yan@yandesbiens

yan@yandesbiens:~$ ./lab --status

Lab status

A single, honest snapshot of where the program stands. Capable AI you can own — on commodity hardware — through self-similar architecture and memory that organizes itself.

1citable outputs
1reproducible
5research threads
3articles
1newsletter issues
2open repos
Unified Fractal Memory: running a routed model larger than VRAM on a single consumer GPU

2026-06-27 · Memory Systems reproducible

We benchmark Unified Fractal Memory (UFM), a residency manager that treats GPU VRAM and CPU RAM as one elastic pool, on a routed Mixture-of-Experts whose expert bank exceeds device memory. On a 23.5 GB RTX 4090, the standard all-on-GPU placement OOMs at a 24 GB bank; UFM runs the same model holding VRAM at 19.6 GB. When the active working set fits the VRAM budget, UFM matches baseline throughput within ~1% and is ~240x faster than naive per-call CPU offloading; when every expert is touched every step (no locality), UFM is transfer-bound and offers no speedup over naive streaming. UFM is a bet on routing locality, not unbounded memory.

read → · cite · repo ↗

threadmaturityevidencenext proof
Memory Systems benchmarked Proof drop #1 (2026-06-27): runs a 24 GB expert bank on a 23.5 GB RTX 4090 where baseline OOMs; within ~1% of baseline throughput when the working set fits the budget; ~240× faster than naive offload; honest no-locality failure case documented and reproducible. Training-time paging (autograd-safe) + OffloadedAdam memory curves; cost-aware eviction vs. LRU ablation; bf16 + an NVMe tier.
Agent Runtime Systems shipped Code others can run; agents whose identity + memory are plain files. The YSON spec and an inspectable-memory demo are not yet written up. A published YSON spec; a reproducible demo of inspectable, file-based agent memory compared against opaque memory stacks.
Commodity Training shipped A live, working from-scratch training stack on one consumer GPU; no public reproducible recipe doc yet. An end-to-end reproducible run with logged loss curves, tokens/sec, and exact hardware/software versions.
Cognitive Architectures prototype A running system. Internal behaviors are not yet formally measured, so all capability claims are marked speculative. Measurable, reproducible behaviors: memory growth over time, belief-revision traces, idle-cycle throughput — before any public capability claim.
Fractal Cognition prototype Rich and documented, but NOT yet benchmarked. All architectural-advantage claims are marked speculative until measured. A controlled, small-scale comparison of the fractal backbone vs. a parameter-matched transformer (loss/perplexity at equal params + equal compute).

The next proof for each thread is in the table above. New outputs ship as proof drops — verifiable first, announced second.

yan@yandesbiens:~$ subscribe --proof-drops

Track the program.

Get each proof drop as it ships. Reproducible, no hype.

→ full research program · → citations