undefined | Better HN

0 pointsbigyabai9d ago0 comments

OpenAI and X's open-weight LLMs are terrible. Both of them deliberately nerfed their releases to prevent competition with actual SOTA models.

I consider all three companies to be demonstrably hostile to local inference. Anthropic at least doesn't lie about it.

0 comments

rvz9d ago

> OpenAI and X's open-weight LLMs are terrible.

They are open weight models that they have released and we know they are not the only ones out there to use and it is better than none.

> I consider all three companies to be demonstrably hostile to local inference.

Google isn't hostile to local inference and also release open-weight models for local inference.

> Anthropic at least doesn't lie about it.

You literally get nothing for defending Anthropic who doesn't care about you and continues to lie to you about "safety" because the truth is local models directly compete against them.

bigyabaiOP9d ago

> we know they are not the only ones out there to use and it is better than none.

Oh, my apologies. It's not fair to compare shoddy upstarts like OpenAI and X to Deepseek or Qwen or GLM or Kimi or MiniMax for not releasing their best model weights as soon as they're out. I'm just not appreciating all of what OSS-120b and Grok 2.5 have to offer to the homelabbing community.

> Google isn't hostile to local inference

I didn't include Google among them. "all three" meant X, OpenAI and Anthropic.

> You literally get nothing for defending Anthropic

My quote was not "Anthropic at least doesn't lie" full stop, was it? They don't lie about being a frontier local model provider.

OpenAI hasn't released a genuinely boundary-pushing model since GPT-2. That's a contemptible output for a trillion-dollar research organization, say what you will about Anthropic's corporate policy.

j / k navigate · click thread line to collapse