AI Chatbots

A recent snackable learning session (by Vincent Verbergt) discussed the current state of AI chatbots. It covered advanced AI models such as GPT and other technologies currently dominating the landscape. The session emphasized the importance of a down-to-earth, pragmatic approach when integrating AI within Lemon.

Battle of the Bots: GPT vs. Kapa.ai

One of the most striking parts of the presentation was the comparison between two AI systems: GPT and Kapa.ai. Both systems were given the same programming question, "How to define an endpoint?" Where GPT struggled with incorrect or hallucinatory answers, Kapa.ai managed to provide the correct information, including the source. This difference stems from the way both models work. Kapa.ai uses a technique called Retrieval Augmented Generation (RAG), where the model bases its answers on current and reliable sources, while GPT relies on a pre-trained dataset without up-to-date knowledge.

What is Retrieval Augmented Generation (RAG)?

RAG is an AI technique in which a model requests specific information from an external data source before generating a response. Instead of a large language model such as GPT relying solely on its pre-trained knowledge, the system searches for the most relevant pieces of information in, say, a database or documentation source. This allows for more accurate and context-specific answers, as was demonstrated during the session. This is an important development for companies looking to use AI for domain-specific applications, such as technical support or internal knowledge bases.

Why is this relevant to you?

Although AI models such as GPT can provide impressive performance, they are not always reliable. This is due to the fact that they often "hallucinate" and can generate erroneous information. A well-developed RAG solution can circumvent these problems by using specific sources, leading to more reliable and verifiable results. This means that at Lemon, we don't just jump on the AI hype, but rather carefully examine how to use this technology responsibly.

The future of AI within Lemon

The session also provided insight into how AI can be applied within Lemon in our daily workflow. Some promising use cases were mentioned, such as generating feature files based on photos or deploying AI for advanced searches within our knowledge bases. However, these are not simple implementations. The technology must first evolve further, especially in terms of reliability and predictability.

Lemon's practical approach

Clearly, Lemon is closely monitoring developments in AI. We continue to experiment and test AI in specific, controlled environments so that we are ready to integrate the technology as soon as it is mature enough. We see the enormous potential of AI, but want to make sure that its implementation in our workflows leads to valuable and reliable solutions for our customers.

‍