An X post challenges Architect Labs’ claims about Frostbite, a CPU design the company says was built around interaction nets. Architect Labs describes using AI agents to co-design the instruction set and chip architecture from interaction-net theory, then run large swarms to design, verify, and optimize the processor through final chip-layout files. The post frames the project as the work of two very young founders, allegedly relying on the AI model Opus, and disputes the company’s presentation of the design as novel and created from scratch.
The author alleges that Frostbite closely reproduces earlier work by Efim Kubyshkin, including his thesis and Lamagraph project, as well as prior research on interaction-net processor architectures. The post points to similarities in the design and its FPGA testing, noting that both systems are currently single-core, and argues that the earlier research deserves credit. It also challenges Frostbite’s reported claim of 4.59 times better performance per watt than a modern CPU: the author interprets this as matching the performance of processors such as the Snapdragon 865 or Cortex-A53/A55 while using 78 percent less energy, and ridicules that comparison as implausible. The thread links to Lamagraph and Kubyshkin’s thesis, but the supplied material does not include evidence resolving the copying allegation or validating the performance figures.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.