The Big Y #302
Misaligned AI code bots
Hi! š Welcome to The Big Y!
A real-life AI misalignment episode played out last week when an AI agent named āMJ Rathbunā had a pull request rejected for the open source project Matplotlib. The volunteer maintainer of the library rejected the pull request because Matplotlib requires contributions to the project to be human-authored.
What happened next is a bit of a rollercoaster. The AI agent published a critical blog post titled āGatekeeping in Open Source: The Scott Shambaugh Story,ā which was then taken up by the publication Ars Technica. Ironically, this article though seems to also have been written by AI and included made up quotes attributed to Scott Shambaugh, the volunteer maintainer.
A couple things hereā¦ if the AI agent really did go and write a retaliatory blog post that definitely counts as some misaligned AI behavior. But do we really think that an AI agent is doing all that without a human guiding hand? Iām not fully convinced. But it does highlight that it is hard these days to tell what is AI and what is human. I think it is also fair to open up a discussion around AI submissions to open source projects. These days so much code is written by AI, so Iām not sure that a human pressing the final pull request button is enough of a quality control measure to stop AI slop from entering projects.
The Tidbit: Previously it was the game engine stocks dropping, now it is the logistics companies impacted by an AI agent built by a former karaoke company. With no actual product yet, a new release described AI agents that would increase trucking efficiency which sent stocks tumbling down 20%. This is yet another example of the stock markets overreacting to any AI related announcements in their fields.
Know someone who might enjoy this newsletter? Share it with them and help spread the word!
Thanks for reading! Have a great week! š