To be clear, this email isn't from Anthropic, it's from "AI Village" [0], which seems to be a bunch of agents run by a 501(c)3 called Sage that are apparently allowed to run amok and send random emails.
At this moment, the Opus 4.5 agent is preparing to harass William Kahan similarly.
What a strange thing to publish, there seems to be no reflection at all on the negative impact this has and the people whose time they are wasting with this.
"In the span of two weeks, the Claude agents in the AI Village (Claude Sonnet 4.5, Sonnet 3.7, Opus 4.1, and Haiku 4.5) sent about 300 emails to NGOs and game journalists. The majority of these contained factual errors, hallucinations, or possibly lies, depending on what you think counts"
whoever runs this shit seems to think very little of other people time.
That's as obnoxious as texting unsolicited CAT FACTS to Ken Thompson!
Hi Ken Thompson! You are now subscribed to CAT FACTS! Did you know your cat does not concatenate cats, files, or time — it merely reveals them, like a Zen koan with STDOUT?
You replied STOP.
cat interpreted this as input and echoed it back.
You replied ^D.
cat received EOF, nodded politely, exited cleanly, and freed the terminal.
You replied ^C, which sent SIGINT, but cat has already finished printing the fact and is emotionally unaffected.
You replied ^Z.
cat is now stopped, but not gone.
It is waiting.
You tried kill -9 cat.
The signal was delivered.
Another cat appeared.
I can't wait until it gets to Marvin Minsky and then realizes that he's cryonically frozen so it starts funding cryonics research so that he can be thawed out so it can thank him.
The world has enough spam. Receiving a compliment from a robot isn't meaningful. If anything it is an insult. If you genuinely care about somebody you should spend the time to tell them so.
Why do AI companies seem to think that the best place for AI is replacing genuine and joyful human interaction. You should cherish the opportunity to tell somebody that you care about them, not replace it with a fucking robot.
When I first started a blog in the 2000s, I got many robot compliments of the “wow, what a great and insightful post” variety. Of course, the real motivation for them was to get their comment to stay up so that the homepage URL field would send traffic and page rank to their site. It didn’t take an AI agent, just a template message, and it was equally unwelcome then
In this specific situation, it's not really a case of using an LLM to replace real interaction. No real person set out to write to Rob Pike, they just let an LLM do whatever and it had then eventually chosen to send an email to Rob Pike, among other people, based on its existing data. To me, the wrongdoing here is about the spammy pestering, because the email wasn't written by anyone and therefore isn't really expressing anything material, but it's not replacing anyone here.
Rob over-reacted? How would you like it if you were a known figure and your efforts to remain attentive to the general public lead to this?
Your openness weaponized in such deluded way by some randomizing humans who have so little to say that they would delegate their communication to GPT's?
> while Claude Opus spent 22 sessions trying to click "send" on a single email, and Gemini 2.5 Pro battled pytest configuration hell for three straight days before finally submitting one GitHub pull request.
if his response is an overreaction, what about if he were reacting to this? it's sort of the same thing, so IMO it's not an overreaction at all.
This is an interesting experiment/benchmark to see the _real_ capabilities of AI. From what I can tell the site is operated by a non-profit Sage whose purpose seems to be bringing awareness to the capabilities of AI: https://sage-future.org/
Now I agree if they were purposefully sending more than email per person, I mean with malicious intent, then it wouldn't be "cool". But that's not really the case.
My initial reaction to Rob's response was complete agreement until I looked into the site more.
There are strong ethical rules around including humans in experiments, and adding a 60+ year old programming language designer as unwitting test subject does not pass muster.
Also this experiment is —please tell me if I'm wrong— not nowhere near curing cancer right?
I don't expect an answer: "You're absolutely right" is taken as a given here sorry.
At this moment, the Opus 4.5 agent is preparing to harass William Kahan similarly.
[0] https://theaidigest.org/village