OpenAI o1-preview: Is this AGI?

Day 257 / 366

Late last night OpenAI finally came out with a new foundational AI model. It’s named o1, and it's supposed to be amazing at reasoning.

https://openai.com/index/introducing-openai-o1-preview/

And unlike their other recent releases which were always “coming soon”, something that I had criticized them on, this one was available for ChatGPT plus users almost instantly. I was able to test it out today as well.

What’s different about these new models is that they have an additional “thinking” step before they answer your question. This means that they are definitely a step up from the plain LLM that just works as a text completion model.

The model even describes its thought process. For instance, here’s how it answers the famous strawberry question

People all over Twitter have been going crazy about this new model. Apparently it’s performing well on coding benchmarks. This is something that I am excited about. I would like to see how good this thing is in coding. GPT-4 looked promising at the start but failed in daily use. I’ll need to spend at least a few days playing around with this before I can get to a conclusion