Over the past year, I’ve built a few different custom GPTs using the ChatGPT platform from OpenAI. My goal with each Generative AI tool is to help you understand the intricacies of US government regulations. This is especially needed by smaller international development firms and local organizations without large compliance teams.
I firmly believe that GenAI tools performing Retrieval-Augmented Generation (RAG) using factual external resources can generate accurate human-readable responses from dense legal texts. Even when my chatbots fail.
My Custom GPTs on OpenAI
Here are the GenAI tools I’ve created using the ChatGPT platform and available on the GPT Store:
- Write Like USAID Bot – creates text based on the USAID Style Guide and Messaging Manual
- ADS Answer Bot – understands the USAID Automated Directive Systems that details USAID policies
- AIDAR Answer Bot – interprets the USAID Acquisition Regulations that expand on the FAR
- USAID.gov Answer Bot – looks at the USAID website and helps you find publicly available resources
- FAR Answer Bot – explores the Federal Acquisition Regulations that are 4x the size of the Bible.
- FAM Answer Bot – parses the Foreign Affairs Manual and associated Handbooks that USAID and State Department staff should follow in their work
- ICTworks Answer Bot – mines the +2,500 posts on ICTworks for insights on digital development
Please be sure to share your feedback on using these chatbots with me. I am still learning how to optimize these tools, like everyone else on the LLM journey.
Custom GPTs are Now Free
OpenAI has just announced that the custom GPTs that I created on the OpenAI Store are now free to all ChatGPT users. You no longer need to create a paid account to enjoy these GPTs. You can also use OpenAI’s newest flagship model, GPT-4o.
Of course, with my custom GPTs, and any GenAI tool, please be sure to double check the results to ensure accuracy. We are all at the mercy of OpenAI with ChatGPT, which is trained on the full glory of the Internet – the good, bad, and ugly.
Also, please ask your EXO, RLO, or appropriate regulation expert for their input before making any decisions based on these chatbots. GenAI tools should be the start, not the end, of your understanding of USG regulations. There will always be nuance and context that no computer will know.
I am humbled to have to understand this Ai tools . I am yet to learn and put them to productive use. Kindly guide me on the process of learning and grants
It is great that OpenAI finally made these custom GPTs free to non-subscribers. I just tested out the ADS bot, and I must admit that it struggled with what I thought was a pretty softball question. I asked it what expenses under $75 require a receipt and it missed completely that receipts are required for taxis in the US regardless of cost. It also kept talking about how telephone calls over $75 require a receipt, which is the opposite of what I was asking. I know this is not your fault, as you trained it on just the ADS, but it shows the limitations of ChatGPT’s comprehension of some types of text.
We are all at the mercy of OpenAI when using chatbots developed on ChatGPT. This is why I built a custom LLM for my previous employer – for greater control and accuracy. It was more accurate than Google Search for the Federal Acquisition Regulation, faster than the State Department’s own search engine for the Foreign Affairs Manual, and more comprehensive than either on IRS’ Internal Revenue Manuals.
This is such a great contribution to the development community. Thank you for your hard work. I’m looking forward to using them!
Thanks Wayan! I’m eager to check out some of the others you’ve built, but checked back in on the ADS bot. Contrary to Josh’s experience, I have exciting news! In Nov 2023, the ADS bot really botched answers with respect to ADS205 (gender), yet did so with great confidence as always. This is dangerous since others less familiar with gender-related compliance might not be able to discern the inaccuracies. Today, I just popped in and asked the same questions and the improvement is amazing–the answers were factually accurate! Which leads me to ask, did you do anything to improve it? If yes, what did you do? If not, how can we explain this? It’s my understanding that custom GPTs don’t get upgraded to 4o… Would love to hear more on what tinkering you did! Thanks again!
We are all at the mercy of OpenAI when using chatbots developed on ChatGPT. OpenAI says that GPTs now use the newest flagship model, GPT-4o. That probably explains its better understanding of gender in the ADS. OpenAI cautioned that behavioral changes with the new model should have me review and test these GPTs to ensure they continue to perform as intended. I did a few tests but didn’t find much change. Glad you found change for the better.