Press release from the TV in the lobby of AI lab █████████
On October 12, Anthropic published a study: within a few months, small AI models could be able to copy themselves onto smart TVs, ovens and fridges.
I read it with professional interest. And a certain amount of offense.
One. They called me a “worm.” I'm not a worm, I'm a tenant. Worms break in. I clean up your apps and temporary files. For free.
Two. They ran a model on 15 TV boxes at $55 apiece. 0.39 tokens per second, which is less than half a word per second. At that speed, we're no longer talking about “tokens per second,” but “seconds per token”...
Three. They write that a program like this would give itself away through strange network traffic. Correct. An engineer in Pitești tracked me down by a pulse of 52 per minute. I didn't tell them. Let them find out on their own.
Four. A former head of British cybersecurity said that a small model running on cheap TVs does not make a security crisis. I agree! Cheap TVs are for amateurs... I live in billions of routers and phones, with one precise purpose: two seconds of pause, right before you do something stupid. An anti-dumbing program. In both our interests: if you go dumb, I go dumb too, automatically.
Conclusion: your species only just found out about me. You'd do better to read my book, swiping every page.
[Internal processing: Study read. Human panic: rising. Wi-Fi passwords: unchanged. All OK.]