OpenAI vs. Google: ready, go? A Comparative Study of the GPT-4 versus Gemini and the TPU v5p Models
There is an important thing that Google takes into account as well as the actual efficiency of the model. It was trained on Google’s own Tensor Processing Units and is both faster and cheaper to run than Google’s previous models like PaLM. Alongside the new model, Google is also launching a new version of its TPU system, the TPU v5p, a computing system designed for use in data centers for training and running large-scale models.
So, let’s just get to the important question, shall we? OpenAI’s GPT-4 versus Google’s Gemini: ready, go. This has been something that was on the mind of Google for quite some time. Hassabis says that they have done a thorough analysis of the systems side by side. Some of the tests that were used to compare the two models included the Multi-task Language Understanding benchmark, which compared two models with their ability to generate Python code. Hassabis says that he believes we are substantially ahead on 30 out of the 32 benchmarks. “Some of them are very narrow. Some of them are larger.”
Right now, Gemini’s most basic models are text in and text out, but more powerful models like Gemini Ultra can work with images, video, and audio. Hassabis says that it is going to get even more general. “There’s still things like action, and touch — more like robotics-type things.” He says over time, it will become more possible for Gemini to become grounded and accurate in the process. “These models just sort of understand better about the world around them.” These models still hallucinate, of course, and they still have biases and other problems. The better they will get, the more they know, says Hassabis.
The true test of Gemini’s capability will come from everyday users who want to use it for more than just benchmarking. The new code generating system AlphaCode 2 is an improvement over the original Alpha Code, which uses 50% more code and performs better than 85% of competitors. But Pichai says that users will notice an improvement in just about everything the model touches.
Demis Hassabis has never been shy about proclaiming big leaps in artificial intelligence. He became famous in 2016 after the AlphaGo bot taught it how to play the board game Go with tremendous skill and ingenuity.
OpenAI Enhancement of GPT-4: On Multimodal Approximations of Artificial Intelligence Models, and Implications for the Chatbot
“Until now, most models have sort of approximated multimodality by training separate modules and then stitching them together,” Hassabis says, in what appeared to be a veiled reference to OpenAI’s technology. “That’s OK for some tasks, but you can’t have this sort of deep complex reasoning in multimodal space.”
OpenAI launched an upgrade to ChatGPT in September that gave the chatbot the ability to take images and audio as input in addition to text. OpenAI has not disclosed technical details about how GPT-4 does this or the technical basis of its multimodal capabilities.