This weekend there was a lot of conversation regarding the need to “slow down” or “pace” frontier AI development. This started when Anthropic’s Dario Amodei published an essay titled “We Must Pace the Frontier”. The essay discussed the need to pace advanced model development after a summer of faster-than-expected progress. There have also been several security incidents that he pointed to. The first thought I had when seeing this was to shrug and sort of think “here we go again”. Amodei has a history of big, vivid warnings and that can sort of dull the affect of his point. It should be noted that this came on the heels of former Anthropic employee Jacob Coxon’s public resignation and media tour claiming that the labs are “gambling with our lives” and that people inside them “earnestly believe that it could kill us all by the end of the decade.” Anthropic’s Evan Hubinger said he puts the number on it as “more than 10% this decade.”
By the end of the day, Elon Musk (SpaceXAI), Sam Altman (OpenAI), Demis Hassabis (Google DeepMind) all posted some form of agreement. This at least made me think a little deeper about the situation. Like most things, there is a lot of passionate arguments on both sides of this debate. I tend to land somewhere in the middle. I believe that the leaders of these labs believe in the message. That makes me think they see something we don’t. I also believe that they have the ability to self pace. I am skeptical about some of the motivation behind some of this because there is a lot of capital tied up in these labs. There is also an ongoing threat to those investments with the rise in open source adoption. At any rate, this will be top of mind in Washington DC. Hopefully these conversations can happen in a sensible and factual way.
It is important to note that all of this is concerning frontier AI development. Specifically, this is the research and training that happens in the labs. It is the technology that we may not get our hands on for a while, if ever. This is not the chat bots and agents that we are using or helping our customers setup. Those have their own concerns, that we often talk about, such as alignment around your data security policies.
My message is unchanged. AI is a powerful tool. The person using it needs to do so responsibly and businesses need to have a strategy for implementing these tools safely. Businesses also need to understand and mitigate the real risk around employees bringing their own AI tools that aren’t aligned with your policies.
The image associated with this entry was AI generated.