Live data from Hacker News

Douglas Lenat's Automated Mathematician Source Code

github.com

11–13 of 13 posts

Re: Douglas Lenat's Automated Mathematician Source Code

#11

Can someone in the know give a little summary of what we’re looking at here? What’s the purpose? How effective is the code/system at accomplishing its purpose? Etc…

Automated Mathematician is a historically significant step in the evolution of classic AI based on evaluating symbols and rules. This branch of AI seems to have hit a dead end although one can never be certain of such things.

Obviously stuff like LLMs produces much more impressive results as of now, that's a given. OTOH who knows - neural networks have also had a long-ish period when OCR seemed to be the pinnacle of what they can deliver before they exploded via Deep Learning/Transformers/LLMs and what not.

Re: Douglas Lenat's Automated Mathematician Source Code

#12
post #11

Can someone in the know give a little summary of what we’re looking at here? What’s the purpose? How effective is the code/system at accomplishing its purpose? Etc…

Automated Mathematician is a historically significant step in the evolution of classic AI based on evaluating symbols and rules. This branch of AI seems to have hit a dead end although one can never be certain of such things. Obviously stuff like LLMs produces much more impressive results as of now, that's a given. OTOH who knows - neural networks have also had a long-ish period when OCR seemed to be the pinnacle of…

Indeed, AFAIK neural networks have caused at least two AI winters before finally breaking through thanks to a few good new ideas and the fact that the needs of computer games incidentally led to the development of a big industry of specialized, programmable, high-performance dot product calculators.

Re: Douglas Lenat's Automated Mathematician Source Code

#13
post #12
post #11

Earlier quoted context omitted.

Automated Mathematician is a historically significant step in the evolution of classic AI based on evaluating symbols and rules. This branch of AI seems to have hit a dead end although one can never be certain of such things. Obviously stuff like LLMs produces much more impressive results as of now, that's a given. OTOH who knows - neural networks have also had a long-ish period when OCR seemed to be the pinnacle of…

Indeed, AFAIK neural networks have caused at least two AI winters before finally breaking through thanks to a few good new ideas and the fact that the needs of computer games incidentally led to the development of a big industry of specialized, programmable, high-performance dot product calculators.

Speaking of winters; there's a good article about Cyc, a successor to Automated Mathematician. Cyc was the last big project in symbolic AI: https://yuxi.ml/cyc
Post reply on HN