Deep learning
Deep learning still uses several architectures. Qwen combines alternatives to attention with sparse experts, and low-precision training keeps improving. The cost of training and the cost of running models both matter, and we have not seen capital move wholesale from one to the other.
Research snapshot · updated Nov 12, 2026
Core thesis
Every major AI capability — LLMs, computer vision, drug discovery, protein folding — runs on deep learning. The transformer architecture (2017) dominates. But state-space models (Mamba), mixture-of-experts, and neuro-symbolic approaches are gaining ground. The 'bitter lesson' holds: scale and compute beat hand-engineering. The question is whether we're approaching diminishing returns on the current paradigm.
State of the art (2026)
The Transformer is still the base of the leading AI models whose designs are public. Hybrid designs are proving workable. NVIDIA's Nemotron-H research reports models that mix Mamba and Transformer layers and match or beat open Transformer models of similar size on accuracy, with faster inference on some workloads. The makers of the leading proprietary models do not disclose their architectures in full, so we cannot say how far they have moved. Much recent progress has come from spending more computation when the model answers, through step-by-step reasoning and checking.
Explore the deeper analysis in CanaryIQ
Explore what is covered below. These previews show the structure; log in to read the full analysis.
Signal stack
Evidence stacked leading → lagging
11 signals
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
1 tracked
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what, and who depends on whom
196 players · 6 layers
Landscape map
Who builds what, and who depends on whom
Catalyst calendar
Dated events that will move the position
3 ahead
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
8 milestones
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers, each with a condition for removing it
20 · 20
Watchlists
Companies, people and papers, each with a condition for removing it
Decision frameworks
What our view means for your job
Locked
Decision frameworks
What our view means for your job
Thesis changelog
When our view changed, and why
8 updates
Thesis changelog
When our view changed, and why
Change our mind
2 disconfirming conditions
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and a record of every time our view on Deep learning has changed. All of it is in CanaryIQ.