Hi All,
GPT-4o is surprisingly good. Admittedly, I expected some progress from OpenAI, but GPT-4o is remarkable for two reasons. It is very human, frighteningly human I would even say. There's a reason it earned the nickname βherβ in reference to the movie from Spike Jonze. Besides, it's probably a model with fewer parameters, because it responds surprisingly quickly. That is, it also looks like OpenAI has mastered a definite improvement in the efficiency of smaller models. Or they have found a way to speed up inference by a large model. When we add multi-modality to this, we have the full picture.
It looks to me that OpenAI is close to achieving what Gemini wanted to be. Do you remember how Google presented a βfakeβ presentation of its multi-modal model last year? It looked too good to be true, and it turned out that the model's capabilities were βboostedβ during video editing. And now we've reached the point where a live model has similar capabilities.
I did a small test of the ability of GPT-4o to program in ABAP. You are invited to watch it. I also recommend previous episodes:
LLAMA3: https://youtu.be/kTs5LZFhiwc
GPT4 and Gemini: https://youtu.be/vkzrwIof88M
Bluesky
