Recent posts

Using LLMs to Explain Historical Code: FLOW-MATIC Investigation

The new generation of code assistance tools powered by Large Language Models (LLM)s may be useful in efficiently categorizing and translating historical software corpora. In this qualitative study, we examine LLM performance in explaining and translating FLOW-MATIC programs. We also test their reliability by injecting OCR-like errors into the text. Released in 1958, FLOW-MATIC targeted business applications and used a heavily English-inspired syntax that later influenced COBOL. We find that the models perform inconsistently for describing the key aspects of a program but fare better when translating business logic into SQL. LLMs appear robust against OCR-like errors in the source code, but we caution against relying on the current technology.

2024-12-14

Chatbots Decoded: Exploring AI (Review)

Chatbots Decoded: Exploring AI is a new exhibit from the Computer History Museum (CHM) opening November 20th, 2024. In development for more than a year, the exhibit covers both the history of chatbots and conversational interfaces as well as the social ramifications of the technology. While there is not an explanation for why GPUs rather than CPUs are so useful for learning or what “attention is all you need” means, both are alluded too and I think, many audiences, including children, will enjoy it.

2024-11-15

Mathematical Analysis of the Royal Game of Ur

From the Board Games Study Journal:

Mathematical analysis of the Royal Game of Ur

Despite many discoveries and proposals for rules for the ancient board game known as the Royal Game of Ur (RGU), no mathematical analysis has yet been performed investigating those rules. In an attempt to fill that gap, this paper presents an initial mathematical analysis of the RGU from an introductory point of view. The paper deduces the overall complexity of the RGU using a state-space and game-tree complexity analysis, allowing the RGU to be compared to the popular games Checkers, Backgammon, Ludo, Chess, and Go. The paper builds upon the fundamental laws of combinatorics and probability to improve the understanding of the game: what patterns should you expect, what moves increase your chance to win, and what moves should you avoid. The paper also presents theorems to predict the probability of future dice rolls and piece movements within the game, allowing basic inferences to be made about strategy in the RGU. The game is further examined by analysing three different influences when determining the best move: advancement and attack (beneficial to the player), and captures (detrimental to the player). These influences are used to deduce explicit equations for the advantage gained by playing each possible move from a position, which allows the formalization of a strategic algorithm to play the RGU.

2024-10-29

A New Look

I’ve given this blog a new look by switching the theme from Noteworthy to Long Form, the latter a theme of my own design.

2024-10-28