MIT Exhibition "AI: Mind the Gap"
"GPT-3 [Generative Pre-trained Transformer] is an AI that can generate text by predicting the words that should follow a given word or phrase."
Keeping up to date with the latest ....
-
Generative AI Product Tracker, from Ithaka (the parent company of JSTOR)"The Tracker is a living document, which we update regularly as new products enter the market or new information about existing products becomes available."
NYT: What Makes A.I. Chatbots Go Wrong?
-
“What Makes A.I. Chatbots Go Wrong?"Metz, Cade. “What Makes A.I. Chatbots Go Wrong?” The New York Times, 29 Mar. 2023, https://www.nytimes.com/2023/03/29/technology/ai-chatbots-hallucinations.html.
Why not train all generative AI bots using "good" writing from reliable sources?
-
Massachusetts Library System, Generative AI LibGuideSelection of articles curated by Jane Stimpson, of the Massachusetts Library System, as part of the MLS Generative AI LibGuide.
SAMPLE CHAT: Gemini explains calculus. TAKEAWAY: Human's language holds meaning; Generative AI holds numerical vectors.
-
Transcript of dialog with Google's GeminiSD asked Gemini: "Are you able to explain the principles of calculus in a way that is comprehensible to an adult with little background in math?" Then I followed up with a question about how Gemini found and formulated the answer so fast:
[SD's prompt / question]
Is your knowledge base cataloged or indexed, so that words and phrases appropriate to calculus, such as "derivative" and "integral," are available to be retrieved and used to compose your answer? Is your large language model divided into subsets, based on the perceived topic?
[Google Gemini's response]
That's an excellent set of questions that touches upon the internal workings of large language models like me!
Is your knowledge base cataloged or indexed?
Yes, absolutely, though not in a way that directly maps to human-readable "catalogs" or "indexes" in the traditional sense.
Think of it less like a library with books arranged by subject and more like a vast, multi-dimensional web of interconnected concepts and relationships. When I'm trained, I don't just memorize isolated facts. Instead, I learn the statistical relationships between words, phrases, ideas, and contexts.
- Embeddings: Every word, phrase, and even concept is represented internally as a numerical vector (an "embedding") in a high-dimensional space. Words and phrases that are semantically similar (e.g., "derivative," "rate of change," "slope of tangent") are located closer together in this space. This proximity allows for efficient retrieval and association.
- Semantic Networks (Implicit): While not explicitly built as a symbolic "semantic network" by humans, the training process effectively creates an implicit semantic network . When you mention "calculus," concepts like "derivative," "integral," "limit," "instantaneous change," "accumulation," "slope," "area under curve," and their associated contexts (physics, economics, engineering) are all highly activated and brought into consideration.
- Pattern Recognition: I learn patterns in how language is used. For example, I learn that "derivative" is frequently followed by phrases like "of a function," "with respect to," and "represents the instantaneous rate of change." This allows me to generate coherent and contextually appropriate text.
SD's takeaways:
-
Humans learn words to convey and understand meaning.
-
Generative AI assigns numerical vectors to words, enabling statistical pattern recognition.
Gemini's explanation of calculus was not nearly as helpful as the articles written by Professor Steven Strogatz for The New York Times, cited below:
-
Professor Steven Strogatz of Cornell University: NYT series on mathSteven Strogatz is the Susan and Barton Winokur Distinguished Professor for the Public Understanding of Science and Mathematics at Cornell University. In addition to several books and articles, Strogatz has authored two series for The New York Times, The Elements of Math and Me, Myself and Math.
What's Not to Like about Generative AI? (Or, what is risked?)
-
"What I Learned Serving on My University’s AI Committee," by Megan Fritts. Chronicle of Higher Education, May 23, 2025."Comparisons are sometimes made between LLMs and calculators, to make the point that AI bans are as futile as calls to return to the abacus.."[...]
"If, on the other hand, the aim of our disciplines is the formation of human persons, then the real threat AI poses is not one of job replacement or grading frustration or having to reimagine assignments but something entirely different. From this perspective, language-generating AI, whether it is utilized to write emails or dissertations, stands as an enemy to the human form of life by coming between the individual and her words." -
"A.I. Might Take Your Job. Here Are 22 New Ones It Could Give You," by By Robert Capps. The New York Times, June 17, 2025."First, a confession: I tried writing this essay with A.I [...]
So why aren’t you reading that version? Well, for starters, it would have gotten me fired: Almost all quotes and experts in the article were entirely made up. [...] Even if the A.I.-written version of this piece was entirely factual, submitting it to my editors would have represented a fundamental misunderstanding of why they hired me. In freelance journalism, as in many fields where the work product is written text, you aren’t just being paid (graded) for the words you submit. You’re being paid to be responsible for them: the facts, the concepts, the fairness, the phrasing. This article is running with my byline, which means that I personally stand behind what you’re reading; by the same token, my editor is responsible for hiring me, and so on, a type of responsibility that inherently can’t be delegated to a machine."
Upcoming Trends: Increase in the Digital Divide
| ACCESS | TRAINING DATA and SOURCE MATERIAL |
USE CASES |
|---|---|---|
| Free |
Scrapes text off of the Internet.
|
AI generated writing (Example: ChatGPT)
|
|
Subscription
|
Premium versions may claim to be based on newer, larger, and cleaner training datasets.
|
For users who wish to have access to:
|
|
AI tools incorporated into other subscription services / products
|
Articles contained within an enclosed environment, under copyright protection, and secured by vendor database constraints and contracts. Generative AI, search engines, and databases: "Answering your AI questions: Search engines, PDF readers, and special collections." |
AI summaries of journal article(s)
|
Energy, Environmental, & Financial Costs
-
How A.I. Is Changing the Way the World Builds Computers.Metz, Cade, et al. “How A.I. Is Changing the Way the World Builds Computers.” The New York Times, 17 Mar. 2025, https://www.nytimes.com/interactive/2025/03/16/technology/ai-data-centers.html.
