idk, I'm starting to worry about the LLM that's producing fundraising emails for Mike Rogers' Senate campaign:
"Ok, our new AI writing app is almost ready to go. Sure, we spelled 'Hemingway' wrong, but at least we captured some of the bleakness in the slogan, right?"
Does Facebook have a problem with AI-generated slop pages? Here's a post about UF football coach Jon Sumrall "clashing" with Nigel Farage about immigration to help you decide!
Been trying out some new open-weight OCR-focused LLMs and found a 4b param model that nails pages like this, including the "vertical" headers. OCR isn't totally solved, but we're getting closer to a scenario where you can get 2 of these 3, one of which is always accurate: cheap fast accurate
Detecting AI-written materials is hard, they say. Tough to be sure, they say. RNC: A person would definitely write a subject line of "Question 1 of 14"
As I sign up for more political fundraising emails, I'm gonna start surfacing the most AI-generated candidate website designs I'm finding. Here's one for Virginia McIntyre, a NY congressional hopeful:
I know I can always spot veterans because they wear this jacket(?) with "U.S. VETERAN" across the back as this clearly AI-generated image from a VA newsletter shows.
Honestly, might be starting to prefer AI writing over this MadLib by, well, mad libs.
I don't often use papal encyclicals in teaching my class on artificial intelligence, but when I do, I use this pope's encyclicals.
What you definitely want to see when you're working on a process that uses AI agents: "The shell can write outside its sandbox."
