I spent months trying to break the quadratic O(N^2) bottleneck of Transformers. Today I'm releasing Pulse-Field v3.0 — an event-driven, neuro-symbolic architecture that runs in O(N) time.
Benchmarks vs GPT-2 style baseline (on CPU):
- Latency: 5ms (vs 60ms)
- Context: Tested up to 100k tokens with <3ms penalty.
- Size: Starts at ~20MB (dynamic growth).
The architecture uses "Event-Driven Routing" instead of dense attention matrices. Tokens travel as impulses through a graph of specialized "crystals" (logic/memory nodes), activating only relevant paths.
This entire core was architected and coded in a 55-minute sprint using a swarm of AI agents (reasoning models) that I orchestrated to overcome the "average output" bias of standard LLMs.
Happy to answer questions about the routing logic!
https://github.com/makimilan/pulse-field-corev
A new link for those who want to check out the project while I wait to release a new post. I deleted the previous repository because everything there was literally fake.
You can check it out; I'm still delayed for the next post and can't do it yet. Please check if my AI is correct and the project is worthwhile if you'd like. I'd appreciate your feedback: https://github.com/makimilan/pulse-field-corev
You might want to read the code your AI agents are producing. Even the agents are aware that the metrics are all made up.
https://github.com/makimilan/pulse-field-core/blob/main/puls...
Thanks, most likely they adjusted something with the latest changes, unfortunately it’s still difficult to cope with hallucinations
Author here.
I spent months trying to break the quadratic O(N^2) bottleneck of Transformers. Today I'm releasing Pulse-Field v3.0 — an event-driven, neuro-symbolic architecture that runs in O(N) time.
Benchmarks vs GPT-2 style baseline (on CPU): - Latency: 5ms (vs 60ms) - Context: Tested up to 100k tokens with <3ms penalty. - Size: Starts at ~20MB (dynamic growth).
The architecture uses "Event-Driven Routing" instead of dense attention matrices. Tokens travel as impulses through a graph of specialized "crystals" (logic/memory nodes), activating only relevant paths.
This entire core was architected and coded in a 55-minute sprint using a swarm of AI agents (reasoning models) that I orchestrated to overcome the "average output" bias of standard LLMs.
Happy to answer questions about the routing logic!
https://github.com/makimilan/pulse-field-corev A new link for those who want to check out the project while I wait to release a new post. I deleted the previous repository because everything there was literally fake.
Where did the repository go? It has disappeared.
You can check it out; I'm still delayed for the next post and can't do it yet. Please check if my AI is correct and the project is worthwhile if you'd like. I'd appreciate your feedback: https://github.com/makimilan/pulse-field-corev
The repository looked kinda fake. It looks like it has been taken down?
Yes, I deleted it and will recreate it now with the changes.