Issue 95: Updating ChatGPT, Cryptomining, and Website-for-Small-Libraries Threads
This week we revisit threads from a month ago, a year ago, and 12 years ago.
- ChatGPT references a non-existent book
- AI generating news articles
- Cryptocurrency miners in Texas forced to shut down due to electricity shortage
- Lament for a future that could have been: A Web presence for every library
Feel free to send this newsletter to others you think might be interested in the topics. If you are not already subscribed to DLTJ's Thursday Threads, visit the sign-up page. If you would like a more raw and immediate version of these types of stories, follow me on Mastodon where I post the bookmarks I save. Comments and tips, as always, are welcome.
ChatGPT references a non-existent book
Been seeing a lot about #ChatGPT lately and got my first question at the library this week from someone who was looking for a book that the bot had recommended. They couldn't find it in our catalog. Turns out that ALL the books that ChatGPT had recommended for their topic were non-existent. Just real authors and fake titles cobbled together. And apparently this is known behavior. 😮
One of the Thursday Threads issues last month focused on Chat-bots Powered by Artificial Intelligence. This weekend I saw this fediverse post from Wanda Whitney; a patron was looking for a book recommended by ChatGPT. This is a problem known as "stochastic parroting". Let's ask ChatGPT to define that term...
What is "stochastic parroting"? Define this term and give an example of how this is a problem for large language models like ChatGPT.
"Stochastic parroting" is a phenomenon that occurs when a language model generates text that is similar to the text it has seen during training, but is not a coherent or logical response to the input. This can happen when the model is simply repeating patterns it has learned from the training data, rather than using its understanding of the meaning of the input to generate a response.This is a problem for large language models like GPT-3 because the amount of data these models are trained on is so huge that they are able to memorize large amount of text and sometimes can repeat it back which might not be relevant to the task or context, this can be mitigated by fine-tuning the model with smaller dataset that is domain specific, or other techniques like controlling sampling temperature and restricting the model's training to a certain task.
From what I've read, that is a pretty good answer. The algorithm behind the large language model uses statistical probabilities to find the most likely next word based on the previous words. Even if that generates nonsense in the real world. I still think that the invention and wide-scale use of large-scale natural language models like ChatGPT means an increased importance on information literacy training. Let's get to it, librarians!
AI generating news articles
Also in that Thursday Threads issue was mention of how AI can change the nature of essay-writing and instructors can adapt to the change by making the assignment about editing rather than writing text. Publishers are trying out just that thing. With headines like "What Are NSF Fees and Why Do Banks Charge Them?" and "What Happens When you Bounce a Check?", these seem to be fairly basic articles based on common knowledge and easy editing. I wonder if anyone—CNET, the publisher, or an academic researcher—has reviewed those articles to see how accurate they are.
This isn't the first time automation has been used to write news articles. Back in 2015, The Verge reported on how the Associated Press was using a fill-in-the-blank system to write financial stories from regulatory filings.
Note! An update to this thread is in issue 97.
Cryptocurrency miners in Texas forced to shut down due to electricity shortage
As Texas and much of the U.S. confronts the challenges of this winter storm, the bitcoin mining industry in Texas is playing a part in supporting the Texas grid during this challenging time by proactively curtailing power. Bitcoin miners in Texas and the Texas Blockchain Council are working with grid operators in Texas as they are closely monitoring the situation, and stand ready to contribute to the state's efforts to ensure reliability and keep families warm and safe during the extreme weather.
A Thursday Thread from a year ago was cryptocurrency’s energy consumption and how cryptocurrency miners were heading for Texas after being forced out of China and Kosovo. Now comes last month's press release from the Texas Blockchain Council on how they shed load as the grid struggled to keep up with heating demands. Still no word on why using energy for fraud-supporting busy-work calculations is a better use of excess capacity than, say, some form of energy storage that would help smooth out the bumps in the grid.
Lament for a future that could have been: A Web presence for every library
Late last year -- just about a week before ALA Midwinter -- came an announcement of a project by OCLC's Innovation Lab to offer an inexpensive website to every small library. At a price of about $5 per month, a library could have a basic desktop and mobile website. At about $40 per month, the library could have a simple inventory and circulation module. You can see what is possible at the Loremville, TN public library sample site and read more information about the project in my write-up of the public demonstration.
Twelve years ago, OCLC announced a project to build a website for every library, and I wrote an excited piece about it for this blog. It seemed like the perfect use of the OCLC cooperative's resource magnification efforts; if anyone could pull off something like this, it was OCLC. Unfortunately, the project didn't last, and I wish I knew more about why it didn't take off. I couldn't find the Loremville Public Library sample site on Wayback, so the only thing that may be left is this low resolution screenshot I made for my 2010 article.
Extra Cat!
