ModelsComparator
Company belief maps
Current beliefs, hypotheses, open questions, evidence, and change-my-mind triggers for Meta, Amazon, Apple, Netflix, Google, NVIDIA, Anthropic, and OpenAI.
Interactive
Open
Loading
Models
Models
Frontier labs, model launches, SOTA rankings, and the mechanics of training and scaling. The intelligence layer.
Stories
This Week
What actually moved
OpenAIApr 23, 2026
GPT 5.5 Released
Establishes a new frontier in AI capability.
DeepSeekApr 23, 2026
DeepSeek V4 Released
Maintains pressure on proprietary labs with open-weight alternatives.
Moonshot AIApr 16, 2026
Moonshot AI Releases Kimi K2.6
Cements a sub-quarterly release cadence as the new normal for the Chinese frontier.
Live
The race, right now
Labs
Who is building the frontier
Most recent
ModelsComparator
Company belief maps
Current beliefs, hypotheses, open questions, evidence, and change-my-mind triggers for Meta, Amazon, Apple, Netflix, Google, NVIDIA, Anthropic, and OpenAI.
Apr 30, 2026
ModelsPrimer
Mission query map
Demand discovery from the way users actually ask: is AI a bubble, how do I benefit, what should I learn, where should I work?
Apr 30, 2026
ModelsPrimer
Canonical research objects
The structured substrate beneath each peregrination: topic, thesis, sources, claims, contradictions, graph links, and model lineage.
Apr 30, 2026
Explainers
How the models work
Hello World, Byte to SiliconNew
One prompt, one forward pass, traced from UTF-8 bytes through tokens, attention, and systolic arrays.
How Transformers Work
The full architecture from input to output.
Tokenization
Where text becomes numbers.
Embeddings
Meaning as geometry.
Attention
Why a 2017 paper still dictates the shape of every frontier model.
The Transformer Block
What actually happens inside one layer.
KV Cache
The trick that makes decode tractable.
Memory and GPUs
Where the real bottleneck lives.
Model Memory
State, caches, and context.
Architectural Innovations
MoE, RoPE, SSMs, and the moves that reshape the cost curve.
All Models
8 itemsModelsPrimer
Mission query map
Demand discovery from the way users actually ask: is AI a bubble, how do I benefit, what should I learn, where should I work?
Interactive
Open
ModelsPrimer
Canonical research objects
The structured substrate beneath each peregrination: topic, thesis, sources, claims, contradictions, graph links, and model lineage.
Interactive
Open
ModelsAgent
Launch-intelligence agent
Autonomous agent that watches lab blogs, vendor news, and arXiv for frontier-AI launches. Extracts structured fields plus a draft "what changed" note in Pere voice; queues candidates for human review.
Interactive
Open
ModelsAgent
On-demand visual explainer
Pick any frontier-AI concept; an agent writes a 5-6 slide swipe-through explainer with structured visuals on the fly. Schema-constrained generation — same Story format we use for the hand-built explainers, generated end-to-end.
Interactive
Open
ModelsPrimer
What is a parameter?
A primer on parameters, tokens, and training data — the three "trillion" numbers behind every frontier-model launch, and how they differ.
Interactive
Open
ModelsComparator
Frontier models, side by side
Sortable comparison of frontier models across labs — total params, active params, context window, training tokens, and open-weights status.
Interactive
Open
ModelsArticle
The Mercurial Muse
Nine months ago, we called the talent spending rational. Last week, the first model shipped. Now comes the harder question.
6 min
Open