Summary
According to a post on X by @VectorCrossProd, an ambassador project credited to divagr1925 uses the Gated DeltaNet state of Qwen3.5-4B as memory. It inserts hand-compiled transformer layers that implement a 32-bit RISC-V CPU, so the model runs instructions during its own forward pass instead of predicting arithmetic. Per an explainer quoted by LilithDatura, the original Qwen weights are left unchanged. Builders called it a striking open-source experiment, but the source snippet is cut off.
AI-assisted summary based on the listed source.
What happened
An ambassador project credited to divagr1925 uses Qwen's Gated DeltaNet state as memory and inserts hand-compiled transformer layers that implement a RISC-V CPU, so the model executes instructions inside its own forward pass instead of predicting arithmetic. The original Qwen weights are left unchanged, per an...
Why it matters
For researchers and builders studying how language models can be made to carry out computation, this is an early experiment to watch rather than a settled result. The next signals to track are whether the project publishes its code and whether independent builders can reproduce how it works.
Signal Intelligence
Signal Strength 78%
Technical label WATCH
Public Interest 42
Category OPEN SOURCE
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 51
Practical Impact Score 26
Novelty Interest Score 70
Consequence Score 26
Curiosity Score 0
Shareability Score 57
Why this is here
VQV surfaced this signal because it is recent, relevant to Open Source LLMs, connected to Scobleizer News Pulse (X).
