Illustration: Allie Carl/Axios
Andreessen Horowitz is pushing back against AI transparency proposals, calling them unconstitutional and uniquely harmful for startups.
Why it matters: Lawmakers in statehouses and on the Hill are introducing bills that run the gamut from requiring companies to share training data sets to providing AI warning labels that content might not be accurate.
Driving the news: a16z is countering with its own policy proposal, shared exclusively with Axios, that the VC firm says would protect smaller companies.
- a16z says startups will have a hard time taking on the "extraordinarily expensive" compliance burdens that come with mandated disclosures, which have penalties associated with them if companies share wrong information.
- The mandates could also violate the constitution, the group says, because the Supreme Court has stated that "unduly burdensome disclosure requirements might offend the First Amendment by chilling protected commercial speech."
- a16z is pitching this idea to policymakers around the country given the high volume of bills proposed.
What's inside: a16z is proposing requiring base model developers to provide a fact sheet they've dubbed "AI Model Facts" that includes information like the date a model stopped training so that people would know the limits of its knowledge.
- They suggest the fact sheet could include the company's terms of service and the model's release date.
- The fact sheet could also show how models perform against standards once they've been "well-established," a16z says, without specifying who should set them.
The big picture: Smaller tech and AI companies that have different business models than the biggest players have been flexing more muscle on policy to ensure larger companies that can afford compliance don't dictate laws.
