My friends often ask for use cases for GPT. Below I provide one recent example.
I attend a weekly educational seminar where real estate professionals present the work they do. I take notes on paper, and later transcribe them to my computer, which is a pain. As a result, I have a backlog of many months worth of meetings. I don’t look forward to it.
I considered paying a virtual assistant in another country to do the work for me, but the contents of these notes didn’t justify the cost.
I considered speech recognition software (not cloud based), but the accuracy is too low for my satisfaction. It’s not fun babysitting a tool, and cleaning up all the errors is a lot of work.
An idea occurred to me: Why not use speech recognition software to write the initial, horrible draft, and then send it to GPT for cleaning?
I used Nerd Dictation, which is built on top of VOSK. It’s pretty bare bones. All it does is enter the word you speak in lower case. No punctuation. No autocapitalization.
Below is an example of the output generated by the speech recognition software: [1]
medlock value fund november third twenty twenty three new line the
fund focuses on discounted and distressed real estate investments period it aims for thirteen to seventeen percent
they focus on five to fifty million dollar properties period little competition in the space period most areor under five million dollars or over fifty million dollars period
when they buy a comma they have twoo do three ways to improve cash flow period they aim for a fixed interest loans period
look at over six hundred deals a year and invest in less than ten per year period they aim for low leverage period alone would be a be percent of the value of the property period
first fund launched in twenty nineteen open parenthesis but they’ve been doing real estate for years close parenthesis period find clothes in twenty twenty one period they bought twenty six properties period and they’ve distributed it one point seven the
total property value of that file was four hundred and seventy five million dollars new line forty four percent of it was multifamily new line thirty percent seven per cent was commercial new line the
in twenty twenty two they launched their second fund new line sixteen properties acquired new line forty four per cent were multifamily new line thirty six percent were industrial new line twenty percent were commercial new law as part of their filed period.
I gave the following prompt to GPT:
You are going to correct for text that has been produced by voice recognition software. Rewrite any text provided to you. The text will not have punctuation, so please add where needed. If the text has the word period, then insert a period, and do not insert another one just before or after. If the text has “new line” or “newline”, insert a newline character and start a new paragraph. If the text says “comma”, insert a comma. If the text has a number spelled out, replace it with the actual number. So “five thousand” becomes “5000”. Also, if something seems off in the text, it was probably due to a misrecognized word. Please correct for it.
Here is the output:
Medlock Value Fund, November 3, 2023.
The fund focuses on discounted and distressed real estate investments. It aims for 13 to 17 percent returns. They focus on $5 to $50 million properties with little competition in the space; most are under $5 million or over $50 million.
When they buy, they have three ways to improve cash flow. They aim for fixed interest loans.
They look at over 600 deals a year and invest in less than 10 per year. They aim for low leverage; a loan would be 80 percent of the value of the property.
The first fund launched in 2019 (but they’ve been doing real estate for years). Fund closed in 2021. They bought 26 properties and distributed $1.7 million. The total property value of that portfolio was $475 million.
44 percent of it was multifamily. 37 percent was commercial.
In 2022, they launched their second fund. 16 properties acquired. 44 percent were multifamily. 36 percent were industrial. 20 percent were commercial as part of their portfolio.
Note that it added the word ‘returns’ after the 13-17%. It also corrected a number of spelling mistakes. It occasionally loses some information. As an example, the original text said “twoo do three ways” and it simply dropped “twoo do”. It should have said “2-3 ways”.
Nerd Dictation does allow me to specify a Python script which processes the output. I could handle things like “period” myself. The beauty of using GPT is that I can achieve this without any programming whatsoever. And occasionally, it improves my writing. Within minutes, it allowed me to use a fairly poor voice recognition tool and produce real usable output.
Granted, I still have to go over the text and correct a few of the remaining mistakes, but it’s probably only 10% of the effort it would have been without using GPT.
I’ll probably write an elisp function to send the output to GPT and insert the results into the buffer. Or make my Python script do that. I’m still exploring.
| [1] | Yes, this may seem like exceptionally poor output. This is my first time using the software and I haven’t figured out how to enable punctuation and have it write out the numbers. The question is: Should I even bother finding out? GPT is solving this for me. |