Trusted News and Notes, 2023-03-30
Happy Thursday! I’m tentatively planning to do one shorter newsletter and one longer analysis piece per week. Thanks for reading!
News and Notes
Business and technology leaders urge delay in development of AI systems
A large group of AI researchers and technology leaders have signed an open letter calling for a six-month pause on the development of AI systems that are more powerful than GPT-4. The letter warns that such systems pose profound risks to society and humanity, that they should be subject to independent review and oversight and asks governments to intervene if the AI labs do not comply with the pause request.
The media ecosystem being what it is, instead of a meaningful debate on the points in the letter, it’s mostly been personal attacks on the motivations of the signers. I’ve been meaning to write a long-form piece on my views on AI safety, so I won’t go into it much here. Broadly speaking, I think the AI safety advocates have some good points about a topic that is still very uncertain, but sensationalist and zealous tactics are going to marginalize their influence with society. For further reading, see these thoughtful articles by Gary Marcus or Arvind Narayanan.
OpenAI adds plugin support to ChatGPT
OpenAI has added a plugin system to ChatGPT that allows it to directly access content from outside its trained knowledge base. There are OpenAI-designed plugins supporting internet browsing and programming, as well as an open API allowing third-party developers to build their own plugin for use in ChatGPT; launch partners include Expedia, Instacart, OpenTable, Wolfram, and Zapier.
This is still in alpha testing and I’m not one of the chosen few who have made it in off the waitlist, so I can’t talk about how it works yet. From watching videos, though, this has some significant business implications.
All of the startups that were building businesses off of the OpenAI API are sweating right now. True, a chatbot isn’t the best interface for many business use cases, but it’s good enough, and interestingly will compete directly with Microsoft’s offerings. ChatGPT + web search, once freely available, will compete very well with Bing AI, and ChatGPT + programming will challenge GitHub Copilot.
Stratechery argues that this is the first step in OpenAI transitioning from an infrastructure company to a platform tech company, and will eventually limit the API’s they offer. I don’t think so. There’s no tool of comparable quality out there right now, so the entire subsidiary AI market will end up dependent on OpenAI’s API’s. That might not be as short-term profitable as selling subscriptions to ChatGPT, but it has a much stronger long-term outlook.
Military organization calls for the creation of a U.S. Cyber Force
The Military Cyber Professionals Association (MCPA), an association of U.S. military cyber officers, published their own open letter calling for the establishment of a U.S. Cyber Force as a new uniformed service. (Full disclosure: I am a somewhat-lapsed member of the MCPA but was unaware of this letter being published.)
I’m not a fan of this idea, which has been kicking around for over a decade. The standard argument in support is “Well, every other domain of conflict has an aligned Service, so the cyber domain needs their own too.” That makes sense on the surface and is emotionally appealing to existing uniformed Cyber personnel from the military services who dream about what a Cyber Force mission would look like, or possibly the uniforms. (The Space Force got Battlestar Galactica as a uniform inspiration; we need some Matrix or Tron or something in there.)
In reality, though, Cyber is more akin to a supporting function than an actual warfighting domain, so this argument doesn’t compute. All of the physical domains have their own associated cyber domain; ground forces have their own “internet,” as do ships, as do aircraft, etc., and they all need to be defended/attacked as necessary, which requires domain-aligned cyber forces for best effectiveness. Add to this that USCYBERCOM exists and already coordinates the service cyber teams, and it’s hard to see how a lot of benefit would accrue from this move.
One Recommendation
Daniel Miessler’s Unsupervised Learning blog was one of my key motivators for starting this project, and I’m totally ripping off taking inspiration from his own weekly newsletter. Like me, he’s a security guy who also has lots of thoughts on technology and society. I periodically hang out in his member-only Discord channel, so stop by.
In Closing
I’ve been using GPT to help me learn CSS (the “layout” language of web design), but spatial relationships are not my strong point.
Standard disclaimer: All views presented are those of the author and do not represent the views of the U.S. government or any of its components.