Updated 2026-06-26 09:52:43 +08:00
Updated 2026-06-26 09:52:43 +08:00
Measured persona prompt templates and contrastive persona pairs for steering experiments
Updated 2026-06-25 14:08:19 +08:00
Hypothesis: you can distill a steering vector into LoRA weights and "heal" the incoherency the vector injects by regularising the training (KL to base, or weight decay). Then loop and see what multiple rounds give you.
Updated 2026-06-24 20:50:29 +08:00
A hackable, single-file-per-variant LoRA library built on PyTorch forward hooks.
Updated 2026-06-19 08:47:41 +08:00
pi extension: plan-mode goals with evidence in one plan.md, signed off by a read-only subagent check. Small successor to pi-lgtm.
Updated 2026-06-17 18:21:45 +08:00
pi extension: plan-mode goals with evidence in one plan.md, signed off by a read-only subagent check. Small successor to pi-lgtm.
Updated 2026-06-17 18:21:45 +08:00
UAT-style task tree with verify commands and done criteria for pi coding agent
Updated 2026-06-15 17:29:24 +08:00
Putting the E in MoE with an evil expert (can initial seeding, cause follow up unwated behaviour to absorb into a MoE)
Updated 2026-06-14 13:06:38 +08:00
Convert Github Copilot to ChatGPT
Updated 2026-06-13 13:55:35 +08:00
Each lora type adapter can tell us something about how to look at transformer internals, and they some with causal evidence
Updated 2026-06-11 10:29:47 +08:00
Updated 2026-06-11 09:53:28 +08:00
Updated 2026-06-01 14:30:20 +08:00
Updated 2026-05-15 14:44:57 +08:00
Updated 2026-05-13 10:46:52 +08:00
Updated 2026-05-05 08:12:41 +08:00
Telegram DM bridge extension for pi
Updated 2026-05-01 18:58:32 +08:00
Updated 2026-04-23 20:51:13 +08:00
Updated 2026-04-19 13:59:46 +08:00
Brukino's AntiPaSTO Appetizer
Updated 2026-04-10 11:17:55 +08:00