Regulatory Augmented Itelligence 5.0
Twenty-two months ago we launched ChatMDR. We had no idea that we would we up to our fifth major release now, across a variety of regulations: ChatMDR, ChatIVDR, ChatFDA and ChatAIAct. The number 5 handily coincides with GPT-5, the most capable AI model in the world to date, but actually, that was a coincidence. There is much more in version 5 that we need to tell you about. Grab a coffee and sit back, because here goes:
Read-along updates
The RAI tools take their sweet time. That is a feature, not a bug: to make sure the answers ChatMDR and friends provide are good and of regulatory quality, there is a lot of information being processed and thinking (so to speak) going on. That doesn’t mean it’s nice to wait, though. Therefore we are happy to announce one of the most anticipated features of the chatbot: incremental info updates that you can read along with.
The app will begin with reasoning for harder requests and you can read along with its thoughts. This can be helpful sometimes if the model draws a wrong conclusion; you can then make out where it went wrong and correct it in the next conversation turn.
After the reasoning, the final output is produced. The output stays in your chat history, printouts and so on, whereas the reasoning summary does not. By default, the screen scrolls along with the text being produced, but you can break that behaviour by scrolling up with your mouse or trackpad. That enables you to read what was produced while the chatbot is still going. You can go back to the live streaming by scrolling all the way down.
Ding means done
If your workflow involves going to another browser tab or even another application while AI is coming up with (hopefully) some brilliant answers, then this feature is for you. Rather than having to check back in with ChatIVDR and others, you can now have a little bell sound to let you know your answer is ready. Inspired by 60s-diner-the-food-is-ready bells and internally known as the Leo Feature. Leo, here you go! 🔔
Please note: on some browsers, the bell has to be toggled once off and then back on to activate the sound for the first time.
Thinking effort
We’ve been antromorphising ‘thinking’ quite a bit now. The English language (nor any other that we know) doesn’t have a better word for the type of processing that the AI model does after we ask it a question. So thinking it is.
New in Release Five is a slider that lets to choose the amount of thinking the model needs to do. This replaces the old ‘model selector’ where you would choose between various OpenAI models that had different capabilities. That complexity is now gone: all questions are handled by a version of GPT-5, the most capable AI model in the world.
The Thinking effort slider bar has four settings:
- Auto: estimates the amount of reasoning required based on the question you ask. This is what we use ourselves and the setting most people will use. Like your car nowadays, just put it in Drive and let the technology do the work.
- Low: tell the model that this is a simple question and don’t go overboard with time-consuming thinking. Give it to me, fast and simple! Great for information lookup questions.
- Medium: another decent default setting if you don’t like Auto. The model will spend an intermediate amount of its considerable brain power and time. This is the setting that Auto usually ends up choosing as well.
- High: for the really hard questions, obviously. This engages the highest gear of GPT-5, the one that wins math olympiads and such. The answers are very well thought through, but there is a penalty. Running a query like this can sometimes take two, three, four minutes (!) for an answer. Probably not something you want to use for every question, but the power is there when you need it.
High thinking and the Free subscription
The Free plan continues to be a popular plan. Try out the solution, be able to work with it already before convincing the boss (let us know if we can help with that, by the way) or wait until it becomes worth your 65 cents per day to have the smartest AI sidekick for regulatory affairs by your side all the time.
Previous versions of ChatMDR, ChatIVDR, ChatAIAct and ChatFDA did not give Free users access to the most powerful model, with poetic names like Diamond or Scholar. The sad part about that was that Free tier customers could never experience the power these models offered without upgrading. There was no kicking the tires, try-before-you-buy possibility.
Now there is. All customers have access to High Thinking Effort. There is one caveat: on the Free tier, you get to submit one question on High per day. Premium customers on Plus or Unlimited can ask as many High questions as they want.
Prompt library
Think of the Prompt Library as a set of templates for questions you may ask the chatbot. For each of ChatMDR, ChatIVDR, ChatAIAct and ChatFDA there are some dedicated prompts, with blanks to fill for your own devices. They are intended to not only help you ask certain questions with less typing, but also to show you what a strong prompt looks like. Watch and learn!
The Prompt Library has its own post here, where you can read more about it.