Google’s recent announcement of its new artificial intelligence model, Gemini, has shaken up the tech world. Presented as a potential competitor to OpenAI’s GPT-4 models, the Google Gemini demo video shows spectacular advances in AI. But the controversies surrounding Gemini’s preparation and actual performance raise questions about its true capabilities.
Gemini’s promise in the demo video
- Google has unveiled Gemini, an artificial intelligence model claimed to be superior to OpenAI’s GPT-4.
- Gemini is a multimodal model capable of processing text, images and video simultaneously and in real time. In the video, it looks as if the AI sees through the video and responds instantly by speaking.
- Emphasis on Gemini’sconversational intelligence, with the ability to conduct complex, nuanced dialogues in real time.
Controversies and doubts
Despite the promises, revelations about the preparation of Gemini’s demonstration video have raised doubts about the real capacity of this AI compared to the presentation of its technological advances in the video :
- Gemini’s response time has been accelerated in video demos.
- The questions put to Gemini would have been carefully prepared in advance to guide the answers. The questions would have been put to Gemini via photos and text, not video. The text responses were read by a voice, then edited.
- Some have pointed to the lack of transparency in the benchmark methodology, with concerns about real comparability with GPT-4.
- Questions about the representativeness of the demo for real, everyday use scenarios.
So what is Gemini’s real performance?
Independent studies, including those conducted by Carnegie Mellon University, suggest that Gemini Pro’s performance is inferior to that of GPT-3.5, despite Google’s claims:
- Difficulty with complex mathematical reasoning and sensitivity to multiple-choice answers.
- Lower performance compared with GPT-3.5 Turbo, especially in terms of accuracy.
- Examination of Gemini’s ability to provide contextualized and accurate answers outside prepared scenarios.
Implications for the future of AI
This raises fundamental questions about the credibility of tech giants in the rapidly expanding field of AI. Trust and transparency are key issues for the future of this technology:
- The importance of transparency and authenticity in communicating advances in AI.
- The need for benchmarks and standardized methodologies for a fair and reliable evaluation of AI models.
- Discussion of the need for stricter regulations governing the presentation and advertising of AI capabilities.
In conclusion, although Google’s Gemini potentially marks a step forward in the field of AI, the controversies and doubts raised highlight the importance of a more transparent and authentic approach to gaining the trust of the public and the scientific community.