Hacker News new | past | comments | ask | show | jobs | submit login

A couple of days ago I gave both of them a prompt similar to:

"Create a map with tect using only the letters B,W,S. The map should be a 20 line by 20 colum text . Each line should contain only B,W or S letters were W are walls, B is blank/background space and S represent "stairs" to outside of the room...."

The query was a it longer with more specs.

Neither ChatGPT nor Bard could give me a good answer. They used other letters , they made 21 or 19 chars lines. They made 5 or 6 line maps. They basically made a mess.

That's my current test for reasoning, analysis and intelligence for these things.




They are both pretty bad. I ask about templates for CI/CD and they imagine parameters that don’t exist, and no amount of wrestling it around can suppress this. People like to cherry-pick examples where they work great and then proclaim it’s the best thing since sliced bread, but it’s just simply not.


My favorite so far is Copilot writing code with variables like “testO1”.

Took me an hour to figure out why it didn’t work.

O != 0


(facepalm emoji)




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: