Recently, the competition between Google and OpenAI has intensified once again. Just one day after the new GPT-4o topped the AI leaderboard, Google launched its latest experimental model, Gemini-Exp-1121, quickly reclaiming the championship title. Just a week ago, Google had released Gemini-Exp-1114, indicating that Google is responding to OpenAI's developments very rapidly.
Jack Rae, Chief Scientist at Google DeepMind, stated that this is a "lightning war," suggesting that the iteration speed of post-training is faster than that of pre-training.
According to official information, Gemini-Exp-1121 has made significant improvements in various aspects, mainly reflected in enhanced coding abilities, reasoning capabilities, and visual understanding. Additionally, the model has achieved a level of style control in complex prompts comparable to the current top models, o1-preview and New Sonnet3.5.
In practical tests, Gemini-Exp-1121 outperformed the new GPT-4o in handling comic understanding, providing more comprehensive answers and clearly presenting information using subheadings and bold highlights. In a classic logical reasoning problem involving animals crossing a river, Gemini-Exp-1121 provided a completely correct answer, demonstrating stronger logical reasoning abilities, while the new GPT-4o made some errors.
Meanwhile, OpenAI is also actively developing new features. Recently, code for a "Live Camera" video function was discovered in the latest version of ChatGPT, marking progress in voice and visual recognition. OpenAI users have experienced this capability for the first time while using the advanced voice mode, indicating an intention to expand the application of this feature in the future.
It is foreseeable that next year, the primary mode of communication with chatbots may gradually shift from traditional text conversations to voice and more intelligent agent services, a transition that could be led by the introduction of the "Live Camera" feature.
Key Points:
📈 Google's new model Gemini-Exp-1121 quickly surpassed GPT-4o after its debut, reclaiming the top spot in the AI leaderboard.
🔍 Gemini-Exp-1121 has improved in coding, reasoning, and visual understanding, showing outstanding performance.
🎥 OpenAI is developing the "Live Camera" feature, which may change the way we interact with AI in the future.