As the rapid advancements in artificial intelligence continue to reshape industries and captivate the public, lawmakers in California have decided to take action. A new bill in the state legislature aims to impose a slew of safety requirements on companies developing foundational AI models.
The goal, ostensibly, is to hold these AI companies accountable and mitigate the risks of their creations causing “major havoc.” However, as economist Noah Smith points out, the bill displays a fundamental misunderstanding of how AI development works.
The proposed legislation “demands that AI companies know all sorts of things that their model can and can’t do before the model is trained on the data.” But as Smith rightly notes, “that’s not even possible; you don’t really know what a model can do before it’s trained.”
Even after a model like the wildly popular ChatGPT is fully trained, it remains incredibly difficult to anticipate all of its potential capabilities – let alone its potential misuses. “Does anyone think we know, right now, whether GPT-4 is actually capable of causing catastrophic harm, in the hands of the right villain?” Smith asks. “I don’t think so.”
The good news is that the bill is so poorly written that Smith expects it will have little real impact in slowing down AI innovation in the state. Though the lawmakers’ intentions may be noble, their grasp of the technology appears shaky at best.
Meanwhile, as this legislative drama unfolds, the AI arms race is heating up on a much broader scale. Tech giants are racing to roll out ever-more powerful AI models to the masses – with billions of social media users set to get their first taste of generative AI capabilities.
Facebook parent company Meta, for example, is preparing to unleash its latest Llama 3 AI model to 4 billion users across its platforms. As Paradigm AI authority James Altucher notes, this will introduce “generative AI” to a staggering number of people who have never interacted with the technology before.
What’s more, Llama 3 is a serious beast, with over 400 billion parameters – seven times more data than its predecessor. This allows it to outperform rival models on a wide range of tasks, from coding to general knowledge.
Crucially, Meta has also equipped Llama 3 with the ability to pull real-time search results from the likes of Bing and Google. This could eventually pave the way for Meta to build out its own in-house search capabilities, posing a threat to Google’s long-dominant position.
Of course, Google is unlikely to cede ground without a fight. The tech titan is expected to announce a slate of new AI functionalities at its upcoming I/O conference. The AI arms race is only just beginning, and the implications for industries, businesses, and everyday users are profound.
As concludes, “it’s still exciting times in the AI space — and it’s not too late to get in.” The AI revolution is accelerating, and those who can navigate it successfully stand to reap enormous rewards. California’s lawmakers may be struggling to keep up, but the tech giants are racing full speed ahead.