I wouldn't say that g++ is AGI. But I looked at some of the optimizer output in assembler and I would say the depth of understanding of the source code bordered on narrow AI. It was certainly more than I was expecting.
I was on the OpenCog mailing list for awhile before I got booted off for being overly critical of the approach instead of contributing to the code base. It doesn't matter now because the project was abandoned after a decade of effort without producing anything. I understand why people wanted to use old school symbolic knowledge representation from the 1980s instead of neural networks when they didn't have a supercomputer or a rack full of GPUs. People actually believed they could write a program whose most obvious application would be to automate $1 quadrillion worth of human labor. -- Matt Mahoney, [email protected] On Mon, Oct 5, 2026, 12:18 PM twenkid <[email protected]> wrote: > Hi Basil, > > *Basil: *"*Are open source inference engines (coded in C++, GPL license, > compiled by GCC which is also a compiler mostly coded in C++ and also GPL > licensed) considered on this mailing list as AI software?"* > > > > *Todor:* > I do consider them AI software. I've reviewed your project several times, > first after I discovered it via your Stack Overflow account, maybe in late > 2021, but I didn't dive deeply enough in order to actually run it, use it > or modify it; the project had very specific requirements besides its > complexity and advanced level of C++. I don't recall any serious and > technical discussions within the activity of this group, though. > > Thanks for sharing the paper about your DSL for GCC MELT. > > GCC is an interesting beast as well - good food for thought that gives > insights into the process of performing gradual sequences of lowering, > while keeping semantic equivalence (or changing it in a desired direction); > and in reverse - elevating representations; the pattern matching that you > exercise with your language. The same goes for programming language > interpreters and compilers in general, as it does for LLVM. Some criticize > GCC for its "patched" nature, but this way it serves as a record of the > historical development and a path of a concrete reasoning process and > gradual development, which is more "realistic" and "educational" regarding > the process of incremental development. It carries "meta-info" about the > evolution of the translation process. > > In the paper you discuss the use of C macros and the difficulty of their > mapping to glue code in Python etc. > They are interesting pattern-matching exercises and also a step within the > logical and historical process of "highering" the code from Machine code, > simple Assemblers, more complex Macro assemblers - at different stages - > etc. > > Overall - these systems are more interesting, > algorithmic-complexity-wise, than most of the typical programs for ML, > which are considered "AI" (excluding the inference libraries themselves, > PyTorch etc., which may be interesting internally). A few short sections > with "processing" of the "vector", stacked layers, "optimize", "train", > "if a < b then repeat". "Half a page of pseudo code" or something as Andrew > Ng once promoted. While in compilers and related systems there is a huge > amount of reasoning and logic, embedded in the design, and more variety and > complexity. > > *Basil: "*In other words, what is an advanced AI software, and what are > examples > of advanced AI whose source code is open source (perhaps partly > generated by open source software, including itself)" > > > > *Todor:* > OpenCog was a big open source AGI project in C++ etc. and the very early > calls for volunteers for Ben Goertzel's AGI projects were for C++ > developers. Modern OpenCog is such software, I guess. > > This is archived in the hyperbook *The Prophets of the Thinking Machines*, > the volume *"World Class AI Institutes and Strategies from Eastern Europe > and the World"* , compared to the* Russian institute* "*AIRI*" - both its > title and its goal ("Strong AI") resembled Ben's *AGIRI*, but *18-20 > years later.* > > ..... > > https://web.archive.org/web/20030206221823fw_/http://www.agiri.org/joining.htm > > > "We are a small team with a big project, and are always eager for bright > minds with appropriate skills, interested in joining our team. > > Currently we cannot pay salaries to any new recruits, but we can offer the > thrill of being involved in what is perhaps the most exciting science and > engineering project on the planet. > > We are looking for two sorts of individuals at the moment: > > 1. *C++ programmers, with the ability to understand moderately complex > mathematical equations* > 2. Applied mathematicians or theoretical computer scientists with > some programming experience > > Our work relies on a large body of non-published knowledge, so new team > members should not expect to be able to contribute significantly right > away. Rather, an initial period of study will be required, followed by > participation on a limited basis by contributing to more peripheral aspects > of the project, followed by full participation in the AI development. > > If you fit into one of the above two categories, and are interested in > contributing, please contact *Ben Goertzel > <https://web.archive.org/web/20030206221823/mailto:[email protected]>*. > ... > > * Preserved at SIGI-2025: > https://github.com/Twenkid/SIGI-2025/blob/main/AI_Institutes_Strategies_The_Prophets_Thinking_Machines_7-9-2025.pdf > > See pp.56-60+ > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/Tc0ac46f2c46f6874-Mb06877ae819f1379ccd3e6ad> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tc0ac46f2c46f6874-M3eca6c0e9ab229db0bc9ec63 Delivery options: https://agi.topicbox.com/groups/agi/subscription
