In the past year, I wrote 2500 articles. That's the equivalent of the work of three full-time journalists.
How is this possible?
Of course, using AI. It might seem that with the trendy models available to everyone, we have so many functions at our fingertips that we can skip the unfashionable stage of data preparation by using some magical trick or wand.
Is this a problem? It depends.
Data preparation for training language models is one of the most difficult parts of the entire process. Theory says it accounts for 80% of the costs in creating "intelligent" solutions, and practice fully confirms this. If someone has already done this hard work for us, there's a good chance we're feeding data to these complex mechanisms.
So how can we still use our potentially weak position to our advantage?
By using an efficient and cost-effective data acquisition strategy, an effective data loading and transformation process, operating, of course, in a low-cost cloud, and taking advantage of the offerings from leading #LLM model providers, automated agents replacing journalists could easily prepare content that satisfies readers.
The time needed to prepare such a summary article is negligible: 60 articles are produced within 60 minutes. Input data, such as a recording from a meeting on the other side of the world, is transcribed and then material is created that captures the most important facts. Besides summarizing, key quotes are extracted from the unstructured recording that engage the reader. The total cost of such an article is just a few cents. Despite concerns about lower quality compared to the work of a live journalist, the benefit to the publisher is over 2.5 million impressions annually.
Shifting the perspective slightly on AI services, we can see that they have become another layer of cloud services. Just like users of #IaaS, #PaaS, or serverless solutions, by using AI provider APIs, we gain access to an even more complex service. Yes, the LLM service is another layer of the cloud!
How to extract value from it?
By combining it with other known layers.
Simply adding the right input with good data, as in the example above, significantly increases value.
Is this example the future of journalism? Certainly not, but the leverage that such a process generates using a new layer of the cloud is impressive.
This humble entry was written by a human :-)
Author: Mariusz Jażdżyk
