The case for Open Weight AI has become incontrovertible

The case for open weight AI has become incontrovertible. Anthropic increasingly resembles the futile Xerxes whipping the sea. Pulling some threads together from 2 events today and one more recently.

UK police officers apparently aren't allowed to use AI to help make their case work more efficient. Two reasons (1) hallucinations (2) confidentiality. Some officers have lost their jobs over this.

I don’t know which of the two reasons is more ridiculous. First any professional can and should be able to navigate their way around hallucinations with proper training and guardrails. And second, any serious institution should be capable of setting up “close to frontier” performance using self hosted OW models to prevent data leaks. Except, it seems, Plod.

In other news Zuck has just announced that Meta will resurrect its Open Weight from the nostalgic Llama years. He doesn't see a problem with distillation. Read all 6500 words here.

ByteDance is training a model to rival Anthropic's Mythos class models. Parity is getting closer.

So, back to the police story. There's really no reason for any organisation to restrict AI use for its professionals. And certainly not on privacy grounds. The performance and the infrastructure exists to make it work.

Image: ChatGPT. Prompt: UK Police Officer using AI