First, you comment was a nice addition to the lproven article and was thought-provoking. >> The story about evolution is totally wrong. Actually, this is exac

First, you comment was a nice addition to the lproven article and was thought-provoking. >> The story about evolution is totally wrong. Actually, this is exactly the story of evolution. Mammals are the simple, stupid things that showed up when the MUCH smarter and capable dinosaurs ruled the world. But there is a HUGE amount of homology in the biological tree of life. That is true in computer evolution too. But more in core hardware and less in platforms and in software - which is what lproven was mostly writing about. Core hardware evolves more like basic metabolic and signaling pathways and channels - very slowly and with stability. Hardware platforms evolved in pretty dramatic step functions. Some of those steps did effectively obsolete everything that came before - like the EISA bus. But computer software is very much like human software - it can be largely rewritten in a evolutionary blink of the eye. > But for the last couple decades not even that is true. It's not true in computers for the same reason it's not true with humans in ecological evolution. We've won. There will likely be no next evolutionary step biologically. That's a change from the history of the last several hundred million years and one of which we don't yet full understand the ramifications. In computer tech, it's why we now have the Magnificent Seven, and why it differs from the Nifty Fifty. The Nifty Fifty didn't have an evolutionary moat. I was very actively in the computer industry from 85 to 2000 and witnessed first-hand the very Darwinian evolution at work in computer tech. My first job (88-90) was at Westinghouse building a bespoke nuclear plant operating system running on minicomputers and Sun workstations. Then a new species arrived running on EISA bus and Windows OS and they kicked our ass even though we'd been doing that type of work for two decades. I also saw it at my second job (91-95 - studied neuroscience in between) at Dataviews - the dominant computer workstation software vendor at the time. We ran on all the major and most of the minor Unix workstations. We'd port the software to your workstation for about $100K and had many takers - which meant that it ran on basically everything. But it didn't run on Windows. And so we hit a cost/produtivity wall and had other companies kick our ass that were Windows-native just as had happened at Westinghouse. The web arrived and HTML/Javascript kicked the ass of everything else. I was an X/Motif expert and trainer in the early 90s. Great tech. Dead end because it's license couldn't compete against free. Is machine learning different? Certainly, there have been some step functions ("Attention is all you need") but mostly it's been a slow evolution of Moore's Law and better backpropogation stacks. I did my early independent research in AI at CMU in '88 and mostly ran my backpropogation models in Excel. Biologically, this isn't so different from a slug brain vs a human brain. Same basic hardware but 100,000x more capable. If I'd told my 20 year old self that in 40 years we'd have backprop models with a trillion parameters that could run on PCs, I'd have said "yeah right - in some very distant future". But it was/is a pretty distant future.