yogthos

17 ms·

331.
by yogthos 5mo ago
There are absolutely fully open source models. These are not frontier models, but they very much do exist. OLMo is one of the models explicitly mentioned as having passed the OSI's validation phase. Pythia was also validated by the OSI
332.
by yogthos 5mo ago
My expectation is that local models will be the default for coding within a year or two. You can already run Qwen 3.6 with MTP at a pretty reasonable speed without needing a huge amount of VRAM. And while it's not as good as current fr
333.
by yogthos 5mo ago
That's absolutely correct. I use LLMs heavily in my work and open source projects. These tools are effective, they can solve hard problems and they allow me to work on a wider range of tasks than could before. But, they're also ja
334.
by yogthos 5mo ago
The primary argument of the article is that it's physically impossible to dissipate enough heat to power even a modest data centre in space. What you refer to as lack of imagination is my understanding of basic thermodynamics which you
335.
by yogthos 5mo ago
But won't somebody think of the robber barons!
336.
by yogthos 5mo ago
That's just an infantile straw man though which has nothing to do with how communism works. Communism simply means that workers own their workplace and are the primary beneficiaries of their labor. In practice this means a mix of state
337.
by yogthos 5mo ago
https://finance.yahoo.com/news/wealthiest-10-americans-own-9...
338.
by yogthos 5mo ago
unions work
339.
by yogthos 5mo ago
nothing about data centres in space makes sense if you understand even a modicum of physics https://taranis.ie/datacenters-in-space-are-a-terrible-horri...
340.
by yogthos 5mo ago
I've always looked at it as a platform to build stuff on top of as well. I expect that we'll be treating this tech the same way we treat stuff like Linux today. It'll become common open infrastructure that will be used to bui
341.
Different models solve number-theory race problem(aicc.rayonnant.ai)
2 points by yogthos 5mo ago|0 comments
342.
by yogthos 5mo ago
You can't really do any serious work with a chat interface though. Also, you're still wasting time figuring out if your providers do what you expect. If you run a local model then you know it's going to work the way you set i
343.
by yogthos 5mo ago
Or whether the model still works the way you want. For example, a lot of people were pretty unhappy with Claude 4.7 and preferred the way Claude 4.6 worked. If you're relying on a service, then you're stuck with whatever changes t
344.
by yogthos 5mo ago
I assumed you were asking about capabilities since that is a question that can be answered. There isn't any comprehensive reporting by companies, so it's just the anecdotal reports the video I linked discusses. And there are also
345.
by yogthos 5mo ago
This is absolutely false as a recent study from Stanford clearly states https://hai.stanford.edu/news/inside-the-ai-index-12-takeawa...
346.
by yogthos 5mo ago
Here's a recent Stanford study showing that Chinese models are basically just as good https://hai.stanford.edu/news/inside-the-ai-index-12-takeawa... For most use cases, you don't actually need frontier perfo
347.
by yogthos 5mo ago
Sure, publications might not equal talent, but the fact is that China leads research in 90% of crucial technologies. So, clearly China leads in a very tangible way here https://www.nature.com/articles/d41586-025-04048-7
348.
by yogthos 5mo ago
Who cares how the script was generated. What he says is entirely factual. He cites plenty of concrete examples too. The idea that the talent in the US surpasses the global research community is laughable. China already tops the world in art
349.
by yogthos 5mo ago
What are you talking about even. Chinese models are what pretty much every AI company in the US is using now because you can run them on prem and customize them, and because hosted versions cost a fraction of US ones. https://www
350.
Chinese memory module makers ramp up production as DDR5 breakthrough hits market(scmp.com)
13 points by yogthos 5mo ago|0 comments
351.
by yogthos 6mo ago
I'd argue there's little rational for having the model talk down to people which isn't malicious. If the user doesn't understand the answer, they can explicitly ask the model to explain it in simpler terms. If you read t
352.
by yogthos 6mo ago
Everyone’s worried about AI taking jobs or whatever, but baked in biases are another very real and more basic problem that doesn't get nearly enough attention. This study took GPT-4, Claude 3 Opus, and Llama 3 and fed them the same 1,8
353.
LLM Targeted Underperformance Disproportionately Impacts Vulnerable Users(arxiv.org)
9 points by yogthos 6mo ago|3 comments
354.
by yogthos 6mo ago
This was basically my realization as well. We are trying to get LLMs to write software the way humans do it, but they have a different set of strength and weaknesses. Structuring tooling around what LLMs actually do well seems like an obvio
355.
by yogthos 6mo ago
I have a subscription and I have not seen any difference in performance during on/off hours. What exactly are you basing this on?
356.
by yogthos 6mo ago
Similar idea, I find tree sitter is nice because it already supports a bunch of languages and it's easily extensible. Once you the AST, you can really have the LLM go to town with it.
357.
by yogthos 6mo ago
Awesome, and feel free to open issues if you find anything missing that would be useful.
358.
by yogthos 6mo ago
I find a lot of the inefficiency also comes from the model just randomly poking around and grepping all the time which is the fault of the harness. I ended up building a Prolog based MCP where I use tree-sitter to parse the code into a grap
359.
by yogthos 6mo ago
Why should you feel the need to make a vapid comment on an article you admit to not having read?
360.
A 200-Person Chinese Team Just Embarrassed Every $500B AI Lab on Earth(curiousmodels.substack.com)
6 points by yogthos 6mo ago|3 comments

More ›