H
Howardismvol. 03 · quiet corner of the web
Howardism · Vol. 03Plate II · No. 02

Architecture, in order.

Notes21TopicArchitectureOldest10 Apr 2026Newest23 May 2026

Model internals: encoder-free fusion, training, scaling, evals.

Architecture articles, sorted by date, newest first.
TitleSummaryDate
The AI-Native Safe-Choice InversionBuying the legacy incumbent used to be "safe"; post-AI, *being* the incumbent = not AI-native; boards give buyers air cover; a counter-positioning play
Building Is Cheap, Arguing Is Expensive"In technical debate, code wins": generate three PRs vs whiteboard; prototype over design doc; reduce design docs
Code as Source of TruthDocs go stale at high coding throughput; check specs/skills into the repo; onboard via Claude; spec-drift verification
Founder-Led Sales DisciplineStay founder-led until PMF; don't offload sales to an AE *or* an agent; explicit tension with Founder As Agent Orchestrator
Jagged Intelligence (Ghosts, Not Animals)"Ghosts not animals": jagged statistical circuits, no intrinsic motivation; car-wash/strawberry failures; stay in the loop, treat as tools
Narrow Wedge into a Legacy MarketDisrupt without being feature-complete: be the best for a narrow customer profile (tech cos outgrowing QuickBooks); Google-Sheets MVP; the wedge-flip lesson
Outsource Your Thinking, Not Your Understanding"You can outsource your thinking but not your understanding"; understanding as the non-delegable human bottleneck; knowledge bases as understanding-tools
Product Velocity as MoatShipping speed as differentiator + trust signal ("you'll scale with us"); a treadmill that must convert into durable lock-in
Software 3.0Karpathy's taxonomy: 1.0 code, 2.0 weights, 3.0 prompting; LLM as programmable interpreter; MenuGen "shouldn't exist"; neural-net-as-host-process extrapolation
The Verifiability ThesisLLMs automate what you can *verify* as computers automate what you can *specify*; RL verification rewards → jagged peaks; "verifiable + labs care"; everything eventually verifiable
Problem-Solution Fit DisciplineIdea-stage thesis: three defenses against premature building (time, resources, belief friction) all eroded; AI as devil's advocate is the antidote to confirmation-bias-with-research-engine
The Bitter LessonSutton 2019: scaled general methods beat hand-engineered structure; recurring justification across the wiki for dissolving harnesses into models; caveat — mechanical verification and character may not migrate inward
Time-Aligned Micro-TurnsThe core interaction-model move: input/output as continuous streams in ~200ms interleaved chunks, no turn boundaries; streaming-sessions inference (upstreamed to SGLang), latency-tuned MoE kernels, bitwise trainer-sampler alignment
TML-Interaction-SmallTML's first interaction model: 276B MoE / 12B active, audio+video+text in / text+audio out, 200ms micro-turns, async background agent; best turn-taking latency of any model; research preview May 2026
Model Spec Midtraining (MSM)New training phase between pretrain and AFT: train base model on synthetic docs discussing the Model Spec; controls AFT generalization; cuts agentic misalignment 54%→7%; beats deliberative alignment baseline
Model Spec ScienceEmpirical study of which Model Spec features best generalize alignment; value explanations > rules alone, specific > general "be ethical" framing; first concrete examples in Li et al. 2026
Opus 4.6 → 4.7 Changes and Multi-Agent Coding Considerations4.6→4.7 delta table + six hazards for multi-agent coding teams: role-based model selection, prompt re-tuning, harness invariants, per-agent context budget, unattended-fan-out safety, independent reviewer
Scale-Dependent Prompt SensitivityLarge models underperform small ones on 7.7% of standard benchmarks due to overthinking; brevity constraints recover 26pp and fully reverse hierarchy on GSM8K/MMLU-STEM
When to Use Claude Opus 4.6 for WorkDecision rules for Opus 4.6 deployment: solver-not-planner, elaboration-load-bearing tasks, brevity constraints, Pareto frontier check
LLM-as-Compiler Knowledge BaseKarpathy's architecture: LLM incrementally compiles raw docs into a persistent interlinked wiki, replacing RAG with a 4-phase ingest→compile→query→lint pipeline
What Are AI Tools?Overview of AI tools landscape and categories