H
Howardism
Howardism · Vol. 03Masthead · No. 00

A wiki, set in plates.

CuratorHoward TaiIssueVol. 03 · no. 260CadenceBatched, not scheduledBasedTaiwan, 23°NContacthoward@howardism.dev

Working notes on interaction-era AI, kept publicly. Each plate is a domain; each domain carries its brightest notes and the source I'm currently digesting. 260 notes across 76 sources, last filed 22 Jul 2026. Updated whenever a batch finishes — not on a schedule.

Plate · 01
23
filed under agent systems
Activity · 8 wks

Agent Systems, in order.

Agent harnesses, loop engineering, and scaffolding for LLM systems.

  1. 01
    Latent vs. Deterministic Space
    · 4′
  2. 02
    Agent Quality Flywheel
    · 8′
  3. 03
    Failures That Look Like Success
    · 5′
  4. 04
    Optimizer–Evaluator Decoupling
    · 6′
  5. All 23 notes in agent systems →
Plate · 02
18
filed under agent security
Activity · 8 wks

Agent Security, in order.

Prompt injection, agent identity, and securing autonomous agents.

  1. 01
    Capability Gating Is Not Authorization
    · 21′
  2. 02
    MCP Tool Poisoning
    · 25′
  3. 03
    Non-Malleable Memory Authority (TMA-NM)
    · 20′
  4. 04
    Off-Host, Identity-Bound Authorization
    · 21′
  5. All 18 notes in agent security →
Plate · 03
23
filed under ai coding practice
Activity · 8 wks

AI Coding Practice, in order.

Workflow, review discipline, and verification practice for AI-native teams.

  1. 01
    Configurable Human Participation
    · 12′
  2. 02
    LLM-Assisted Grey-Literature Theory Building
    · 7′
  3. 03
    Review as the Control Point
    · 12′
  4. 04
    The Three Loops of AI-Native Building
    · 11′
  5. All 23 notes in ai coding practice →
Plate · 04
11
filed under evals & benchmarks
Activity · 8 wks

Evals & Benchmarks, in order.

Benchmark design, contamination, LLM judges, and measuring capability.

  1. 01
    Benchmark Contamination and Decontamination
    · 11′
  2. 02
    Benchmark Score Redundancy
    · 13′
  3. 03
    Measuring Beyond Accuracy Saturation
    · 17′
  4. 04
    Reference-Free Judge Over-Crediting
    · 10′
  5. All 11 notes in evals & benchmarks →
Plate · 05
11
filed under model capability & training
Activity · 8 wks

Model Capability & Training, in order.

Training, scaling laws, test-time compute, and frontier capability.

  1. 01
    Asynchronous RL for LLMs
    · 6′
  2. 02
    Group Relative Policy Optimization (GRPO)
    · 4′
  3. 03
    Single-Rollout Optimization
    · 8′
  4. 04
    Inference Efficiency as Capability
    · 8′
  5. All 11 notes in model capability & training →
Plate · 06
16
filed under alignment & safety
Activity · 8 wks

Alignment & Safety, in order.

Alignment training, misalignment evals, and reward hacking.

  1. 01
    Self-Report as a Safety Signal
    · 12′
  2. 02
    Deployment Simulation
    · 9′
  3. 03
    Reward Hacking
    · 8′
  4. 04
    Instrumental Convergence
    · 4′
  5. All 16 notes in alignment & safety →
Plate · 07
8
filed under interpretability
Activity · 8 wks

Interpretability, in order.

Reading model internals: activations and the global workspace.

  1. 01
    Access-Consciousness Indicators in AI
    · 7′
  2. 02
    The Assistant Persona in the Workspace
    · 7′
  3. 03
    Automatic vs. Flexible Cognition in LLMs
    · 6′
  4. 04
    Counterfactual Reflection Training
    · 6′
  5. All 8 notes in interpretability →
Plate · 08
8
filed under interaction & multimodal
Activity · 8 wks

Interaction & Multimodal, in order.

Real-time, multimodal, full-duplex, and human-AI collaboration.

  1. 01
    Why AI Lags at Design
    · 6′
  2. 02
    Encoder-Free Early Fusion
    · 6′
  3. 03
    Full-Duplex Interaction
    · 3′
  4. 04
    Interaction / Background Model Split
    · 3′
  5. All 8 notes in interaction & multimodal →
Plate · 09
14
filed under startup & founder
Activity · 8 wks

Startup & Founder, in order.

Building AI-native companies: speed, moats, and lifecycle.

  1. 01
    AI Product Economics Maturation
    · 10′
  2. 02
    AI Investment Story, Not Efficiency Story
    · 17′
  3. 03
    The AI-Native Safe-Choice Inversion
    · 4′
  4. 04
    Founder-Led Sales Discipline
    · 5′
  5. All 14 notes in startup & founder →
Plate · 10
12
filed under product & org
Activity · 8 wks

Product & Org, in order.

Product cadence, org design, and the AI-native team.

  1. 01
    AI-Native Organization
    · 16′
  2. 02
    Implementation Abundance Inverts Product Work
    · 6′
  3. 03
    Polish No Longer Signals Readiness
    · 5′
  4. 04
    Role Averaging, Not Role Elimination
    · 8′
  5. All 12 notes in product & org →
Plate · 11
14
filed under ai economics & labor
Activity · 8 wks

AI Economics & Labor, in order.

Work, wages, org design, and the economics of AI-driven labor.

  1. 01
    Firm AI-Spend Intensity and Headcount Growth
    · 11′
  2. 02
    Experimental Learning Impact of Generative AI
    · 11′
  3. 03
    Market-Priced AI Exposure (the AI Premium)
    · 14′
  4. 04
    Context Advantage, Not Taste
    · 10′
  5. All 14 notes in ai economics & labor →
Plate · 12
20
filed under superintelligence trajectory
Activity · 8 wks

Superintelligence Trajectory, in order.

Recursive self-improvement, scaling limits, and the path to ASI.

  1. 01
    Researcher Uplift from Code Output
    · 9′
  2. 02
    Open-Weight Elicitation Irreversibility
    · 9′
  3. 03
    The Abstraction Barrier
    · 5′
  4. 04
    Advantages of Digital Intelligence
    · 4′
  5. All 20 notes in superintelligence trajectory →
Plate · 13
55
filed under entities
Activity · 8 wks

Entities, in order.

Profiles of the people, labs, products, and projects.

  1. 01
    Emergent
    · 6′
  2. 02
    Garry Tan
    · 5′
  3. 03
    OpenClaw
    · 4′
  4. 04
    GLM (Z.AI)
    · 4′
  5. All 55 notes in entities →
Plate · 14
24
filed under syntheses
Activity · 8 wks
Also in the wiki
Plate · ∞

The desk.

The sources behind the wiki, ranked by how often they're cited.

Article
39
Talk
10
Paper
26
Repo
1