The UK government is pushing companies such as OpenAI, Anthropic and Google to explain the inner workings of their large language models (LLMs). While the code for some models is public, models such as GPT-3.5 and GPT-4 are not, and OpenAI is very reluctant to share many details.
The UK is preparing to host a new global artificial intelligence summit, which will bring together governments, companies and researchers to examine the risks posed by artificial intelligence and discuss how to reduce them.
One of the reasons companies are reluctant to share their internal data is that such behavior could reveal proprietary information about their products. If malicious actors know more inside information, it can also make artificial intelligence models vulnerable to cyberattacks.
According to the Financial Times, one of the things the government wants to check are model weights, which define the strength of connections between neurons in different layers of the model. Currently, AI companies are not required to share these details, but there have been calls for greater transparency in this regard.
The UK will hold its first summit at Bletchley Park in November. Bletchley Park holds an important place in computing history as it was here that Nazi messages were decrypted. The Turing test related to artificial intelligence is named after Alan Turing, who also cracked codes there.
The Financial Times noted that Google's DeepMind, OpenAI and Anthropic all agreed in June to open their models to the British government for research and security purposes. Unfortunately, the parties did not agree on the scope and technical details of the opening at that time. Now, the level of openness required by the government is quite high.
Ultimately, for the summit to be successful, attendees must fully understand how the models work so they can better understand their dangers. Whether they have adequate access to the models is another question.