AI offers a new resource to students and researchers. This page presents some experiments in AI applicable the computation and statistics in Economics. Caution is needed. Developments in AI appear at breakneck speed. I date the experiments because even as I post them, the information is out of date. No one can be a “expert” in the use of AI, because development is so rapid. Even those who work at the forefront, specialize, and can only comment on one aspect of AI. A researcher working on images will be uninformed about the latest tweaks to large language models.
Gemini Evolves in a week (February 23, 2024)
Gemini, Google’s response to OpenAI’s CHAT GPT system has come under fire for offering biased information. One caution – in response to the criticism of its bias, Gemini developers appear to have created amnesia and records of earlier sessions no longer appear. This is an important lesson; copy AI sessions to your desktop as a Word document (remember to date them)
Here are two examples of how Gemini evolved in a week:
Experiment 1 (Image generation)
This experiment prompted Gemini to develop an image of people looking at a screen, using the cubist style. I then posed some politically incorrect additions to the prompt. (View experiment 1 here). Gemini retreated when confronted with obvious absurdities such as portraying Black Vikings.
Experiment 2 (Defining a “hot” term … “settler colonialism”)
On other tasks, Gemini also performed curiously. In a first attempt at defining settler colonialism with an example, used Israel. A week later, the definition had been refined, and the European settlement of North America became the example (View Here). Within a day, when prompted for two examples, it offered the United States and Australia (View here).
The key ideas are 1) AI is far from ready for prime time, 2) the process of refining its corpus (the knowledge base AI mines to create a response) remains a mystery, and 3) as its output becomes part of the corpus will the separation of fact from fiction became impossible?
Experiments in Regression
Open AI (Chat GPT 3.5 and Char GPR 4.) is making rapid progress in computation. Here are three experiments, starting with one in mid-January 2024.
Experiment 1 (Regression Experiment 1 – January 15, 2024). This experiment uses a “sequential prompt” to run a regression mode. It uses GPT 4.0. (View here)
Experiment 2 (Regression Experiment 2 – Feb 12 2024). GPT 3.5 allowed access to special Chats such as Data Analyst. This experiment uses an example from a simple econometrics quiz to run regressions and compute elasticities. It uses a sequential prompt where each step of the problem becomes a prompt. (View here.)
Experiment 3 (Regression Experiment 3 – Feb 12, 2024) Here prompt becomes the complete set of questions that comprise the same quiz used in experiment 2, submitted as a block. This experiment shows the capability of Chat 3.5 to create output in Word and prepare code. (View here)
