> These companies only offer closed versions of their powerful models (despite the word "open" in "OpenAI"). You pay them for chat and/or API access, and they are allowed to look at everything you do.
How should I reconcile this claim with 'Google Vertex' and 'Amazon Bedrock' being offered as providers with zero data retention on https://openrouter.ai/anthropic/claude-opus-5.5#providers?
unless you're comparing the most extreme specific harnesses to claude code that's just false.
..and I have a lot of disdain toward Anthropic , & I think GLM is great (although too expensive outside the z plans), but it's just wrong by every scorable metric.
I tried it for a day running everything twice, once with opus 5.5 and once with deepseek. If you notice no difference... I don't know what to say.
I think maybe it's people working on tiny self contained things where code quality doesn't matter that much? But in a big codebase, the open models just aren't good enough to make good changes
I'd move in a second if it weren't so. But it looks like it will be about 5X more expensive for me. So.
Your token usage absolutely scales with parallel agents. So.
For now.
I like hyperbole as much as the next guy. And I'm cynical by nature. But this just seems like a wildly pessimistic view that gives literally zero charity to anyone working at or leading a frontier lab. And the fact the author goes on to say that they "already knew this" just points to confirmation bias.
The rest of the article is actually pretty decent and encourages people to investigate and run open weight models. That is a totally reasonable position that I don't think requires the heavy-handed in-group signal that precedes it.
Seems to me we crossed that line a while ago.
But now I have two new issues:
1. I have fomo from not being face-in the bleeding edge of things.
2. I am ambivalent about z.ai. China -based company is problematic, but they should get my money since they do good for the world. Switching to Mistral for GLM makes my problem inverse since the money doesnt go exactly to to the right place but I trust a french company.
Lets hope the near future resolves our need for personal VRAM and ultimately full self hosting. Coding must be free again. This is the way.
You either practice enough nuance to know that labeling a company like these "evil" is doing a disservice to the critical thinking needed in this discourse, or any company in this space (and likely just any company in general) is fundamentally evil. Picking and choosing based on your feelings is pointless.
>They are chinese, but they should get my money since they do good for the world.
Love the implication that them being Chinese is the moral dilemma lol. Also love that a company just operating in a competitive space is doing "good for the world" because... what, they offer this service cheaply? lol
The moral dilemma is not with ethnicity but rather about the CCP.
cheaply is nice, but the main point is that GLM is open weights. A resounding YES: z.ai are doing the good for the world.
At my own stuff I can do what I want, at work I have no agency and it's not looking like anything is changing soon.
I think the only thing that matters is control.
Muse is run by Zuckerberg, which is a red flag. Grok is run by a racist.
I guess I could go with the Chinese models - I'm not working on anything dangerous.
Bro come on.
My guess is the average Joe will get rejected, but if you're a big company who cares about trade secrets, etc - of course the major cloud providers will provide a ZDR option.
EDIT: strangely Azure shows up as ZDR on openai's models but not anthropic's: https://openrouter.ai/openai/gpt-6.1-sol#providers
For clarity, I do not think you should use abusive language with Ai because of the impact it has on you, Ai only simulates emotion through language use, there are no hormones at work.
Since you seem to know, it's one thing to not retain any data long term, but is the data used for training when it goes through the system? Does ZDR at Azure mean that in practice?
For all the major providers (except Deepseek), API use is simply never used for training (chat use often is).
My point (and the author's) was that even though they commit not to train, they do still retain data (security audits, etc).
With a ZDR agreement, they commit not to retain anything for any purpose.
I looked into it a couple of months ago and it wasn't a win.
(note: it's important to correctly apportion blame to the amount of power you have. The janitor at openai doesn't really have enough say to be at fault)
I promise you none of them believe that. It takes 10 minutes with an LLM and the right set of questions to figure out that were are maybe like 10% of the way there towards AGI.
I like the odds and am happy to play this lottery. Bring it on!
With all due respect, dude, who cares.
Oh noes! We are not giving the billionaire steering reality without a democratic mandate enough benefit of the doubt. I'm sure they will feel mildly sad now.
It is a valid and admirable goal to defend people that are being shat on online, but keep in mind that you are in fact not a billionaire + it is in fact in your own best interest that people without a democratic mandate do not get to control the world.
Controversially: I agree with them. They're correct to think that. They are, and they are.
At the very least, the former is true even if the latter is murky. I trust them over Facebook and Google and China and open source "raze the commons" labs.
I trust Anthropic more than OpenAI when it comes to ASI catastrophe prevention, but I still trust OAI more than those other parties. But both still might fuck it up.
You say you trust them more than Google, but Google's actually kept their models under much better lock-and-key. But even if we do take for granted that they're more trustworthy, that still doesn't make them actually trustworthy.
Harm reduction is a good thing to apply here. It means picking the least-bad option whenever possible.
Best case: you don't need to use any LLMs.
Second best case: you have the hardware on hand to run a local model, and can work within those constraints.
Third best case: you use an open-weight model provided by an inference provider in your country that uses renewable energy.
And on down the line until you get to the worst possible actions: giving money to anthropic, openai, or grok.
OpenRouter lists a few dozen providers for GLM 5.3 and GLM 5.3 Flash. Pick a provider in your country and keep at least some of that money from going to china (not counting all the hardware).
verdverm•57m ago
Great piece, thanks for putting some of my thoughts in organically sourced words.
rnd0•54m ago
Of course there are differences with the impact AI is having economically and the devastation wrought by datacenters -but that doesn't matter to the media or the politicians; they simply want their greenbacks, specifics and effects be damned.
verdverm•50m ago
rnd0•36m ago
Everything will have to become local and community-based, so I applaud your efforts on that -that's one of the few things I see that can make any sort of difference in the long term.
verdverm•3m ago
Timothy Miller, Adam Tooze, and Sherry Turkle are my favorites right now. Just started Daron Acemoglu's new book this week, strong contender for list expansion.
lokar•48m ago
BeetleB•28m ago
Nah. Back when Uber was boasting about all the laws they were flouting, I was pretty anti-Uber, and I was among the minority here on HN. Most of the comments were about how problematic the taxi system is, and how it's great Uber was breaking the cartel.
lokar•20m ago
verdverm•13m ago
lokar•7m ago
tensor•47m ago
But yes, great piece. It really captures why I use open models on european hardware. It's like we haven't learned by now that competition is good and we should be encouraging a diverse and open ecosystem.
verdverm•44m ago
I've observed similar with my negative comments towards Big Ai. Anecdata and vibes caveats
senordevnyc•28m ago
verdverm•20m ago
senordevnyc•10m ago
jpadkins•16m ago
Also I don't understand how 'the media' is separate from social media. People decided to shift their media consumption from magazines, cable tv, corp news outlets to social media. Aren't they all the media (in function)?