abrax3141

I've found, as I'm sure many of you have, that using GPTs for coding has many pitfalls. I won't even being to make a list (although respondents might want to, just to have it all in one place.) I've taken to basically GAN'ing several GPTs againts one another, asking each to complete or critique or modify (or all of the above) code genereated both another, and then back and forth, or round and round. This works (at least for me) extremely well, but as a result I can't really take advantage of any of the GPT-based coding tools becasue they don't really allow this sort of interaction, so I have to zip the whole thing, give it to the other bot (or at least the relevant components), etc. which is very clunky, but does work very well. It would seem like there should be a tool for this mode of work. Is there?

I'm looking for a collaborator on an obscure yet historically significant programming project. As some of you know, I previously worked with the team that reanimated ELIZA (https://news.ycombinator.com/item?id=44300641). I'm now working on reanimating the original Logic Theorist (LT), the legendary theorem-proving program written in IPL-V by Newell, Shaw, and Simon at RAND in the mid-1950s (https://news.ycombinator.com/item?id=43822619). LT was one of the first AI programs ever created. It proved theorems in Principia Mathematica, and is considered a founding work of artificial intelligence. And IPL-V, a direct forerunner of Lisp, was explicitly designed to support AI programming (more precisely, symbolic heuristic programming). Fortunately, IPL-V is well-documented, and I've written an interpreter for it in Lisp that (mostly) correctly runs the original LT (https://news.ycombinator.com/item?id=43822619). But I've hit some nuances that I'm having trouble thinking through, and feel like I need fresh eyes and someone to rubber duck with. (Because I'm the only person on earth currently working in IPL-V, I can't ask ChatGPT or Stack Overflow as there is no active community for it, and no historical information other than a few badly OCR'ed publications! I've actually tried feeding ChatGPT and Claude the IPL-V manual and having them help, but, as anyone who has used LLMs to program well knows, they're good for boilerplate in common modern languages, but pretty poor at understanding subtle programming issues, especially in a dead language!) So I'm looking for an Lisp hacker with a fascination for AI and computing history, or who just wants to be able to say they are one of two hackers alive who have worked on one of the first AI systems and languages. The only reward I can offer, aside from that badge of honor, is academic co-publication, as there's significant academic interest in early AI. (Once LT runs, there are several other of the earliest AIs that we should be able to breathe life into!) So if you're intrigued by the idea of bringing the first AI programs back to life on the first AI programming language (really, one of the first ever programming languages, and almost certainly one of the most esoteric!), DM me.

(Part of) the team who recently reanimated the original ELIZA (https://news.ycombinator.com/item?id=42746506) is now working to reanimate what are often considered the first AIs -- the symbolic heuristic programs built by Newell and co at RAND in the 1950. Most of these programs were written in IPL-V, the fascinating (if ugly) machine level language for a stack machine that was the first to introduce list processing, recursion, and symbolic computing, among other innovations that we take for granted today.

There were many implementations of IPL-V. According to the manual (https://bitsavers.org/pdf/rand/ipl/Information_Processing_La...) implementations existed for at least these machines: B220, CDC 1604 and G-20, Ferranti Mercury, IBM 650, 704, 709-7090, 1620, Philco 2000, UNIVAC 1105 and 1107, and the AN/FSQ-32 (whatever that was). And we know that it was also on the IBM 360 and 7094.

If you have access to a code archive for any of these machines, and can search it for the IPL-V interpreter, we would greatly appreciate it. (A print out is fine -- preferred in fact!) We are building one ourselves in (ironically) Lisp, but it would be amazing to be able to run an original stack, as we did with the 7094->CTSS->MAD stack in reanimating ELIZA.

We (obviously) open source everything we do.

You can post here, or DM me (my email is in my "HN about"). Thanks!