Google Gemini: Marketing Tool or Multimodal Power?

Introduction

Google Gemini, the next fearsome AI model, has received a lot of attention after a video presentation of its alleged multimodal capabilities. However, observers discovered that certain portions of the film contradicted Gemini’s actual performance, prompting concerns about the company’s marketing strategy. This article will look at the expectations and actuality of Google Gemini. More about Bard vs. Gemini read in our blog post What’s Changed and Is it Better?

Misleading Perceptions

In the promotional video, Gemini was portrayed as an AI capable of reacting to voice instructions, recognizing user artwork, and interacting with the surroundings.

At first glance, this material appears to be rather amazing. The video showcases an intriguing process in which Google Gemini recognizes user drawings. From initial sketches determining it’s a bird, identifying the species (duck), noticing the blue color on the duck, and concluding it’s a rarity among ducks. Furthermore, this AI has a sense of humor; for example, when a rubber, blue duck shows on the screen, Gemini responds with a funny “What the quack!” while explaining the material it’s made of. Placed on a world map, Gemini notes that a duck wouldn’t survive there.

Additionally, the AI demonstrates creativity by creating games, recognizing “rock-paper-scissors,” and performing magic tricks. It surprisingly excels in finding similarities between two connected or seemingly unrelated objects. Its impressive ability to detect the movement of paper under a plastic cup or predict a dot-to-dot drawing of a crab before completion showcases its capabilities.

In summary, Gemini is shown as an extremely intelligent system capable of distinguishing things. Some of them are correlating forms, predicting events, merging musical instrument pictures with their sounds, and much more. These advanced capabilities set it apart from other artificial intelligence models.

The truth, however, is quite different. The questions were written ant the replies weren’t created in real time, nor were they voice-activated. As Parmy Olson reported for Bloomberg, Google admitted that things didn’t go exactly as portrayed. Instead, researchers used static images input into the model and combined successful responses, partially showcasing the model’s capabilities.

“We created the demo by recording material to test Gemini’s abilities in various challenges”, a Google representative explained. “Then we prompted Gemini using static images from the recorded material and encouraged it through text”, he added.

Google recorded human hand activities, showed them to Gemini one by one as static pictures and spoke with the model via text (not voice). Then it picked the finest interactions to compile into the film using speech synthesis.

“We created the demo by recording material to test Gemini’s abilities in various challenges. Then we prompted Gemini using static images from the recorded material and encouraged it through text.”

Google representative

The video description emphasizes latency reduction and shortening of Gemini’s reactions and responses for greater conciseness, crucial for the purposes of this demonstration material. Oriol Vinyals, Vice President of Research and Deep Learning at Google DeepMind and a key figure in leading the Gemini project, noted that the demo video actually illustrates ‘…how multimodal user experiences created with Gemini could look like.’

Processing static pictures and text through huge language models currently needs enormous computer resources, making real-time video interpretation impracticable. For artificial intelligence scientists, this was a major signal that the video depicted the situation in an unworthy manner.

The edited processes and “cherry-picked” moments created an impression that Gemini could do much more than it was actually able.

Google Gemini: The Most Powerful AI Yet?

Gemini, according to Google, will be the most powerful AI ever created. It will be capable of language understanding, multimodal interactions, visual interpretation, code production, data management, and analytics. Google plans for Gemini to serve as a crucial component across the majority of its products and services.

Google highlighted multimodality as a fundamental aspect of Gemini from the start. This capacity to handle many sorts of information at the same time, such as text, graphics, and sound, constitutes a significant advancement over current AI models.

Source: https://www.techopedia.com/google-gemini-goes-live-heres-what-to-expect

Google intends to give developers access to Gemini, allowing them to build their own AI apps and APIs. Unlike past models with restricted access, this marks a departure by opening up to the development community.

The parameters of an AI model are an important aspect in determining its strength. While ChatGPT 4.0 contains 1.75 trillion parameters, Gemini has between 30 and 65 trillion parameters, according to claims. This parameter increase signifies potentially revolutionary power for Gemini.

Google Gemini in Practice

Despite marketing claims, the true potency of Gemini is yet unknown. According to independent reports, Google spent a large amount of money training its model with powerful hardware and vast volumes of data. However, the question of how much real value people would notice in regular use remains unanswered.

Conclusion

While Google Gemini claims to revolutionize AI, problems and differences between marketing promises and the model’s real capabilities raise concerns. The question arises regarding how much power Gemini has truly achieved.

Gemini’s picture recognition skills are noteworthy when seen in isolation and presented more precisely (as shown in this Google blog) . It seems comparable to the multimodal GPT-4V (GPT-4 with vision) artificial intelligence model from OpenAI. It can also recognize the content of static images. However, when completely integrated for promotional reasons, it gives the appearance that the Gemini model is more competent than it actually is. It is generating excitement among many people.

In short, a more accurate portrayal of Gemini’s specific skills, devoid of marketing gimmicks, would reveal that they are very outstanding. However, the advertising editing gave the impression of larger powers than it actually has, piqued the public’s attention.