Anthropic unleashes Claude Mythos 5, its most powerful AI to date, with some strings attached.
Anthropic, the AI startup backed by Elon Musk, has made its most powerful language model, Claude Mythos 5, available to the public for the first time. However, don’t expect to get your hands on its most advanced features right away – the company is keeping the high-risk capabilities behind guardrails, or virtual fences that limit how the model can be used.
What this means: You’ll still get to try out the AI’s impressive language generation capabilities, but you won’t be able to access the truly cutting-edge features that could pose risks to users or the AI itself.
### Guardrails for Safer AI
To mitigate potential risks, Anthropic has implemented a 30-day data retention requirement for both Claude Fable 5 and Claude Mythos 5. This means that any interactions with the AI will be logged for at least a month, allowing Anthropic to track and defend against novel attacks. The company also wants to monitor all human access to sensitive data, which is presumably stored securely behind the guardrails.
The exact nature of these high-risk capabilities is unclear, but it’s likely that they involve advanced features such as self-modifying code or the ability to generate malicious content.
### Claude Mythos 5: A Beast of a Model
Claude Mythos 5 is a behemoth of a model, designed to tackle complex tasks such as natural language understanding and generation. It’s the latest iteration in Anthropic’s Claude line, which has been steadily improving since its inception. The new model boasts massive improvements in performance and scale, making it a serious contender in the AI landscape.
The company claims that Claude Mythos 5 has surpassed its human-level counterparts in certain tasks, such as:
204 million parameters (a measure of the model’s complexity)
10 times the performance of Claude Fable 5, the previous publicly available model
100 times the performance of the original Claude model
### What’s Next?
Anthropic’s cautious approach to releasing high-risk capabilities is a departure from some other companies in the AI space. However, it’s also a sign of the company’s commitment to responsible AI development. By keeping the model behind guardrails, Anthropic can ensure that users are protected from potential risks while still allowing them to explore the AI’s capabilities.
As AI continues to advance at breakneck speeds, it’s clear that companies like Anthropic will play a crucial role in shaping the future of the technology.
