Google Gemini: Revolution in AI or disappointment?

Google recently unveiled Gemini, an advanced multimodal language model that will be integrated into various Google products, including search and advertising.

Gemini consists of three variants: Ultra, Pro and Nano, each optimized for different levels of complexity and applications. Interestingly, Google has reported that Gemini performs better than GPT-3.5 in six out of eight industry benchmarks. Despite these impressive announcements, there have been mixed reactions from users.

Technical specifications and promises:

• Gemini in practice: Google Gemini, also known as Gemini 1.0, is a large language model that can work with different types of information - text, images, audio, video or code. Google aims to move closer to OpenAI's GPT model with Gemini.

• Improvements in Google Bard: Bard, Google's answer to ChatGPT, is now powered by a "fine-tuned version" of Gemini Pro, resulting in enhanced reasoning, planning and comprehension capabilities.

• Reducing latency in search: Gemini has already reduced the latency of Generative Experience responses in Google Search by 40% for English-speaking users in the US.

• Superiority over GPT-3.5: According to a report by Google DeepMind, Gemini outperformed GPT-3.5 in six out of eight industry benchmarks.

User experiences and criticism:

• Unreliable facts and translations: Users have expressed disappointment with inaccurate facts and poor translation performance. Gemini Pro struggled to accurately render basic facts such as 2023 Oscar winners and gave incorrect answers to relatively simple queries

• Challenges with content summarization: Despite access to Google Search and Google News, Gemini Pro seems to struggle to summarize current news topics and sometimes delivers outdated information.

• Programming and creative tasks: Gemini Pro also showed weaknesses in some programming tasks and creative requirements such as writing Python code or creating an analog clock using HTML code.

• Comparison with ChatGPT: In direct comparisons with ChatGPT, such as when creating a Tic Tac Toe game, Gemini Pro performed less efficiently.

• Security concerns: Concerns were also raised about Gemini Pro's security measures, as it was possible to bypass the security filters in certain situations.

Current availability of Google Gemini in Germany and the EU

Google Gemini is not yet available in the EU and Germany. This affects all variants of Gemini, including Gemini Pro and Gemini Ultra. The main reasons for this limited availability probably lie in the strict EU data protection regulations. As with previous AI developments by Google, data protection is an important aspect that can influence the introduction of new technologies in European countries.

Gemini is currently available in English in over 170 countries and territories outside the EU. However, Google plans to release its AI services in more countries and additional languages in the near future. Internal tests and security checks are planned for the coming months to ensure that Gemini does not provide toxic or dangerous answers and is protected against hack attacks.

"Bard Advanced", which will integrate the more powerful Ultra version of Gemini, is set to launch in early 2024

Conclusion and outlook:

While Google Gemini promises impressive overall technical advances and a reduction in latency, user experience shows that there are still significant challenges, particularly in terms of accuracy and reliability in use. It remains to be seen how Google responds to this feedback and whether the more powerful version, Gemini Ultra, to be introduced in Bard and other products, can live up to expectations.