The lost cause of the Lisp machines (www.tfeb.org)

This piece argues that the era of dedicated Lisp machines is definitively over: Symbolics’ decline (officially bankrupt by 1993) and the demise of the Symbolics 3600-era hardware in the 1980s show special Lisp hardware was a short-lived niche whose advantages evaporated as RISC and commodity systems improved. The author dismisses common romantic claims — that Lisp machines’ development environment or microcoded hardware made them uniquely valuable, or that they were dramatically faster — noting concrete evidence (early RISC boxes running Kyoto Common Lisp outperformed Symbolics on Gabriel benchmarks) and explaining that many of the Lisp-machine conveniences arose from constraints of the time (tiny memory, slow CPUs), not from essential hardware features. For the AI/ML community this is a cautionary, practical lesson: modern compilers (SBCL, LispWorks) and toolchains, plus technologies like LLVM and just-in-time compilation used by languages such as Julia, have removed the historical reasons for bespoke Lisp hardware. The “Lisp all the way down” idea still has appeal for interactivity, but it would need isolation and security guarantees absent from 1980s designs — and none of that requires new silicon. The broader implication: rather than nostalgia for specialized boxes, progress now comes from better compilers, runtime systems, and leveraging commodity accelerators (GPUs) — a reminder to prioritize software and system architecture over chasing vintage hardware myths.