As stated in the previous blog entry, other AI models have the tendency to hallucinate information.  A common example of this happens when you ask these models to provide sources for their information. This video shows exactly what happens when you do this. While these models are able to generate references, they direct to non-existent pages. As this article from TECHE shows, they can also occasionally come up with fake academic journal entries which never existed. In essence, other models don’t provide real references.

Why does this happen? It’s important to note that these AI models, specifically large language models or LLMs, only aim to generate text that is convincing to the human user. There are no inbuilt mechanisms to determine whether the output is truthful or not. Moreover, these large language models are trained on data scraped from the web with a cutoff date. For example, some models are trained on information from 2021 or before, but not after. This can also lead to gaps in the knowledge of these AI models. Basically, we can’t rely on other AI models to be factually accurate.

Poffices.AI Gives Real References

If other models hallucinate fake sources, then how is Poffices any different? The major difference is that Poffices.AI crawls the web in real-time for relevant articles and entries to gather information. This is how it is able to stay up to date with accurate information.

In conclusion, whereas other AI models have the issues of hallucinating fake sources and coming up with outdated information, Poffices is able to get by these problems using its own real-time web search functionality.