The Business Model

This is an experimental feature that aims to test artificial intelligence applied to our archive. The TTA's AI - Overview - Donate

AI comes at costs: that's is why

The AI Business Model

Dear User,
now that we've gotten excited about new technologies, let's move on to the sour notes.
AI comes at a cost.
OpenAI provides its service by charging on a pay-as-you-go basis.
Below you can see a breakdown of the cost based on the fact that we have about 2,500 users and (assuming) 10 percent of those users have a desire (or need) to interact with AI.

The short version:

in the scenario I just outlined (10% of users using the AI Assistant) it takes 315$ per month (284€ at today's exchange rate).

We therefore have two possible scenarios:
a) we give up providing this feature (due to lack of funds)
b) we provide the service for a fee (with a donation-paywall?).

Given that our (Andy and me) agreements were to run a nonprofit site (and remain so) we must at least return expenses.

The question, then, is this: is the benefit we bring to TTA (and the traditional music world) greater with the introduction of AI, or is the nonprofit guise of TTA inescapable?

——

The long version:

to estimate the monthly cost of offering our OpenAI Assistant (TTAssist) to our subscribers, we’ll need to consider several factors:

Model Used: The cost per token varies depending on the AI model you’re using (e.g., GPT-3.5 Turbo, GPT-4).
Usage Volume: The number of interactions you expect per month.
Token Consumption: The average number of tokens per interaction, including both user input and assistant response.

You can think of tokens as pieces of words, where 1,000 tokens is about 750 words.

Identify the AI Model and Pricing

I believe that the only model applicable to our needs is the GPT-4, and so I will not consider other models (such as GPT-3.5 Turbo).
As of my last update in September 2024, here are the approximate costs (please check the OpenAI Pricing page for the most recent rates):

GPT-4 (8K context):
$0.00500 per 1,000 tokens for input
$0.01500 per 1,000 tokens for output

Estimate Usage Volume

Since we have 2,500 subscribers, we need to estimate how many will use the assistant and how frequently:

Active Users per Month: Let’s assume 10% of subscribers will use the assistant regularly.

2,500 subscribers * 10% = 250 active users
Interactions per User per Month: Assume each active user has 10 interactions.
250 active users * 10 interactions = 2,500 total interactions per month

Estimate Token Consumption

An average interaction includes both the user’s question and the assistant’s response:

User Input: ~100 tokens (about 75 words)
Assistant Response: ~200 tokens (about 150 words)
Total Tokens per Interaction: 100 (input) + 200 (output) = 300 tokens

Calculate Monthly Token Usage

Total Tokens per Month: 2,500 interactions * 300 tokens = 750,000 tokens
Input Tokens Total: 2,500 interactions * 100 tokens = 250,000 tokens
Output Tokens Total: 2,500 interactions * 200 tokens = 500,000 tokens

Calculate Monthly Cost

For GPT-4 (8K context):

• Input Cost: (250,000 tokens / 1,000) * $0.00500 = $37.50
• Output Cost: (500,000 tokens / 1,000) * $0.01500 = $225.00
• Total Cost: $37.50 (input) + $225.00 (output) = $262.50

Add a Buffer for Variability

It’s prudent to include a buffer (e.g., 20%) to account for unexpected spikes in usage:

GPT-4 Total with Buffer: $262.50 * 1.20 = $315.00

Conclusion

Estimated Monthly Cost:
Using GPT-4 (8K context): Approximately $315.00 per month

All this is to say that a couple of bucks/month each is enough.