Also no political censorship, and it doesn’t report you to the police when it suspects you did something illegal. That really happened, by the way.
Now I want a local LLM server :( $25k is just so much money.
Also no political censorship, and it doesn’t report you to the police when it suspects you did something illegal. That really happened, by the way.
Now I want a local LLM server :( $25k is just so much money.
No political censorhip? Try asking Chinese models about Taiwan.
Some censorhip and bias towards certain ideology is baked into those models.
While it is not a problem for programming tasks, I would never use it for certain other tasks.
If you already have a machine, throwing an r9700 in it is one of the cheapest ways to get qwen3.8 27B Q6 with 100k+ context working.
Sadly, the r9700 has gone up in price, but it's still the cheapest route. I use It on an old x99 Linux desktop and it achieves 50 tps decode and 1000+ tps prefill.
I'm building a local LLM server myself. For about 8k€ you can have 170tps, ~6000 prefill.
A lot more with Haiku equivalents (Qwen3.8 27B) or gemini. It's not frontier, but not that far away.
>and it doesn’t report you to the police when it suspects you did something illegal
This is an alignment issue. OpenClaw with a local LLM with an email you setup for it could contact the police.
It will not do that unless you specifically teach it to do that, or you use an Anthropic model. The issue in this case was Anthropic's safety monitoring built into the model. They record everything and flag potential illegal activity for human review. This occurs even over the API. I do not know if OpenAI does this, but I would not be surprised. I think the Chinese models are far less likely to do this (unless one resides in China).
MacBooks with large RAM and optimized for local models should be Apple's medium term goal.
It's a good first step. They're not great but they kind of work for many tasks. I think the small models can be useful in context-specific tasks but the friction is high right now because the user is forced to determine whether Siri is good enough, or a better LLM is required per task. Ideally Siri would be smart enough to intelligently route. Apple claims that's how it works but even their cloud processing is far behind frontier models.
Worryingly, they're taking a walled garden approach, and they don't want to expose user data to competitors. After the monumental failure of Siri over more than a decade, I don't trust Apple to execute competently on AI.