Monday, November 10

Tag: Artificial Intelligence

Chat GPT 4o just got some kind of upgrade. It’s now the first model I’ve ever tested to pass the 4×4 grid test, the test that has brought every other model to its knees.
News Feed, Reddit

Chat GPT 4o just got some kind of upgrade. It’s now the first model I’ve ever tested to pass the 4×4 grid test, the test that has brought every other model to its knees.

The raw conversation, zero shot: https://ibb.co/JQWC2XJ https://ibb.co/b54j4d3 https://ibb.co/JQbfmpt https://ibb.co/2Wv7tHs In short, the challenge is for the AI to create a 4x4 alphanumeric grid that is filled with interesting relationships and secrets and creative references buried inside of it. It's a pretty intense challenge that every model has failed spectacularly up until now. Most fail to follow the basic instructions and their grids aren't alphanumeric and they include all manner of symbols in them, even when repeatedly asked not to. For those that do manage to finally create a grid (including ChatGPT before tonight in previous tests) they end up hallucinating all sorts of things about the grid they just created. They'll claim numbers are there which aren't, etc. So my standards...
The AI Report