Applied Artificial Intelligence - S1 Issue 2
APAI (Applied AI) - learn how to implement AI solutions
“The terms have changed, long live the terms” - OpenAI have announced on Dec 17th new terms of use for their services to take effect on Jan 31, 2024. OpenAI was under pressure in spring 2023 to not use your data for training. So for a while they said they would not. Then mid 2023 they indicated there will be a business usage tier and one of the things included in that tier will be privacy. It was clear that OpenAI needs user data to keep improving their service, but also the only way to get enterprises on board was to offer privacy at the business tier. The old adage holds: “if you are not paying for it, you are not the customer”
The OpenAI email announcing the upcoming change.
You can have privacy if you are paying for it. If you continue to use ChatGPT as a regular user account, your data (input/output content) will be used to train their system to make it better, and sell it. If you want to avoid having your data used to train their systems, you must have a business account or use the API.
So, what’s the takeaway from this?
No, don’t run away and delete your OpenAI account. But do use the API to access OpenAI services. There’s a large number of free apps, wrappers, tools that let you use their OpenAI API key or you can input your own.
Do you have a preferred tool to access OpenAI’s GPT? Do you need help finding some good starting ones? Let me know in comments…
This Issue’s Table of Contents
Weekly Lessons - Lesson 3 - Use the API, or a business OpenAI account
Dog Food Module - Week 2 of Creating Voice Audio DigitalStereo Startup
Weekly Lessons
Lesson 3. Use a business account to access ChatGPT
Use a business OpenAI account, the API or an application to access OpenAI Services. Do not use a personal ChatGPT account.
Using ChatGPT directly has privacy issues and OpenAI has set expectations that they will use your data for training. You should use an app to interact with OpenAI’s GPT.x No need to pay for the use of the app - ChatGPT clones, lookalikes and wrappers are mushrooming like crazy. Pick a reputable one, try using Microsoft Bing, or Opera’s built in chat agent.
Imagine you are running for President 10+ years from now. Now imagine you are using AI to summarize all your server’s emails but you are not worried because all the OpenAI communication is happening over APIs. However, you never counted on your overworked and not-so-well-trained office assistants who use ChatGPT directly using their various free accounts. And such it happens, that somebody posts on social media a prompt that allows them to force ChatGPT to disclose training data pertaining to you, your office and your past. Or, if by then OpenAI makes it harder to leak training data, a judge can subpoena OpenAI training data, which you do not control.
The Dog Food Module - Part 2 of Launching DigitalStereo Labs
This is part 2 in my “working in public” coverage of taking DigitalStereo Labs from idea to launch. DigitalStereo is an app that lets you generate highdef audio from text.
In part 1 I describe the overall approach over the next 4 weeks.
In this part 2, I will describe the use case and what are next steps.
Imagine you have a product - such as a temperature control device like Nest that controls your smart home. Now, imagine you need to create support videos on how to use your product. You create your video slides, and now you create the transcript for the audio voiceover. You upload the transcript to DigitalStereo Labs, generate the audio, and you have professional grade audio in a manner of minutes. You can then upload the video to your Youtube support channel - If you need to make any changes to the audio, it is as simple as editing the actual text file and download a new TTS audio.
No need to hire voice actors, no need to re-hire voice actors if you need to make last minute changes.
This will as well allow you to build a fully indexed, searchable via AI assistants vectorstore database of all your transcripts. Therefore your staff may search for templates, approved boilerplates, you can alter/update transcripts at any time. As well, your support staff may search for the appropriate video via transcript keywords that are fully indexed.
Most people think Text-to-speech and think that either you will speak to interact with a ChatGPT clone or the AI will speak to you. While that is a valid use of TTS, in our case the use case is less interactive.
DigitalStereo Labs TTS purpose is for users to create audio voiceovers for social media videos, audio clips for media apps (think video games), and to allow for fast iterations of this process.
Now, what does this do for voice actors? Well, actually they can sell their talents at scale. They can record their voice, train an AI model, and sell that AI voice model via platforms such as DigitalStereo Labs - the individual per session is going to be smaller, however, their licenced voices can be used across the country or when they have a cold or are otherwise undisposed
Where are we at with DigitalStereo Labs?
This week the team has been working on following:
Landing page
Beta Preview signup page
Stripe account for DigitalStereo
Removing baseline NeuralDreams functionality not needed, such as Video datasrouce upload, PDF doc upload
Name one feature of text to speech usage you may be interested to hear about once the DigitalStereo preview is online? Such as reading your substack newsletter?