wolttam

I don’t quite get the classification of alien considering they were grown with human data (and humans were the ones that invented the architecture to make it work in the first place)

The difference is that Amazon’s to-consumer revenue is very low (~5%) while Apple’s is very high (~95%)

I run self hosted models, don’t go anywhere near Anthropic/OAI, and I share this view.

Maybe it’s also folks adapting to the new landscape ahead of them, not desperately holding on to the old norm.

That is a high expectation of people being driven by predictable human emotions, lacking the ability to think critically through everything in front of them.

The real issue of our media landscape (information feeds controlled by the elite), probably could have been nipped in the bud some years ago but it may be a little too late for that.

    curl > some_file
    bash some_file

(Also, how in the world would a server be able to tell where you’re piping the response?)

Edit: thanks for the replies, that makes perfect sense

There's always going to be some nerds out there that want to make and release a model as open, and share their knowledge along the way. I'm not too concerned.

The fear you're sharing has been shared since day 1 of open-weight models and it has yet to come to fruition.

Further, regardless of what any of these labs license their model as, the end user who has downloaded the weights can still use them for virtually any purpose (synthetic data, classification, etc) as much as they want and there is no way to reasonably enforce a license on those outputs.

So even if you can't run a service using this model's weights (without a license), the weights still do provide utility for training new models with any other license.

> "A REAL, ultra-photorealistic, high-end 35mm cinematic first-person perspective..."

Real nice prompting there, Google. It also still looks like a video game rendering.

The financials depend on revenue that isn’t materializing because:

- the utility of the models is not that great (see almost every company outside of tech and even many within tech reporting “no measurable increased revenue or productivity as a result of adopting the technology”.

- China has been undercutting them hard with open-weight models that are “good enough” for 95%+ of use-cases, leaving only the hardest problems for OpenAI and Anthropic’s models, which it turns out isn’t where there is a ton of money to made (curing various forms of cancer is valuable but not “let’s dump trillions on these companies” valuable - and once their models can solve it, it probably won’t be long before others do too).

People want these models on their own hardware. People will (eventually) want these models running on robots that can do real work for them, and they won’t want those things tethered to an insecure and privacy eviscerating cloud subscription.

The money to be made is in selling hardware, as it should be.

Anthropic’s S1 still shows them not even breaking even on revenue/expenses during what was probably their biggest year of adoption, right before people started going “oh, I can do this with an open model”

I’m not saying anything new here. What are we missing?

It is namespaced under the pylance project. I also don't see any announcement here, just a markdown file.

Perhaps the post title could use a tweak

If the model is predicting syntactically incorrect tokens in code you probably have bigger issues

There is no “earning potential” to be gained in a world where you simply ask machine to do X and it does it for you.

We might not quite be there but we’re trucking towards that point at full speed.

Xiaomi-Robotics-1

Because Opus 5 is the best at their benchmark.

It's a bit fascinating to me how all of these labs have rolled their own (it sounds like) completely unique RL environments.

I wonder how much the industry will centralize around a few key patterns, and if we'll start to see some RL framework projects gain traction and be used by the bigger labs

P.S. weights by November 4th

It really is the determinism aspect for me. Okay, it’s programming if you sample at temperature 0 with a fixed seed on a fully deterministic inference stack and you have made your prompts the artifact that you iterate on (repo full of natural language that “builds” into the running software it describes with the use of the LLM sampling at temperature 0)

The fact that Grok refused is hilarious.

> Podman, by default, uses crun as the container runtime

Not anymore, they switched to runc (but it can use any OCI compliant runtime)

I’d argue that they already are trained on this idea, since the idea of humans being lazy and dumb and relying heavily on technology is, well, not new.

Just whether those parts of the network see activations during trivial coding tasks (I’d have to assume they actually do sometimes, depending on the input prompt)

It just won’t be a chat box. You need something that is always there and listening (at least during work hours), asking questions and figuring out the entire scope of the business, before they will be able to come up with (useful) ideas.

Sorry to be pedantic but it looks like it would have fetched up to $50k, and I have to imagine they would have tried as hard as they could to keep that number as low as possible.

If anything I am confused by how high cache costs still are across many models

The really expensive part of inference - the compute and memory traffic required to perform prompt processing and decode - is paid for roughly once per token in a multi-turn session when prompt caching is working properly.

The difference to the provider's costs between:

- Submitting 100 back-to-back requests, each adding 1k tokens to the context

- Submitting 1 100k token context request

Is negligible. Yet the consumer is going to pay a lot more in the former case for all those repeatedly submitted input tokens that you get charged the cached rate for (but which required no additional compute on the provider side).

DeepSeek V4.1 has an insanely optimized KV-cache that takes around 900MB per million tokens, making it feasible to move to disk between requests. Makes sense they're cheap

“This person likes semi-colons a whole lot”

I’ve already seen that trick snuck in by newer models. “I won’t use em-dashes; I’ll use semi-colons!”

Weights in 2 weeks they say

Since there is not a lot of hope for me to reasonably run DS V4.1 on my 2 x DGX Spark setup (I know, spoiled), both the “readable prose” and the fewer tokens per task claims (relative to the base GLM-5.3-Flash) are most welcome.

I’ll absolutely be giving this a go once weights land (a QAT FP4 experts release would be most appreciated)

Exactly, there is no ”best.”

Most people are probably looking at: https://artificialanalysis.ai/models/comparisons?compare=mus...

This is still a welcome model, even if I won't be running it at home

I disagree with this framing entirely.

It's unsurprising and perhaps even desirable that different LLMs return similar results to an identical prompt.

If you want variation, prompt for it.

LLMs already do that but the advantage they have is being able to walk you through a plausible explanation for why they’re offering one position over another.

Using the output of a “decision” model without insight into the reasoning for a given decision seems very trusting.

Aw man, my beloved .lan for my home network

I kind of think it's crazy to not have the ability to filter/rewrite DNS requests/responses on your home network (by any means, dnsmasq is just one of them)