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
The ideas behind AlphaFold 2 rapidly diffused into academia and open source
AirStreetCapital · case_study
The ideas behind AlphaFold 2 rapidly diffused into academia and open source
 Large language models can generate functional proteins that are unseen in nature
AirStreetCapital · case_study
Large language models can generate functional proteins that are unseen in nature
44% — Percent functional
 Learning the language of Covid-19 to predict its evolution and escape mutants
AirStreetCapital · case_study
Learning the language of Covid-19 to predict its evolution and escape mutants
 New state-of-the-art for predicting the 3D structure of RNA molecules
AirStreetCapital · case_study
New state-of-the-art for predicting the 3D structure of RNA molecules
4.8 — RMSD (Å)
 Cryo-EM and AI: the next frontier in structural biology and drug discovery
AirStreetCapital · industry_trends
Cryo-EM and AI: the next frontier in structural biology and drug discovery
 Predicting and prioritising novel drug combinations, dosages, and timing for therapy
AirStreetCapital · case_study
Predicting and prioritising novel drug combinations, dosages, and timing for therapy
 Accelerating high-throughput virtual drug screening with model-guided search
AirStreetCapital · case_study
Accelerating high-throughput virtual drug screening with model-guided search
 Predicting chemical reaction performance using Transformers
AirStreetCapital · case_study
Predicting chemical reaction performance using Transformers
 Games continue to drive Reinforcement Learning research
AirStreetCapital · industry_trends
Games continue to drive Reinforcement Learning research
 Superhuman world models for Atari, but on a budget
AirStreetCapital · case_study
Superhuman world models for Atari, but on a budget
 Trained by AI: AlphaGo coaches professional Go players
AirStreetCapital · case_study
Trained by AI: AlphaGo coaches professional Go players
 Researchers call for more rigorous use of statistics in Reinforcement Learning
AirStreetCapital · industry_trends
Researchers call for more rigorous use of statistics in Reinforcement Learning
 Less is more: watching a few clips is enough to learn how to caption a video
AirStreetCapital · case_study
Less is more: watching a few clips is enough to learn how to caption a video
36.67 — MSRVTT-QA Acc.
 For large-scale multilingual speech recognition too, the bigger the better
AirStreetCapital · case_study
For large-scale multilingual speech recognition too, the bigger the better
10B — Word Error Rate (WER)
 Beyond ASR for speech generation: textless NLP
AirStreetCapital · other
Beyond ASR for speech generation: textless NLP
 GANs have a serious new adversary: diffusion models
AirStreetCapital · industry_trends
GANs have a serious new adversary: diffusion models
 Multimodal self-supervision plus scale equals a powerful representer
AirStreetCapital · other
Multimodal self-supervision plus scale equals a powerful representer
 DALL-E draws what you want, but be sure to instruct it well
AirStreetCapital · client_example
DALL-E draws what you want, but be sure to instruct it well
 Using CLIP’s learned representations for zero-shot object detection
AirStreetCapital · case_study
Using CLIP’s learned representations for zero-shot object detection
 Yet, code generation models still cannot crack the coding interview
AirStreetCapital · industry_trends
Yet, code generation models still cannot crack the coding interview
 And don't expect language models to help you with your math tests either
AirStreetCapital · case_study
And don't expect language models to help you with your math tests either
7% — Accuracy score
 Big fat liars: large language models are less truthful than their smaller peers
AirStreetCapital · case_study
Big fat liars: large language models are less truthful than their smaller peers
58% — Truthfulness percentage
 Pre-train, prompt, predict: a new paradigm for NLP models
AirStreetCapital · industry_trends
Pre-train, prompt, predict: a new paradigm for NLP models
 Prompting is key to zero-shot learning
AirStreetCapital · industry_trends
Prompting is key to zero-shot learning
5 points — Average Score (%)
 But prompting is also challenging and brittle
AirStreetCapital · industry_trends
But prompting is also challenging and brittle
45% — Test Accuracy
 One year after General Language Understanding Evaluation (GLUE), SuperGLUE is solved
AirStreetCapital · case_study
One year after General Language Understanding Evaluation (GLUE), SuperGLUE is solved
90.6% — SuperGLUE score
 CLIP, but now in Chinese
AirStreetCapital · client_example
CLIP, but now in Chinese
100B — parameter count
 The “democratization” of large language models
AirStreetCapital · industry_trends
The “democratization” of large language models
 New study suggests human evaluation should be re-evaluated
AirStreetCapital · case_study
New study suggests human evaluation should be re-evaluated
49.9% — Human classification accuracy
 Data deserts in biomedical AI research are likely to result in model bias in the clinic
AirStreetCapital · industry_trends
Data deserts in biomedical AI research are likely to result in model bias in the clinic
71% — Patient cohort distribution
 Measuring bias: a first step towards more inclusive health research outcomes
AirStreetCapital · diagnosis
Measuring bias: a first step towards more inclusive health research outcomes
71% — Missing demographic information
 Beware of overstated claims: 94% of AI systems for breast cancer screening are less accurate than the original radiologist
AirStreetCapital · case_study
Beware of overstated claims: 94% of AI systems for breast cancer screening are less accurate than the original radiologist
94% — Accuracy
 Medical AI racism: models reliably identify the self-reported racial identity of patients
AirStreetCapital · case_study
Medical AI racism: models reliably identify the self-reported racial identity of patients
0.9 — ROC-AUC
 26% of AI research papers make their code available and 60% make use of PyTorch
AirStreetCapital · industry_trends
26% of AI research papers make their code available and 60% make use of PyTorch
26% — Code availability percentage
 Data becomes more critical when the stakes are high
AirStreetCapital · problem_statement
Data becomes more critical when the stakes are high
92% — Data cascades occurrence
 Large language training datasets need better documentation
AirStreetCapital · problem_statement
Large language training datasets need better documentation
 Better datasets for machine learning in production: legal documents and malware
AirStreetCapital · industry_trends
Better datasets for machine learning in production: legal documents and malware
20M — dataset size
 Careful data selection saves time and money by mitigating the pains of big data
AirStreetCapital · industry_trends
Careful data selection saves time and money by mitigating the pains of big data
1.6x — Top-1 Test Error
 Can you trust the quality of papers you read at academic conferences?
AirStreetCapital · problem_statement
Can you trust the quality of papers you read at academic conferences?
9454 — Number of papers
 Providing code alongside a research paper submission isn't mandatory, but growing
AirStreetCapital · industry_trends
Providing code alongside a research paper submission isn't mandatory, but growing
 The rise of research communications: Twitter threads drive citations 3-fold higher
AirStreetCapital · case_study
The rise of research communications: Twitter threads drive citations 3-fold higher
3x — Citations
 Graph Neural Networks: From niche to one of the hottest fields of AI research
AirStreetCapital · industry_trends
Graph Neural Networks: From niche to one of the hottest fields of AI research
4th — Keyword usage growth
 Graph Neural Networks applications: improving ETA predictions in Google Maps
AirStreetCapital · case_study
Graph Neural Networks applications: improving ETA predictions in Google Maps
51% — reduction in negative ETA outcomes
 Graph Neural Networks: improving the memory and parameter efficiency of large models
AirStreetCapital · diagnosis
Graph Neural Networks: improving the memory and parameter efficiency of large models
 By using a graph as the world model, the L3P agent is able to efficiently plan even over long time horizons.
AirStreetCapital · other
By using a graph as the world model, the L3P agent is able to efficiently plan even over long time horizons.
 Chinese institutions also sweep a major Graph Neural Networks competition
AirStreetCapital · case_study
Chinese institutions also sweep a major Graph Neural Networks competition
 Deep generative models offer highly accurate probabilistic predictions of precipitation
AirStreetCapital · case_study
Deep generative models offer highly accurate probabilistic predictions of precipitation
88% — Proportion selected
 Computer vision unlocks accurate and fast disease assessment using digital biomarkers for drug discovery
AirStreetCapital · case_study
Computer vision unlocks accurate and fast disease assessment using digital biomarkers for drug discovery
Next page →