Overcoming ChatGPT's Knowledge Gap
Large Language Models (LLMs) like ChatGPT have revolutionised how businesses access and process information.?
With more and more people using ChatGPT a critical limitation often goes unnoticed. ChatGPT, like all LLM’s have a? ‘knowledge cut-off date’. This knowledge cut off date varies between LLM’s. Depending upon the platform and the model version being used the date can be a year or more before present day.? This can lead to outdated or inaccurate information, particularly when referencing recent events or when relying on output produced in relation to rapidly changing industries.
The knowledge cut-off date refers to the point in time up to which the LLM was trained on data. For instance, if an LLM has a knowledge cut-off date of January 2023, it won't have any information about events or developments that occurred after that date unless explicitly provided with updated information.
ChatGPT and most (though not all) LLM’s attempt to get around this limitation by leveraging a search engine. For ChatGPT it is the “Browse with Bing” feature. This tool effectively connects the model to the internet thereby allowing it to access current information. The problem with that, however, is that the internet is filled with an array of content much of which is unreliable or inaccurate.?
To make best use of a feature like Browse with Bing it is critical that users employ strategic prompting techniques.
The key lies in crafting targeted queries that guide the model's internet search. Instead of general prompts, specify date ranges, authoritative sources, and the type of information you're seeking. This approach ensures relevance and accuracy across various business contexts, from market trends to regulatory changes.? It is also important to guide the model to rely on a minimum number of internet sources before it generates its output. Typically, I have found that a minimum of 8 to 10 is a good guide when prompting.
Consider this example for researching recent changes in employment law:
领英推荐
"Using Browse with Bing, research changes to Australian employment law regarding flexible work arrangements from December 2023 to September 2024. Focus on authoritative sources such as government websites, industry associations, and reputable business journals. Provide a summary of key changes and their potential impact on businesses. Take your time and do not generate your response until you have found at least 8 different URLs on which to rely"
This prompt demonstrates several best practices:
By structuring your prompts in this manner you're more likely to receive accurate, up-to-date information that reflects recent developments in your field.
It's also crucial to verify the sources ChatGPT uses. Always check the dropdown menu showing the searched sites to ensure the information comes from reliable sources. If needed, refine your prompt to exclude less authoritative websites.
By mastering these techniques business professionals across all sectors can bridge the gap between ChatGPT's base knowledge and the latest information available online.?
Understanding and overcoming the limitations of LLMs is crucial. By effectively using the Browse with Bing feature and employing strategic prompting, you can turn ChatGPT from a static knowledge base into a dynamic, up-to-date powerful assist.