My mental bias always kept me away from Chinese models. Because i know that china is a surveillance state and all the things we know about CCP. But after what we learned about OpenAI and how they most likely used user data to basically cheat in an open competition i think it does not matter which AI provider you use all of them will own your data and all of them can spy on you. So I am willing to switch to Chinese models. This way we help them develop and improve models some day we can run them locally.
The Chinese labs have released interesting papers to accompany their releases too, especially DeepSeek and Kimi. This improves their standing among a few of us, who really like to see and read the papers with details about what they have changed and how their models work.
These models are open-weights. Anyone can host them, you don’t have to use chinese servers even though most of them offer zero data-retention policies.
The new meta model is fast and very cheap as well, and when used through OpenCode you get quite a lot of free tokens. But meta is also THE surveillance company, so probably also not a good choice in your case.
Is it? You're still putting a lot of thought and guidance into the agent's harness, the final code is just a tiny bit of that. It's like giving a junior developer final code vs explaining the whys and nuance. Which I'm not sure I want to give Meta
You're naive if you're thinking the scumbags running the US companies aren't using your data.
In any case old rules apply: if privacy is a concern don't share the data. I share all my work-related code because it's worthless, but I don't and would never share company business and process details, access to production/user data, etc.
Meanwhile I know of people connecting all the kind of MCPs for datadog/sentry/jira/concluce/production databases to their harnessess..lol.
It does. I believe he is one of the worst people and contributed to misery of humanity. I had nothing against him until he dismantled USAID. The richest man in the world did not go to a party on weekend to make sure that poorest men on the world have less help. He was basically on the side of HIV.
So i never use his products.
Beside don't read too much in to benchmarks. They are alrrady ruined by Goldhart principle. Tgese models have already seen most of the data.
I actually use Claude for finding stuff and recommendations. like "Find me a decent shoe with blah blah conditions" this query now goes to LLM Chat instead of google search. I use google search to find the url of something that i know exists like find the website of a company or finding address of a store.
The issue is Claude is losing money serving you that, and while Google makes money serving you the same result. On top of that, you are paying Claude to get that same result.
Unless Claude adds ads, it isnt going to be sustainable for both Anthropic or you - and congrats, you've invented Google Search.
Google and ChatGPT are both running LLMs on free user queries. If Google is doing it more profitably, it’s because they have a much more mature ad business, and/or they are running a cheaper LLM. There’s no fundamental difference in the business model in that specific product line.
Regarding Claude, it’s true they must be losing money on free users since they promised they won’t put ads there.
Speculatively, I think people are overestimating the cost of inference for the consumer free tier chatbots. I suspect it’s effectively a marketing cost. The primary expenses are compute to train the next model, stock compensation, and heavy-duty enterprise inference (which pays for itself).
I agree. But then again you and I are contrbuting to that. We invest our money into S&P 500 hoping it'd grow (and that greed / need is what is contributing to companies trying to make a profit).
It isnt Netflix deciding, sure I'll make money, it is more shareholders are demanding profits .. and Netflx is forced to act.
So if Ads are a disease, and we (you + me + anyone with a stake in the US economy's growth) are the virus.
> But that’s it. I never use Google outside of that. It’s just ads, and most of the web is just ads or AI slop.
Funnily enough, their core revenue driver (Google Ads) is very broken too. I'm trying to run some ads, but for a week they haven't been showing due to some invisible combination of flags when the campaign was created. There's no way of knowing they're not showing from the dashboard, it only becomes apparent when you try and preview the ads with one of your search terms.
I know everyone is long Google but that experience seriously makes me question how valuable their ad business will stay in the future.
Because legally then it's a copy and not a format conversion.
This story has a lot of details left out and is a magnet for illinformed commentators: the reason the originals are destroyed is because it is legal to do so for the purposes of format conversion, but not legal to retain the originals under copyright law.
One thing about local llm projects is that they don't mention the maximum context length which for this project is 4k. I think for it to be used with a harness you need at least 200k and reliable tool calling and structured output which obviously degrade at high quantizations.
I always wanted to learn more about scripting. But today I am not as passionate as before because LLMs write working scripts most of the time. I am wondering if it is true for most programming techniques and quirks. Are we going to write code to solve low level problems?
The other day there was a blog post about learning SIMD. I think in future "programmers" will just nag about the speed of the program and the coding assistant will eventually introduce SIMD to the source code.
It is a little sad but we have to go with the current if we want to survive.
I can see crafting line by line when I’m retired or for a personal project. It’s eventually going to be like the people now who still operate restored manual linotype machines. “Look here I have the last functioning one in the midwest. See how the plates work…”
I still think that learning the shell environment makes a lot of sense these days. But mainly for interactive use. It's amazing the scripts your LLM can make you with, say, fd and fzf.
They could have fixed this if they cared.