\n However, there is still a challenge getting the model to answer with up-to-date \u201cgeneral information\u201d (for example, if the question relates to events that have occurred after the model\u2019s training). You can see a \u201creal life\u201d example of this when you use Bing Chat<\/em> versus ChatGPT<\/em> to search for a new TV show called \u201cPoker Face\u201d which first appeared in 2023:\n<\/p>\n \n \n You can see from the Bing Chat screenshot in Figure 3 that it takes an intermediate step before answering the question, which is to extract the intent of my question and perform an internet search. The Bing Chat answer specifically references Wikipedia.org as its primary source (as shown by the footnote indicator on the results) that helped it to answer this question.\n<\/p>\n \n To illustrate how to build an internet search that will let JetchatAI<\/em><\/strong> answer questions about recent topics, we\u2019ll create a new function (similar to the weather function<\/a>) that searches Wikipedia<\/a> and uses the results to augment our OpenAI chat requests.\n<\/p>\n NOTE: this post is not<\/em> intended to be a production-quality implementation of a Bing Chat-like feature. Rather, it walks through the principles<\/em> whereby the chat function API<\/a> can extract meaning from the user\u2019s query and execute a search on their behalf, and then summarize the results and thereby answer questions on topics that did not exist in the training data.<\/p><\/blockquote>\n \n Figure 4 shows how JetchatAI would answer this query before<\/em> adding the ability to search more current data. The app is configured with the \n \n I\u2019m reasonably sure the model is hallucinating the answer because the ChatGPT website did not respond with an answer, and every time I restart JetchatAI, it produces a different answer. This is probably due to configuration differences (e.g., temperature) between the model being used in JetchatAI and the ChatGPT website examples in Figures 1 and 2.\n<\/p>\n \n In order to access more accurate, up-to-date information, we\u2019ll need to implement a way for the app to query an internet search index. For this demo, we\u2019ll just use Wikipedia\u2019s MediaWiki API<\/a> \u2013 it\u2019s not as comprehensive as an internet search engine like Bing, but it will provide enough data to build a working prototype.\n<\/p>\n \n The existing Figure 5: HTTP requests to Wikipedia to implement the \u2018search\u2019<\/em>\n<\/p>\n \n With these two HTTP requests we can now create a chat function \n There are four different spots in the function declaration that act like a \u201cprompt\u201d that affects how the model interacts with the function. I tested two different styles of \u201cprompt\u201d, one that was very broad and another that was designed to limit answers to TV\/Movie topics only. These values are all set in the AskWikipedia.kt<\/strong><\/a> file.\n<\/p>\n![\"\"](\"https:\/\/devblogs.microsoft.com\/surface-duo\/wp-content\/uploads\/sites\/53\/2023\/10\/a-screenshot-of-a-computer-description-automatica-1.png\")
![\"\"](\"https:\/\/devblogs.microsoft.com\/surface-duo\/wp-content\/uploads\/sites\/53\/2023\/10\/a-screenshot-of-a-phone-description-automatically.png\")
![\"\"](\"https:\/\/devblogs.microsoft.com\/surface-duo\/wp-content\/uploads\/sites\/53\/2023\/10\/a-screenshot-of-a-television-show-description-aut.png\")
gpt-3.5-turbo-16k<\/code> model, which returns details of an old TV show…\n<\/p>\n![\"\"](\"https:\/\/devblogs.microsoft.com\/surface-duo\/wp-content\/uploads\/sites\/53\/2023\/10\/a-screenshot-of-a-chat-description-automatically-1.png\")
HttpClient as a chat function<\/h2>\n
currentWeather<\/code> function performs two HTTP requests (to locate the user and then to retrieve the weather for that location). The pull-request<\/a> for this blogpost shows that we can copy-paste that exact code to perform a \u2018search\u2019 on Wikipedia and then retrieve the content for the first matching result:\n<\/p>\n val wikiSearchUrl = \"https:\/\/en.wikipedia.org\/w\/api.php?action=opensearch&search=$query&limit=1&namespace=0&format=json\"\r\n val wikiSearchResponse = httpClient.get(wikiSearchUrl) {contentType(ContentType.Application.Json)}\r\n if (wikiSearchResponse.status == HttpStatusCode.OK) {\r\n \/\/ CODE OMITTED to parse the Wikipedia search response\r\n wikipediaTitle = responseJsonElement[1].jsonArray[0].jsonPrimitive.content\r\n }\r\n val wikiTitleUrl = \"https:\/\/en.wikipedia.org\/w\/api.php?action=query&prop=extracts&exsentences=10&exlimit=1&titles=$wikipediaTitle&explaintext=1&format=json\"\r\n val wikiTitleResponse = httpClient.get(wikiTitleUrl) {contentType(ContentType.Application.Json)}\r\n if (wikiTitleResponse.status == HttpStatusCode.OK) {\r\n \/\/ CODE OMITTED to parse the Wikipedia content response\r\n wikipediaText = itemPageElement?.get(\"extract\")?.jsonPrimitive?.content\r\n }<\/pre>\naskWikipedia<\/code>, which will attempt to use the Wikipedia search engine to identify a top result, and then visit that entry and use the page summary as context for our chat request. The query is hardcoded to only return a single result \u2013 in a production application, we might choose to return a number of top results (possibly using embeddings to evaluate how likely they are to be useful)\u2026\n<\/p>\nPrompt engineering challenges<\/h2>\n
![\"\"](\"https:\/\/devblogs.microsoft.com\/surface-duo\/wp-content\/uploads\/sites\/53\/2023\/10\/a-screenshot-of-a-chat-description-automatically-2.png\")
![\"\"](\"https:\/\/devblogs.microsoft.com\/surface-duo\/wp-content\/uploads\/sites\/53\/2023\/10\/a-screenshot-of-a-phone-description-automatically-1.png\")
<\/a>