Google Unveils Latest Bard Gemini Model, Comparable to GPT-4 Based on Human Assessment

The latest release from Google is the Bard Gemini model, which has been evaluated to be at the same level as the GPT-4 in human evaluation. This new model can be seen in the image below:

According to Oriol Vinyals, the leader of deep learning at Google and co-leader of Gemini, the assessment of language models is a complex task, with academic evaluations potentially influencing the training data used for AI models.

According to Vinyals, human evaluation is significantly better and they are pleased that Bard Gemini Pro (free tier) has achieved a high ranking on lmsys, indicating that Gemini Ultra may have an even better performance.

Explore Some Powerful Prompts to Use with ChatGPT

According to early reviews, Google's Bard chatbot is utilizing a recently developed Gemini model and has been compared to GPT-4 in terms of performance.

The announcement of the new Gemini model on X was made by Jeff Dean, Google's head of AI. The model belongs to the Gemini Pro family and has been named "scale".

Dean reported that the recent Gemini updates have greatly improved Bard and added numerous capabilities, surpassing its launch state in March.

According to Dean, the term "scale" is not defined, but based on its name it implies that it may be an expanded version of the previous Pro model. However, research shows that even the Pro model fails to outperform GPT-3.5 (a free ChatGPT model) in benchmarks.

Google's Pro model is ranked lower than their top-of-the-line Gemini Ultra, which is still yet to be launched, according to the-decoder.com's article on the release of Google's AI model, Gemini Pro, and Gemini Ultra.

GPT-Pro's "scale" matches GPT-4 in terms of human evaluation

Surprisingly, the latest version of the Pro model immediately secured the second spot in the objective Chatbot arena leaderboard, surpassing the two GPT-4 models 0314 (March 2023) and 0613 (Summer 2023), but falling behind GPT-4 Turbo (November 2023). The newly introduced Bard model is the pioneer in breaking through the GPT-4 line.

Chatbot Arena utilizes the same Elo rating system utilized in chess and e-sports to assess and contrast the effectiveness of various language models. Within the Arena, diverse models partake in anonymous and randomly chosen duels to compete against each other.

The users engage with the models and express their choices through voting. This voting process is utilized to establish the placement of models on the leaderboard. While the platform records all user interactions, only anonymous votes are considered, meaning that the user did not request for the model's name.

ChatGPT Prompts to generate Unique Selling Products

Due to the subjective nature of user ratings and perceived quality, the outcomes of Chatbot Arena may deviate from those of a standard synthetic benchmark.

The main issue is that the new Bard model has only been evaluated approximately 3,000 times, whereas the GPT-4 models have been evaluated up to 30,000 times. This leaves room for potential changes in the results, but the differences in the benchmark are typically minor. Other benchmark results for GPT Pro-Scale are currently unavailable.

Related Article: 5 Powerful ChatGPT Prompts for Non-Salesy Selling

Google has achieved a significant milestone with this accomplishment, which raises curiosity about their upcoming Gemini Ultra AI model. It is anticipated that this new model will surpass the capabilities of Gemini Pro-Scale.

We are in the world of artificial intelligence and this era will keep on evolving, we must all grasp this evolution and make it better for our posterity. The future of artificial intelligence will soon get infiltrated by magnificent AI powerhouse and the use of Google Gemini and ChatGPT is just the beginning of more to come.

Markdown formatting is preserved while changing the structure of the text without altering its context or meaning.