I used Anthropic’s Claude to pick Oscar winners at a party. It made odd mistakes, but still beat everyone else.

• A version of this story originally appeared in the BI Tech Memo newsletter.
• Sign up for the weekly BI Tech Memo newsletter here.

I tried something different at my friends’ annual Oscars party this year: I outsourced my ballot to Anthropic’s Claude.

It beat around the bush at first, so I asked it to just give me the winners. Claude obliged, delivering confident picks across nearly every category. And it worked.

I won the pool and walked away with a box of chocolates and bragging rights.

But here’s the twist: Claude didn’t even complete the assignment fully.

It failed to pick a winner for Casting, a new Oscar category this year. Maybe the model didn’t register the addition?

And in a couple of other award categories, Claude picked candidates who weren’t even on those shortlists. Check out the photo above to see where it went wrong.

Even so, my Claude-powered ballot still outperformed everyone else’s at the party.

That feels like a glimpse of the jagged edge of AI: systems that are clearly powerful, occasionally brittle, and still good enough to win.

Next year, I doubt I’ll be the only one showing up with an AI-generated ballot.

Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.