Here’s Our Experience with Google’s New AI Tools

When it comes to weaknesses, the new AI tools share similar drawbacks that other LLM AI programs have. Unfortunately, accuracy cannot be guaranteed, so it is essential to verify any information provided by the AI bot before relying on it.

Moreover, the quality of the AI’s writing leaves something to be desired. As seen in the above email draft, the bot begins every sentence with the letter “I,” which is considered poor writing form. This issue persisted in many of the bot’s drafted emails, necessitating time and creativity to rectify. In addition, the bot occasionally fabricates details that were not mentioned in the original prompt, potentially causing trouble for the sender if the email is not edited first.

For example, in one case, when drafting an email to request an overdraft extension from a bank, the bot assumed we had been loyal customers for over 10 years with an excellent credit history. However, removing this information resulted in the core argument of the email being lost.

Furthermore, the available options for these tools are limited. Similar to other tools, the success of Duet’s responses heavily relies on the quality of the prompt provided. The more comprehensive the prompt, the more accurate Duet’s response will be. However, this expectation becomes burdensome as it contradicts the purpose of the tool, which is meant to save time. Additionally, crafting an effective prompt is often considered a highly valued skill, further complicating the usability of these tools. For Google’s model to truly be useful, it should be able to generate valuable output from minimal input.