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
AI takes over the workforce
misc · 2023 · context
AI takes over the workforce
 Future optimism gaps
misc · 2023 · data_table
Future optimism gaps
83% — Percentage of respondents
 How tech will drive AI’s growth by solving for human values
misc · 2023 · quote_slide
How tech will drive AI’s growth by solving for human values
 How we can build needed trust in AI through equity
misc · 2023 · quote_slide
How we can build needed trust in AI through equity
 Why ethics should be at the center of new AI tech
misc · 2023 · quote_slide
Why ethics should be at the center of new AI tech
 Why humans plus AI are key to revolutionizing healthcare
misc · 2023 · quote_slide
Why humans plus AI are key to revolutionizing healthcare
 How AI will help people be more creative
misc · 2023 · quote_slide
How AI will help people be more creative
 What it will take to help people trust AI for democracy
misc · 2023 · quote_slide
What it will take to help people trust AI for democracy
 How AI can reduce the friction in work and life
misc · 2023 · quote_slide
How AI can reduce the friction in work and life
 Scorecard: Reviewing our predictions from 2019
AirStreetCapital · section_divider
Scorecard: Reviewing our predictions from 2019
 AI research is less open than you think: Only 15% of papers publish their code
AirStreetCapital · industry_trends
AI research is less open than you think: Only 15% of papers publish their code
15% — Code Availability
 Papers With Code tracks openly-published code and benchmarks model performance
AirStreetCapital · context
Papers With Code tracks openly-published code and benchmarks model performance
3,000 — State-of-the-Art leaderboards
 Facebook’s PyTorch is fast outpacing Google’s TensorFlow in research papers, which tends to be a leading indicator of production use down the line
AirStreetCapital · market_landscape
Facebook’s PyTorch is fast outpacing Google’s TensorFlow in research papers, which tends to be a leading indicator of production use down the line
75% — PyTorch adoption rate
 PyTorch is also more popular than TensorFlow in paper implementations on GitHub
AirStreetCapital · industry_trends
PyTorch is also more popular than TensorFlow in paper implementations on GitHub
47% — Share of implementations
 Language models: Welcome to the Billion Parameter club
AirStreetCapital · industry_trends
Language models: Welcome to the Billion Parameter club
175B — Parameter count
 Bigger models, datasets and compute budgets clearly drive performance
AirStreetCapital · analyze_data
Bigger models, datasets and compute budgets clearly drive performance
 Tuning billions of model parameters costs millions of dollars
AirStreetCapital · industry_trends
Tuning billions of model parameters costs millions of dollars
$10M — Training cost
 To achieve the needed quality improvements in machine translation, Google's final model trained for the equivalent of 22 TPU v3 core years or ~5 days with 2,048 cores non-stop
AirStreetCapital · analyze_data
To achieve the needed quality improvements in machine translation, Google's final model trained for the equivalent of 22 TPU v3 core years or ~5 days with 2,048 cores non-stop
22 — TPU v3 core years
 We're rapidly approaching outrageous computational, economic, and environmental costs to gain incrementally smaller improvements in model performance
AirStreetCapital · data_table
We're rapidly approaching outrageous computational, economic, and environmental costs to gain incrementally smaller improvements in model performance
10^112 — Computational cost
 A larger model needs less data than a smaller peer to achieve the same performance
AirStreetCapital · analyze_data
A larger model needs less data than a smaller peer to achieve the same performance
 Low resource languages with limited training data are a beneficiary of large models
AirStreetCapital · case_study
Low resource languages with limited training data are a beneficiary of large models
 Even as deep learning consumes more data, it continues to get more efficient
AirStreetCapital · industry_trends
Even as deep learning consumes more data, it continues to get more efficient
2x — Training efficiency factor
 Yet, for some use cases like dialogue small, data-efficient models can trump large models
AirStreetCapital · case_study
Yet, for some use cases like dialogue small, data-efficient models can trump large models
68.2% — 1-vs-100 Accuracy
 A new generation of transformer language models are unlocking new NLP use-cases
AirStreetCapital · industry_trends
A new generation of transformer language models are unlocking new NLP use-cases
 Computer, please convert my code into another programming language
AirStreetCapital · case_study
Computer, please convert my code into another programming language
90% — translation success rate
 Computer, can you automatically repair my buggy programs too?
AirStreetCapital · case_study
Computer, can you automatically repair my buggy programs too?
 NLP benchmarks take a beating: Over a dozen teams outrank the human GLUE baseline
AirStreetCapital · industry_trends
NLP benchmarks take a beating: Over a dozen teams outrank the human GLUE baseline
90.6 — Benchmark Score
 What’s next after SuperGLUE? More challenging NLP benchmarks zero-in on knowledge
AirStreetCapital · industry_trends
What’s next after SuperGLUE? More challenging NLP benchmarks zero-in on knowledge
69% — Weighted accuracy
 The transformer’s ability to generalise is remarkable. It can be thought of as a new layer type that is more powerful than convolutions because it can process sets of inputs and fuse information more globally.
AirStreetCapital · case_study
The transformer’s ability to generalise is remarkable. It can be thought of as a new layer type that is more powerful than convolutions because it can process sets of inputs and fuse information more globally.
 Biology is experiencing its “AI moment”: Over 21,000 papers in 2020 alone
AirStreetCapital · industry_trends
Biology is experiencing its “AI moment”: Over 21,000 papers in 2020 alone
21,000 — Number of publications
 From physical object recognition to “cell painting”: Decoding biology through images
AirStreetCapital · case_study
From physical object recognition to “cell painting”: Decoding biology through images
 Deep learning on cellular microscopy accelerates biological discovery with drug screens
AirStreetCapital · case_study
Deep learning on cellular microscopy accelerates biological discovery with drug screens
 Ophthalmology advances as the sandbox for deep learning applied to medical imaging
AirStreetCapital · case_study
Ophthalmology advances as the sandbox for deep learning applied to medical imaging
 AI-based screening mammography reduces false positives and false negatives in two large, clinically-representative datasets from the US and UK
AirStreetCapital · case_study
AI-based screening mammography reduces false positives and false negatives in two large, clinically-representative datasets from the US and UK
9.40% — Sensitivity
 Causal Inference: Taking ML beyond correlation
AirStreetCapital · industry_trends
Causal Inference: Taking ML beyond correlation
 Causal reasoning is a vital missing ingredient for applying AI to medical diagnosis
AirStreetCapital · case_study
Causal reasoning is a vital missing ingredient for applying AI to medical diagnosis
25% — Clinical accuracy
 Model explainability is an important area of AI safety: A new approach aims to incorporate causal structure between input features into model explanations
AirStreetCapital · other
Model explainability is an important area of AI safety: A new approach aims to incorporate causal structure between input features into model explanations
 Reinforcement learning helps ensure that molecules you discover in silico can actually be synthesized in the lab. This helps chemists avoid dead ends during drug discovery.
AirStreetCapital · case_study
Reinforcement learning helps ensure that molecules you discover in silico can actually be synthesized in the lab. This helps chemists avoid dead ends during drug discovery.
98% — Activity
 Have your desired molecule? ML will generate a synthesis plan faster than you can
AirStreetCapital · case_study
Have your desired molecule? ML will generate a synthesis plan faster than you can
90.40% — Test set accuracy
 Graph neural networks: Solving problems by making use of 3D input data
AirStreetCapital · context
Graph neural networks: Solving problems by making use of 3D input data
 Graph networks learn to guide antibiotic drug screening, leading to new drugs in vivo
AirStreetCapital · case_study
Graph networks learn to guide antibiotic drug screening, leading to new drugs in vivo
 Enhancing chemical property prediction using graph neural networks
AirStreetCapital · case_study
Enhancing chemical property prediction using graph neural networks
10x — Mean-squared error (MSE)
 AI sifts through chemical space using DNA-encoded small molecule libraries (DEL)
AirStreetCapital · case_study
AI sifts through chemical space using DNA-encoded small molecule libraries (DEL)
72% — Hit rate
 Language models show promise in learning to predict protein properties from amino acid sequences alone
AirStreetCapital · case_study
Language models show promise in learning to predict protein properties from amino acid sequences alone
 COVID-19: Analyzing symptoms from over 4 million contributors detects novel disease symptom ahead of public health community and could inform diagnosis without tests
AirStreetCapital · case_study
COVID-19: Analyzing symptoms from over 4 million contributors detects novel disease symptom ahead of public health community and could inform diagnosis without tests
4 million — Odds ratio
 Drug discovery goes open source to tackle COVID-19. This is a rare example of where AI is being actively used on a clearly-defined problem that's part of the COVID-19 response.
AirStreetCapital · case_study
Drug discovery goes open source to tackle COVID-19. This is a rare example of where AI is being actively used on a clearly-defined problem that's part of the COVID-19 response.
 Missed out on strawberries and cream this year? A controllable synthetic video version of Wimbledon tennis matches
AirStreetCapital · case_study
Missed out on strawberries and cream this year? A controllable synthetic video version of Wimbledon tennis matches
 Attention turns to computer vision tasks like object detection and segmentation
AirStreetCapital · industry_trends
Attention turns to computer vision tasks like object detection and segmentation
Next page →