Explore

Slice the corpus on any axis.

Filter by firm, year, function, metric or slide category — then see the result as a narrative arc, a treemap of slide types, or a catalog of slides.

clear
51,416 slides across 1229 decks match.
Treemap Arc Catalog
Once a “Llama rip-off”, developers are increasingly building on China’s Qwen
AirStreetCapital · industry_trends
Once a “Llama rip-off”, developers are increasingly building on China’s Qwen
>40% — Model derivative share
 Why are researchers going Chinese?
AirStreetCapital · industry_trends
Why are researchers going Chinese?
 World models step out of the clip: real-time, interactive video arrives
AirStreetCapital · industry_trends
World models step out of the clip: real-time, interactive video arrives
 The Genie in three acts: from sketches to image-prompted persistent worlds
AirStreetCapital · process_diagram
The Genie in three acts: from sketches to image-prompted persistent worlds
 Training agents inside of scalable world models
AirStreetCapital · case_study
Training agents inside of scalable world models
100x — Success rate (%)
 China's video generation matures: a strategic divergence
AirStreetCapital · industry_trends
China's video generation matures: a strategic divergence
 OpenAI launches Sora 2: controllable video-and-audio inches toward a world simulator
AirStreetCapital · case_study
OpenAI launches Sora 2: controllable video-and-audio inches toward a world simulator
55% — GPQA Diamond accuracy
 Generated worlds enable practical open-ended learning
AirStreetCapital · industry_trends
Generated worlds enable practical open-ended learning
 How should we measure progress on open-endedness?
AirStreetCapital · diagnosis
How should we measure progress on open-endedness?
 From tools to collaborators: AI agents as partners in discovery
AirStreetCapital · case_study
From tools to collaborators: AI agents as partners in discovery
 AlphaEvolve: a coding agent for algorithm discovery and engineering impact
AirStreetCapital · case_study
AlphaEvolve: a coding agent for algorithm discovery and engineering impact
0.7% — Target metric (aggregated)
 A universal interface model for biology?
AirStreetCapital · case_study
A universal interface model for biology?
 Scaling to Universal Atomistic Models
AirStreetCapital · case_study
Scaling to Universal Atomistic Models
3.7B — Val Loss
 From property predictions to material generation
AirStreetCapital · case_study
From property predictions to material generation
106 — Number of SUN structures found
 Language models in Chemistry: from property predictors to strategy-aware planners
AirStreetCapital · industry_trends
Language models in Chemistry: from property predictors to strategy-aware planners
 Robot chemists scale discovery at 10x human speed and 1,000 experiments per day
AirStreetCapital · case_study
Robot chemists scale discovery at 10x human speed and 1,000 experiments per day
1,000 — experimental throughput
 LLM-driven tree search writes expert-level scientific software across domains
AirStreetCapital · case_study
LLM-driven tree search writes expert-level scientific software across domains
40/87 — Public leaderboard percentile
 Scaling laws in genomics: predictable gains with compute, data, and context
AirStreetCapital · industry_trends
Scaling laws in genomics: predictable gains with compute, data, and context
 Scaling laws for proteins unlock broader and more useful generation
AirStreetCapital · industry_trends
Scaling laws for proteins unlock broader and more useful generation
46B — Validation Loss
 AlphaFold 3 reproductions: strong on familiar chemistry, weak on novelty
AirStreetCapital · case_study
AlphaFold 3 reproductions: strong on familiar chemistry, weak on novelty
 AMIE address multimodal diagnostic consultations in longitudinal care and w/oversight
AirStreetCapital · case_study
AMIE address multimodal diagnostic consultations in longitudinal care and w/oversight
59.1% — Diagnostic Accuracy (%)
 Advancing multimodal foundation models and LLM decision support for healthcare
AirStreetCapital · case_study
Advancing multimodal foundation models and LLM decision support for healthcare
75% — clinical improvement
 Diffusion language models: parallel denoising challenges autoregression
AirStreetCapital · industry_trends
Diffusion language models: parallel denoising challenges autoregression
 Tokenizer-free LLMs via dynamic byte patches
AirStreetCapital · case_study
Tokenizer-free LLMs via dynamic byte patches
 Attention sinks aren't a bug...they're the brakes
AirStreetCapital · other
Attention sinks aren't a bug...they're the brakes
78% — sink metric prevalence
 The trouble with benchmarks: vibes aren't all we need, but increasingly all we've got
AirStreetCapital · problem_statement
The trouble with benchmarks: vibes aren't all we need, but increasingly all we've got
100-point — Maximum Arena Score
 The trouble with benchmarks: safety-washing
AirStreetCapital · diagnosis
The trouble with benchmarks: safety-washing
71% — Attack failure rate
 LLMs are professional yes-men, and we trained them to be that way
AirStreetCapital · key_messages
LLMs are professional yes-men, and we trained them to be that way
 Brain-to-text decoding: decoding brain activity during typing
AirStreetCapital · case_study
Brain-to-text decoding: decoding brain activity during typing
19% — Character error rate
 Can vision models align with human brains...and how does that alignment emerge?
AirStreetCapital · case_study
Can vision models align with human brains...and how does that alignment emerge?
 Scaling laws for brain-to-image decoding: data per subject matters (and costs bite)
AirStreetCapital · case_study
Scaling laws for brain-to-image decoding: data per subject matters (and costs bite)
$1,093/hr — Pearson R
 ATOKEN: a unified visual tokenizer for images, video, and 3D
AirStreetCapital · case_study
ATOKEN: a unified visual tokenizer for images, video, and 3D
 Merging the virtual and physical worlds: pre-training on unstructured reality
AirStreetCapital · industry_trends
Merging the virtual and physical worlds: pre-training on unstructured reality
 The architectural divide: knowledge insulation vs. end-to-end adaptation
AirStreetCapital · framework_other
The architectural divide: knowledge insulation vs. end-to-end adaptation
 Emergent reasoning moves into the physical world
AirStreetCapital · industry_trends
Emergent reasoning moves into the physical world
 Reading the road: processing driving tasks in a unified language space
AirStreetCapital · case_study
Reading the road: processing driving tasks in a unified language space
 Computer Use Agents (CUA) have improved by leaps and bounds, and still fall short
AirStreetCapital · industry_trends
Computer Use Agents (CUA) have improved by leaps and bounds, and still fall short
59.8 — Mean Normalized Score
 Could small language models be the future of agentic AI?
AirStreetCapital · industry_trends
Could small language models be the future of agentic AI?
10-30x — Cost reduction
 Headline AI agent designed innovations require domain-expert audits
AirStreetCapital · case_study
Headline AI agent designed innovations require domain-expert audits
 Model Context Protocol becomes the “USB-C” of AI tools
AirStreetCapital · industry_trends
Model Context Protocol becomes the “USB-C” of AI tools
3x — Research paper citations
 The explosion of AI agent frameworks...
AirStreetCapital · industry_trends
The explosion of AI agent frameworks...
 ...and an explosion of diverse AI agent research papers
AirStreetCapital · industry_trends
...and an explosion of diverse AI agent research papers
40,551 — Number of research papers
 Building agents that remember: from context windows to lifelong memory
AirStreetCapital · industry_trends
Building agents that remember: from context windows to lifelong memory
2,437 — Research paper count
 AI conferences’ capacity crisis
AirStreetCapital · industry_trends
AI conferences’ capacity crisis
29k — Number of papers published
 RIP AGI, long live Superintelligence
AirStreetCapital · other
RIP AGI, long live Superintelligence
 So what will frontier superintelligence actually cost?
AirStreetCapital · industry_trends
So what will frontier superintelligence actually cost?
$300bn — Capital expenditure
 Days spent at the frontier
AirStreetCapital · competitive_analysis
Days spent at the frontier
249 — Days/Weeks at #1
 More for less: trends in capability to cost ratios are encouraging (Artificial Analysis)
AirStreetCapital · industry_trends
More for less: trends in capability to cost ratios are encouraging (Artificial Analysis)
19.5 — Intelligence to Price Ratio
Next page →