Transcript
0:00 ever lit $10,000 on fire with OpenAI? As you can see here, for this list enrichment, $10,000 of OpenAI spend was wasted because we used the wrong prompt. The worst part is this could have been done for about $1,000. Here's how AI works. You're charged based on two basic factors. One is how smart the model is, and two, how much information it has to process. It's a multiplication equation, and the smarter the model, the more context it has to process. The more expensive the output, the more you pay.
0:28 Now when it comes to a job like AI industry segmentation which is a very common use case in clay for go to market use cases you're going to use clientent to scrape the web give a website search the website summarize what the company does and then determine based on that company summary maybe a few other questions which category do they best fit into based on your custom AI segments. So for example this Clayton prompt is going to take this website and it's going to research the site and decide which bucket they fit into. It's going out and it's actually scraping the whole the whole website, maybe for other websites, which is a lot of context. And then it's making a decision. That's a lot at once for a very smart model. And you can see here it's costing us 1.5 cents per call. 1.6 2 cents a call, right? 2.8 cents a call, right? 3.4 cents per call. This is super expensive.
1:13 Now, on the other hand, we can create a very cheap summary of what a company does using 40 Mini, which is a very cheap model. And then we can use a smarter model 40 to have the final um decision made. and use the intelligence of 40 to have a a more accurate output, but decreasing the context window of 40 by giving it less input, less information. We're only giving it the company summary that we generated previously. We're outputting our decisions for 0.003 cents over here. So, as you can see here, the total AI cost between these two prompts is 0.0 is 4/10en of a penny for the same output as opposed to about 2 cents on average, which significantly decreases the cost while still maintaining a high degree of accuracy. There you have it folks, a way to significantly reduce your AI cost while still maintaining a high degree of accuracy.
Summary
- Spending $10,000 on OpenAI was largely wasted due to incorrect prompting.
- AI costs are determined by model intelligence and the amount of information processed.
- A common use case involves segmenting companies based on website data.
- High-context prompts can lead to costs of 1.5 to 3.4 cents per call.
- Using a cheaper model for initial summaries and a smarter model for final decisions can drastically reduce costs.
- The optimized approach can lower costs to 0.003 cents per output while maintaining accuracy.
- Effective prompting strategies can lead to significant savings in AI expenditures.