I created UnrestrictedIntelligence.com as a proof of concept. I wanted to be able to further experiment from his OODA network with some features of a large language model accessible via a connection to OpenAI.
Core functionality of the site was established during flight. By using code to write code (a “no-code” feature called Bubble) and learning how to use those methods to interact with OpenAI’s famous GPT models, we built the first version and spent a few hours I was able to go online with (Announced here in early January.) More features have been added since then. We can now demonstrate the capabilities of large-scale language models used in domains such as:
- intelligence analysis
- Competitor Information Evaluation
- cyber threat intelligence
- Board governance
- Academic Professor/Tutor
Over 10,000 questions have now been submitted via the site, each with an AI-generated answer.
This site is also a source of many lessons for me personally. Here are some of the lessons I learned:
Cambrian Explosion: A new feature Cambrian Explosion is coming. I’m not a developer, but I’ve leveraged my domain knowledge by using no-code tools and learning how to send and retrieve results from an AI provider (OpenAI via API) through a computer interface. , I was able to create a tool for my own use. and my community. There are no barriers to getting in here (if I can, anyone can). Millions of people can build microsites and personalized AI tools like this to add value to their niche.
Work crushing: Yes, many jobs and entire career paths change. The Unrestricted Intelligence site is just a proof of concept. However, it is clear that using AI-enabled tools improves analyst performance. If analysts performed better, would they need the same number for the same amount of work? Are there organizations that decide they don’t need analysts now? What about lawyers? What if we could reduce the number of lawyers needed because the productivity of lawyers who can use such tools is so high? All career paths should be examined. Remember that you are the only one responsible for your career. Whatever you do, think about how you use AI tools.
Rapid Engineering: This is a cute phrase that means asking a good question. Prompto When he first heard the word engineering, he fell in love with it. And I had heard early on from friends and business associates how people started making a living from rapid engineering. Sure enough, within weeks social media was ablaze with the term, and some people started selling prompt engineering lessons. It’s important to get the question right using the current large language model. But is it really engineering? Moreover, as time goes on, these systems get better and better. Perhaps the point here is that whether it’s a computer, an analyst person, or a team asking a question, you have to be good at asking good questions.
OpenAI. This company should probably change their name to reflect who they are now. Hats off to them for changing the world. But what is open about their approach? What is their model? how do they work? how are they trained? How are errors detected? How do the leadership team’s views and biases affect results? What is their roadmap? What do you do with data submitted as queries? How is user data protected? Almost all of these answers seem closed, but this company should probably be called ClosedAI.
Company data: Perhaps the biggest advantage of OpenAI and tools like my web app is that they help people think about what such tools can do with their data. Currently OpenAI is very bad at it. They have a new plugin architecture that allows it even more, and solutions leveraging Azure and OpenAI APIs offer some limited functionality here. But it’s all suboptimal and far from what an enterprise that cares about data security needs (see What leaders need to know about natural language processing). It’s time for the company leader to articulate her AI requirements for data (we hope the Unrestricted Intelligence web app helps users think about what those requirements are). For example, companies should start articulating the need for neural search for all internal data retention. AI-enabled summaries should also be a corporate requirement. You also need the ability to manipulate all the data you hold.
Today’s AI is NKR, which is incorrect: The big advantage of a well-trained language model is that it has great potential. Mistakes, errors, and outright hallucinations have the potential to go the other way. Recognizing the types of flaws in current systems can help ensure decision makers incorporate ways to mitigate risk. The best intelligence analysts already know that they work to reduce their biases and protect themselves from cognitive traps. Great leaders know to believe everything they think. Now we have to understand that even new computer tools are biased and can mislead us.
AI helps build AI tools. As I said, I’m not a developer. I’m also not a power user of no-code tools. I was able to get the Unrestricted Intelligence site up and running, but if I wanted to add more advanced features, it would have taken more technical skills than I could acquire. I could ask for a well-formed query that would help add AI has helped us build better AI tools. This has helped me a lot in my other coding projects as well.
Hackers are trying to hack: I love hacking spirit. I have always admired great and persistent actors pushing their systems to their limits and exploring the realms of what is possible with technology. Yes, of course, I mean hackers operating with permission. The Unrestricted Intelligence site was designed as a proof of concept, where people were encouraged and allowed to ask all sorts of questions. Some people decided to use it in ways that surprised me. One creative thinker kept sending increasingly complex queries designed to generate responses that revealed something that should have been hidden from the user. This hidden part is different for each section of the site. This is a long pre-prompt sent to her OpenAI before the actual user prompt. I could read the log because this creative genius finally did. I certainly learned from it.
Asking good people for feedback can help you: I got a lot of feedback on this from friends and colleagues. This includes friends pointing out bugs. This is always true for all applications, but in the age of AI, I find it even more important to have friends and colleagues point out problems and provide feedback. We humans really have to unite.
AI can be used to keep content a little more civil. Some people on the internet get a kick out of being mean. I don’t know why they would use my site to ask nasty questions, but some did. Maybe it was to test if it’s different from ChatGPT’s reply. I wanted to reduce this sort of thing on my site and changed the background prompt to list the worst bad words and reply “this is not the right question” if someone uses them I thought of doing But it’s getting too long and it’s easy to get around. Instead, I changed the code to a simple “if the request is lascivious, stop processing it and reply that it’s not on the right topic”. It worked really well. There was one person who surprised me by the way he asked a very twisted question and got a very twisted response, but it’s so twisted that I can’t discuss it further. , the idea of ​​using AI to keep content more civil seems to have worked.
Turing test: This test proposed by Alan Turing was originally called the imitation game. It was devised as a test to see if responses from computers were indistinguishable from human responses. Turing proposed that human evaluators judge natural language conversations between humans and machines designed to generate human-like responses. For 70 years, this concept has helped humans think of a computer as his concept of intelligence. But now it’s almost been shown that computers can pass this test, but they’re still not intelligent. It is necessary to consider new countermeasures. Just as all AI is not the same, we also need to understand that not all humans are the same.
These and many other lessons will help in the continued research of artificial intelligence. Please continue to provide insight. Also, please continue to test UnrestrictedIntelligence.com.
