BRM-loading · layering pigment…
BRM-loading · layering pigment…
Open R&D · computing · 10 open problems · substrate contribution paths
Open problems in computing R&D the conventional industry has not closed despite annual spend in the billions. From LLM hallucination grounding to logical-qubit threshold to heterogeneous-accelerator compilation, each row names structural failure + substrate path. Average time-to-impact for substrate contributions: 31 months.
cumulative annual R&D spend addressed · USD millions
v1.2026.05.16 · 10 problems · max spend $2,000M · block 946,772
Open-problem ledger · substrate-contribution pathways · Each row names a structurally unsolved R&D problem, the reason the conventional industry has not closed it, and the substrate contribution path the BRMSTE substrate offers. Substrate modes: physics-decomposer · sieve-inference · cross-civilizational-historical-anchor · agentic-routing · three-body-failure-audit. See /legal/liability.
| # | Open problem | Why unsolved | R&D ($M) | Substrate contribution | T-to-impact (mo) |
|---|---|---|---|---|---|
| 01 | Post-quantum cryptography migration post-quantum-crypto-migration | standardisation complete (CRYSTALS-Kyber, Dilithium) · integration risk + hybrid-fallback attack surface | 2,000 | agentic security audit across deployed lattice variants | 30 |
| 02 | AI alignment (scalable oversight) ai-alignment-scalable-oversight | human feedback doesn't scale · inverse-scaling of oversight cost with model capacity | 500 | agentic three-body audit (model behavior × human eval × audit residue) | 30 |
| 03 | Hallucination grounding in LLMs hallucination-grounding-llm | factuality loss appears intrinsic to next-token prediction at >1T parameters | 300 | agentic kernel routing (factual grounding sieve + confidence threshold + residual audit) | 21 |
| 04 | Neuromorphic chip energy efficiency (10x brain) neuromorphic-10x-brain | spike timing + learning rules unstandardised · digital integration poor | 400 | physics decomposer on membrane-potential × synaptic-plasticity × power-dissipation coupling | 42 |
| 05 | Quantum error correction (logical-qubit threshold) quantum-error-correction-threshold | surface-code threshold requires <0.1% gate error · current ~0.1-1% | 800 | agentic per-qubit coherence monitoring + three-body audit (gate-fidelity × environment-coupling × calibration-drift) | 60 |
| 06 | Interpretability of foundation-model internals interpretability-foundation-internals | scaling limits of mechanistic interpretability · high-dim latent spaces resist decomposition | 200 | sieve inference on attention-head specialisation + residual-stream contribution | 24 |
| 07 | Memory-compute co-design bandwidth scaling memory-compute-bandwidth | HBM bandwidth caps ~900 GB/s · transformer inference memory-bound | 600 | physics decomposer on latency × power × footprint tensors for chiplet fabrics | 30 |
| 08 | Compilation optimisation for heterogeneous accelerators heterogeneous-compilation | 50+ accelerator types · divergent ISAs · no unified IR portable across vendors | 300 | agentic code-generation routing (operator decomposition + hardware-specific kernels) | 21 |
| 09 | Scaling RL to continuous control at billion-step scale rl-continuous-billion-step | sample efficiency plateaus · reward specification remains brittle | 150 | agentic reward audit (specification drift × environment mismatch × learned-policy residue) | 30 |
| 10 | Byzantine-resilient distributed ML at scale byzantine-resilient-distributed-ml | poison-attack detection cost scales · honest-fraction assumptions break | 200 | three-body audit (participant commitment × data-integrity × audit-trail) | 24 |
“10 open problems in Computing · AI · chip architecture. Cumulative annual R&D spend addressed: $5,450M. Each row names the substrate-contribution path. The substrate publishes what it can move; counterparties verify the path before applying.”
Wave OPEN-RD · computing · v1.2026.05.16