Cognitive Blind Spot 2: Bandwagon Surfing or Do What May Be Fashionable
October 6, 2023
Note: This essay is the work of a real and still-alive dinobaby. No smart software involved, just a dumb humanoid.
The litigation about the use of Web content to train smart generative software is ramping up. Outfits like OpenAI, Microsoft, and Amazon and its new best friend will be snagged in the US legal system.
Humans are into trends. The NFL and Taylor Swift appear to be a trend. A sporting money machine and a popular music money machine. Jersey sales increase. Ms. Swift’s music sales go up. New eyeballs track a certain football player. The question is, “Who is exploiting whom?”
Which bandwagon are you riding? Thank you, MidJourney. Gloom seems to be part of your DNA.
Think about large language models and smart software. A similar dynamic may exist. Late in 2022, the natural language interface became the next big thing. Students and bad actors figured out that using a ChatGPT-type service could expedite certain activities. Students could be 500 word essays in less than a minute. Bad actors could be snippets of code in seconds. In short, many people were hopping on the LLM bandwagon decorated with smart software logos.
Now a bandwagon powered by healthy skepticism may be heading toward main street. Wired Magazine published a short essay titled “Chatbot Hallucinations Are Poisoning Web Search.” The foundational assumption is that Web search was better before ChatGPT-type incursions. I am not sure that idea is valid, but for the purposes of illustrating bandwagon surfing, it will pass unchallenged. Wired’s main point is that as AI-generated content proliferates, the results delivered by Google and a couple of other but vastly less popular search engines will deteriorate. I think this is a way to assert that lousy LLM output will make Web search worse. “Hallucination” is jargon for made up or just incorrect information.
Consider this essay “Evaluating LLMs Is a Minefield.” The essay and slide deck are the work of two AI wizards. The main idea is that figuring out whether a particular LLM or a ChatGPT-service is right, wrong, less wrong, more right, biased, or a digital representation of a 23 year old art history major working in a public relations firm is difficult.
I am not going to take the side of either referenced article. The point is that the hyperbolic excitement about “smart software” seems to be giving way to LLM criticism. From software for Every Man, the services are becoming tools for improving productivity.
To sum up, the original bandwagon has been pushed out of the parade by a new bandwagon filled with poobahs explaining that smart software, LLM, et al are making the murky, mysterious Web worse.
The question becomes, “Are you jumping on the bandwagon with the banner that says, “LLMs are really bad?” or are you sticking with the rah rah crowd? The point is that information at one point was good. Now information is less good. Imagine how difficult it will be to determine what’s right or wrong, biased or unbiased, or acceptable or unacceptable.
Who wants to do the work to determine provenance or answer questions about accuracy? Not many people. That, rather then lousy Web search, may be more important to some professionals. But that does not solve the problem of the time and resources required to deal with accuracy and other issues.
So which bandwagon are you riding? The NFL or Taylor Swift? Maybe the tension between the two?
Stephen E Arnold, October 6, 2023
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
