Creating a Better StackOverflow With ChatGPT
We became a little bit 'angry' by OpenAI turning off internet search in ChatGPT a week ago, so we decided to fix the problem.
Join the DZone community and get the full member experience.Join For Free
By now, you've probably heard the news about OpenAI turning off internet searches on ChatGPT. At AINIRO, we believe that software evolution by itself is a higher purpose, so we created AINIRO Oracle, the Search Engine that uses ChatGPT to answer questions.
To understand our motivation, imagine the following use case where you want to find the answer to the following question:
Who was the 35th President of the US' wife?
The process of finding an answer to the above question with a traditional search engine tends to be ridiculously complex, with a lot of unnecessary manual work.
- Find out who was the 35th President. This typically includes finding all Presidents and then wading through this list until you're at number 35.
- Find the Wikipedia article for the 35th President, and then scan it for his wife.
The above question is a simple one, but more complex questions can easily take up to 10 to 20 minutes of your time. With our Oracle engine, the process takes 30 seconds and is 100% automatic. Here's the answer to the question: Who Was The 35th President Of The US' Wife? Below is another example article it automatically produced given the specified question.
We've only played with it for a couple of days, obviously, since we released it yesterday. But in general, I'd say this "feels" one million times better than Wikipedia, StackOverflow, and Google. No need to scroll through endless data of irrelevant content to find the answer to your question, and no need to click 5 links in Google before a relevant article pops up. If you're a software developer, you can probably relate to the following after having played with it for a while ...
It's the AI equivalent of StackOverflow.
How the Oracle Machine Works
First, it takes your query and sends it to DuckDuckGo. Then it scrapes the top 5 results in parallel and creates a context. This context is sent to ChatGPT with instructions to answer the original question. Due to a kick-ass web scraper, combined with some intelligent prompt engineering, we're able to make sure it always puts the answer to your question at the top of the generated article while adding additional relevant content further down.
The result becomes a search engine, or an Oracle answering machine, that "cuts through the crap" and answers your questions with the most significant parts of its answer at the top — significantly simplifying the task of having your questions answered.
One crucial difference between our Oracle engine and ChatGPT is that our Oracle engine arguably has access to "a bajillion times as much data" in addition to, of course, real-time data. The last point is kind of important if you're searching for information related to contemporary information, such as, for instance, "Who Won US Open in 2023?" If you ask ChatGPT who won US Open, it will have no idea. If you ask our Oracle engine who won US Open in 2023, it will provide a perfect answer for you, often adding images that are relevant to your answer.
The Better StackOverflow
Claiming to have invented a better StackOveflow in two days is a pretty strong statement. However, I will prove it by doing a site search on Google for three keywords towards StackOverflow, for then to reproduce the same queries towards our Oracle Machine. All my searches at StackOverflow are done with a "site search," and I chose the first article Google could find.
- StackOverflow — What is OOP?
- StackOverflow — What is dependency injection?
- StackOverflow — What are the original 23 design patterns from GoF?
- Oracle — What is OOP?
- Oracle — What is dependency injection?
- Oracle — What Are The Original 23 Design Patterns From GoF?
Besides that, it didn't have a large enough context to list all 23 design patterns. I'd say at least two of the above three questions are far better answered by our Oracle Machine.
Social Media Argument Machine
In addition to the above obvious use cases for finding answers to questions extremely fast and extremely accurately, it also seconds as "a social media answer machine," implying if somebody asks you about anything in some social media context, you can write it out as a question to our Oracle Machine, and provide a highly accurate answer to the question in some few seconds.
The Oracle Machine will do its best to try to preserve its sources and also allow you to reproduce its DuckDuckGo query. In addition, it creates a permanent high-quality article you can share and send to others by email, use as Facebook comments, Twitter replies, etc. The Oracle Machine will also cache its answers such that if another user asks a question that has more than 95% similarity with a question it has previously answered, it will return this answer instead of generating a new answer. However, this is only true for five days since otherwise, it would become stale over time, incapable of answering questions about recent events.
All of these traits make it (I hope) a highly valuable addition to your internet tooling, delivering a somewhat unique value proposition and simplifying your life in general.
Published at DZone with permission of Thomas Hansen. See the original article here.
Opinions expressed by DZone contributors are their own.