brianletort.ai
Industry

Companion view

The AI Shockwave Timeline.

The events that reset frontier assumptions — and the impact each one sent through the stack.

The LLM Evolutionary Tree explains where models came from. This view tracks the shocks: the moments that bent the curve on cost per useful token, reasoning at inference time, or where the bottleneck lives. Each shock carries a before-to-after delta, grouped by archetype and tagged by layer — hardware, software, networking. Click any shock to watch its impact ripple downstream.

Frontier deltas, 2024 → 2026

Context window

0M
from 32K

Gemini 1.5 → DeepSeek-V4

Reasoning · AIME 2024

0.0%
from 13.4%

GPT-4o → o1 → DeepSeek-R1

Inference token cost

0.0×
from

Blackwell → Rubin (reported)

KV cache footprint

0%
from 100%

DeepSeek-V2 → V4 @1M

15

Shocks

9

High-magnitude

5

Archetypes

2024-02 → 2026-05

Span

Click a shock to trace its downstream impact

Archetype
Layer
Magnitude
All shocks
HHardwareSSoftwareNNetworking202420252026SGemini 1.531xHNVIDIA Blackwell & …30xSDeepSeek-V2-93.3%SGPT-4o~9x fasterSLlama 3.1 405Bfirst frontie…SOpenAI o1+61 ptsNModel Context Proto…one open prot…SDeepSeek-V32.788M H800 G…SDeepSeek-R179.8%SClaude 3.7 Sonnetuser-controll…NAgent2Agent Protocolvendor-neutra…HNVLink Fusioncustom CPU/AS…HNVIDIA Rubin & Vera…10x lowerSDeepSeek-V4 Preview8xSClaude Opus 4.84x fewer
Archetype (color)
EfficiencyReasoningContext & interfaceOpen-frontierInteroperability & network
Layer (shape)
H · HardwareS · SoftwareN · Networking & protocol
Magnitude (ripple)
HighMedium

Methodology.

  • A shock resets an assumption. Inclusion requires that an event materially changed capability per unit of compute, reasoning at inference time, or where the system bottleneck lives — not merely that it topped a benchmark.
  • Impact as before → after. Every shock carries one to three deltas across capability, economics, and systems, with exactly one headline. Improvements read green; the arrow shows whether the metric went up or down.
  • Five archetypes, three layers. Color encodes archetype (efficiency, reasoning, context, open-frontier, interoperability); node shape encodes layer (hardware, software, networking); ripple size encodes magnitude.
  • Reported metrics, labeled. Figures are official vendor-reported numbers unless independently reproduced. Forward-dated or preview items carry lower confidence.
  • Curated, not exhaustive. The timeline captures the largest cross-layer inflection points defensible with primary English-language sources. Last updated 2026-05-29.

AI-readable exports.

Companions

Reset. Ripple. Repeat.