- Unfolding AI
- Posts
- chatGPT Deeper Dive | Issue 12
chatGPT Deeper Dive | Issue 12
The Unfolding:ai weekly newsletter about AI for business professionals
Paul Bratcher
11 Oct
chatGPT Deeper Dive | Issue 12
Welcome to Issue 12 of unfolding ai, wow, has it really been 12 Issues. It seems like the summer has flown past. The coffee shops are in full on ‘pumpkin spice mode’, I haven’t yet given in to Mince pies… it’s just a matter of time.
 | Are you getting the most out of your subscription. If you visit the ‘burger’ menu you can login, see your referrals and read all the back-issues |
This week a focus on chatGPT. The chatGPTv (vision) model is still rolling out. I have seen plenty of ‘examples’, but have not yet tried it in a real business mindset. So before all of that arrives, we thought it would be good to just go over what is in chatGPT right now.
Best regards,
Paul, Co-founder (and newsletter editor)
Understanding the cut-off dates and what web browser integration actually does.
The basic model for chatGPT (its trained ‘corpus’ of knowledge), is september 2021 for version 3.5, for version 4 it is January 2022. The addition of ‘browse with bing’, or a ‘search plugin’ such as BrowserOP (my favourite) results in an extra query (sometimes) where relevant (according to bing search) additional data is pulled in.
This is in effect ‘pasted into’ your prompt, just the same as if you had searched in google and copy pasted the text into the prompt.
It does not have the same weight, value or integrations for inference as the core trained data, in essence its a ‘fine tune at a prompt level’.
It is useful, but not ‘game changing’.
 | chatGPT making a call out to ‘browserOp’ plugin, this data is then pasted into a prompt continuation to combine and generate the result to the question. You can ask chatGPT to explain itself |
> what caused browserOP plugin to be used ChatGPT
The BrowserOp plugin was used to fetch real-time information about the predicted weather for the next quarter of 2023. This was done to provide you with the most current and accurate data in response to your query about upcoming weather forecasts, how they compare to previous years, and whether these patterns could be indicative of climate change impacts. Would you like to explore further on this topic?
> how was it combined with the trained LLM data ChatGPT
The BrowserOp plugin fetched real-time information from various sources to answer your specific query about the predicted weather for the next quarter and its relation to climate change. Once the plugin returned this information, I integrated it with my pre-existing training data to provide a comprehensive response.
Here's a step-by-step breakdown:
User Query: You asked about the predicted weather for the next quarter, how it compares to previous years, and whether this could be an impact of climate change.
BrowserOp Activation: To provide the most current information, I used the BrowserOp plugin to fetch real-time data about the predicted weather for the next quarter of 2023.
Data Integration: The real-time data fetched was then analysed and integrated with my trained language model data, which includes general knowledge up to January 2022.
Response Formation: Finally, I formulated a response that combined the real-time data with contextual information from my training to answer your query comprehensively.
This approach allows for a more nuanced and up-to-date response than relying solely on my pre-trained data. Would you like to know more about any specific aspect?
Reply