Read our blogs, tips and tutorials
Try our exercises or test your skills
Watch our tutorial videos or shorts
Take a self-paced course
Read our recent newsletters
License our courseware
Book expert consultancy
Buy our publications
Get help in using our site
537 attributed reviews in the last 3 years
Refreshingly small course sizes
Outstandingly good courseware
Whizzy online classrooms
Wise Owl trainers only (no freelancers)
Almost no cancellations
We have genuine integrity
We invoice after training
Review 30+ years of Wise Owl
View our top 100 clients
Search our website
We also send out useful tips in a monthly email newsletter ...
ChatGPT embedded within an Excel workbook? What's not to like? |
---|
How one determined person embedded the published ChatGPT 2 source code and weightings within an Excel workbook to give an insight into how AI works |
In this blog
When OpenAI released the underlying source code for GPT-2 in 2019, Iman Amand decided to create a version of Excel which included ChatGPT 2 embedded within it. You can download this from GitHub (all 1.25 gb of it!).
Almost halfway ...
The file has an .xlsb extension, which means that it holds information in binary rather than readable text format. This allows Excel to read and write large volumes of data more quickly (having said which, this isn't going to run quickly and may well seize up your version of Excel).
What you see when you open the workbook.
Making sure that I had manual calculation enabled, I typed in these words and then recalculated the spreadsheet:
The model has determined that the most likely next word is you - not a bad prediction.
There are 172 worksheets in the workbook, a small number of which are shown below:
A few of the worksheets.
A quick scan through some of the worksheets shows how deceptively mathematical AI tools really are:
A sample sheet - this one contains 2304 columns and 768 rows of (I presume) weightings, making 1,769,472 cells in total. It's not good bedtime reading matter.
When you type in some keywords, the model generates the next word from the Predicted_Token sheet:
The formula refers to row 2 column 2 on the Predicted_Token sheet.
This sheet contains this formula:
The id_to_tokens worksheet contains a list of 50,258 words, each of which is assigned a number.
And that's about as far as I got! However, this spreadsheet will allow you to reverse engineer how the ChatGPT algorithm works (if you have more patience and determination than this author).
You can see more background on the creation of this tool (including a video from its creator) at this ArsTechnica site.
Some other pages relevant to the above blog include:
Kingsmoor House
Railway Street
GLOSSOP
SK13 2AA
Landmark Offices
99 Bishopsgate
LONDON
EC2M 3XD
Holiday Inn
25 Aytoun Street
MANCHESTER
M1 3AE
© Wise Owl Business Solutions Ltd 2024. All Rights Reserved.