious part of most agent frameworks. One needs a clear view of what, exactly, is being sent to the underlying language model, and what's coming back. Everyt…

I think MTP Gemma4 support is still WIP https://github.com/ggml-org/llama.cpp/pull/23398 ?…

> quantization 12b means 12G @ 8 bits/param (basically lossless) and 6G at 4 b/p (generally accepted 'pretty close' level). Not too bad? But TBD how well the base model performs…