Skip to content

LLMs and Tool Use – Schneier on Security

Last March, Microsoft researchers announced a plan to compile millions of APIs into a compendium accessible to large language models (LLMs). This milestone in the race to teach LLMs how to use tools could supercharge the potential of AI. While LLMs can recommend pizza toppings or draft dialogue, they can’t place orders. However, a tool-using AI could do it all. The ability to manipulate tools also allows AIs to improve themselves and provide better explanations for their reasoning. However, tool use comes with risks, and caution should be exercised in granting AIs access to certain combinations of tools. OpenAI and Meta are leading the way in enabling tool use, but regulators must consider AI-specific regulations and prepare for the societal impact of tool-using AI. In the best-case scenario, tool-using AI could accelerate fields like drug discovery and provide more opportunities for citizen engagement. It is important to be prepared for the power and potential of tool-using AI.

Key points:
– Microsoft plans to compile millions of APIs accessible to LLMs, allowing them to use tools and supercharge AI potential.
– Tool-using AI can place orders, improve itself, and provide better explanations for its reasoning.
– Caution is needed in granting access to certain combinations of tools to mitigate risks.
– OpenAI and Meta are enabling tool use, but regulators must consider AI-specific regulations.
– Tool-using AI can impact society positively, such as accelerating drug discovery and enhancing citizen engagement.
– It is crucial to be prepared for the power and potential of tool-using AI.

Leave a Reply

Your email address will not be published. Required fields are marked *