Google's Gems Offer a Gentle Start to AI Prompt Engineering

While I'm no sales expert, the advice seemed valuable. Having a transcript of the chat saved in the sidebar is a nice feature for reviewing the session later.

However, there are some limitations. The Gem maintains a consistent tone but doesn't refer back to earlier points in the conversation, only moving forward. In a real coaching session, you'd expect the coach to connect later discussions with earlier ones. This seems to be a broader issue with large language models, likely requiring better use of the context window.

Additionally, the Gem relies on general sales knowledge from its training data. For more focused use cases, it could benefit from retrieval-augmented generation (RAG), which would allow it to tap into external databases for domain-specific knowledge.

Another potential improvement could be storing simple background knowledge as sentences, similar to OpenAI's "memory" function. This would let users incorporate recurring facts into their chats without repeating them each time.

The most significant limitation is that Gems don't retain past conversations. Each time you use a Gem, it starts fresh, with no memory of previous sessions. This is a major drawback if you want to use the Gem repeatedly, as you can't build on past discussions.

Despite these shortcomings, Gems are a great way to get started with prompt engineering, especially for those new to the concept. They offer a practical introduction to crafting effective prompts and using Gen AI tools.