Using LLMs to monetize every keypress
Published by marco on
The article Adding AI-generated descriptions to my tools collection by Simon Willison includes the following snippet,
First of all, I’m not surprised that he asked it to shorten its descriptions. The initial versions were typically and unbearably wordy.
That he noticed and fine-tuned the generated output to waste less of his readers’ time is less interesting for me[1] than that he, once again, wrote about how much it cost to run the tool. I think it’s great that he explains how much it costs. At the same time, I take it as a sign of how quickly we acquiesce to sea-changes in our lives without even noticing that anything has changed.
Before these cloud-based tools, I have never once had to think about how much using a tool costs me past the initial purchase or annual subscription cost. The brave new world of cloud-based coding assistants has now increased the billing granularity to commercialize individual keystrokes.
In the good old days, we used to buy a tool and use it. It didn’t phone home. You got an update when you bought it in a store or when you downloaded and installed it. The next step was automated but optional updates. Some tools had automated and non-optional updates. After that came subscription-based licensing, where you rented rather than owned software.
In this next phase, you will neither own nor rent software; instead, you will pay for each move of your mouse. This is, of course, a coup for the companies running the software. It is a downgrade for a way of life that was heretofore more decoupled from immediate and constant monetization.
This is the same type of change that came for blogging and creation and run online many years ago. That’s why many people no longer do anything for free; instead, they funnel their “content” through official monetization platforms or include links to monetization platforms. The system commercializes and marketizes more of what we do every day.
Some technological changes were empowering, e.g., releasing filmmakers from the burden and cost of obtaining film. Now, those same filmmakers—or the next generation of them—will once again be yoked to a finite resource for which they will have to pay as they go.
The hope is that everyone will integrate these subscription-based, per-resource cloud resources into all of their creative workflows. In the programming world, this pattern has heretofore been the domain of B2B cloud services. Now it’s coming for everything. Everything will be a subscription. You’ll be dinged at every possible juncture.
You can either ignore the price as you work and be surprised by the bill at the end of the month … or you can start changing your work patterns to accommodate the way the tools want you to work. We’ve seen this pattern before, and it can be (relatively) benign. It’s how electricity works—but electricity is largely state-controlled and the prices are set at a point where most people hardly ever need to think about it.
Even for electricity, though, this is the case for some but there are an increasing number for whom this is not the case, for whom electricity has largely been privatized and subjected to the whims of flash commodity pricing. There are people who turn off their air-conditioners because they can’t afford to run them.
Do we want to use this same pattern—privatizing and unit-pricing essential
commodities to gate access to creation and innovation tools to those who can afford them? I personally don’t think so.
Using shorter descriptions also makes it easier to spot mistakes, which, according to AI search engines give incorrect answers at an alarming 60% rate, study says by Benj Edwards (Ars Technica), happen a lot. This result squares with my personal experience, in which I am still trying to find a pattern where I incorporate LLM-based tools without losing efficiency to constantly having to correct it.
The following citations aim to summarize the article.
“Error rates varied notably among the tested platforms. Perplexity provided incorrect information in 37 percent of the queries tested, whereas ChatGPT Search incorrectly identified 67 percent (134 out of 200) of articles queried. Grok 3 demonstrated the highest error rate, at 94 percent.”
↩“For the tests, researchers fed direct excerpts from actual news articles to the AI models, then asked each model to identify the article’s headline, original publisher, publication date, and URL. They ran 1,600 queries across the eight different generative search tools.
“The study highlighted a common trend among these AI models: rather than declining to respond when they lacked reliable information, the models frequently provided confabulations—plausible-sounding incorrect or speculative answers. The researchers emphasized that this behavior was consistent across all tested models, not limited to just one tool.”