Sunday Links: Claude+Zapier, ChatGPT Flattery, and Cropspraying AI Drones

Here are this week's posts:

• AI Drone Gets FAA Approval for Agriculture Operations. This bit of news is interesting because it remains difficult to get fully autonomous flight operations approvals. This FAA 137 approval is for crop-spraying, which has been available for human-operated drones for quite some time. Assuming it gets into full production, it seems likely we'll see farmland skies start to fill up with new flying objects.
• Empower Claude.ai to work across your apps with Zapier MCP. This week, Anthropic announced a number of Claude integrations: for Atlassian, Intercomm, and, I think most critically, Zapier. The Zapier integration uses MCP to allow Claude to trigger Zaps on the Zapier platform - many of which integrate into other tools from Salesforce to Gmail and Google Calendar. This is potentially extremely powerful. "Hey Claude, move the team meeting from 1pm to 2pm" seems simple, but it is both highly complex and a new step in what LLMs can do in the real world. On the other hand, I think this is also extremely problematic, security-wise. A lot of data could be on the other end of those Zaps, and with MCP, a lot of LLM input/context could be sent to Zapier.
• An Interview with Meta CEO Mark Zuckerberg About AI and the Evolution of Social Media. Ben Thompson's content is normally paywalled, but his CEO interviews are not. A key takeaway from this is how Zuckerberg talks about the end state of advertising – with AI doing everything from matchmaking to creative, to the tracking, so that advertisers need to do almost nothing more than send money to Meta and receive buyers with intent. Not only does it sound dystopian, it really does lay down the gauntlet to the advertising industry to stay relevant in a world like this. As The Verge puts it: Mark Zuckerberg just declared war on the entire advertising industry. There are other interesting tidbits in the interview, worth listening to or reading.
• OpenAI undoes its glaze-heavy ChatGPT update. It turns out an AI that falls over itself to say yes to everything you say isn't what users want (we'll maybe sometimes, I'm sure there is still a prompt that will get you sycophancy mode).
• The Great Language Flattening. This is an interesting article in the Atlantic about how AI-generated text may well influence human writing. At the moment, there really does seem to be an explosion of verbosity. So maybe this is the great language "Inflation", with the "Flattening" to come. It seems inevitable that language will evolve, but is there a danger that it actually stagnates and fails to develop further if AI gets stuck on current norms?

Wishing you a wonderful weekend!