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
Die Studie baut auf Hebel entlang der Wertschöpfungskette sowie drei Fragen auf – Sonderauswertung 2014 zur Produktoptimierung
RolandBerger · 2013 · framework_other
Die Studie baut auf Hebel entlang der Wertschöpfungskette sowie drei Fragen auf – Sonderauswertung 2014 zur Produktoptimierung
 Passend zum Fokus-Thema der CFO-Agenda 2014: "Produkt" haben wir eine Vielzahl an Studien – Kommen sie auf uns zu!
RolandBerger · 2013 · closing_ask
Passend zum Fokus-Thema der CFO-Agenda 2014: "Produkt" haben wir eine Vielzahl an Studien – Kommen sie auf uns zu!
 IT'S CHARACTER THAT CREATES IMPACT
RolandBerger · 2013 · closing_ask
IT'S CHARACTER THAT CREATES IMPACT
 Philippines is expected to continue its double-digit climb towards ~$35B by 2025, largely fueled by e-commerce
Bain · 2023 · market_sizing
Philippines is expected to continue its double-digit climb towards ~$35B by 2025, largely fueled by e-commerce
$35B — GMV ($B)
 Metro Manila leads in digital participation; gap widens beyond capital
Bain · 2023 · geographic_map
Metro Manila leads in digital participation; gap widens beyond capital
 Steep DFS growth projected to continue from a low base
Bain · 2023 · industry_trends
Steep DFS growth projected to continue from a low base
 HVUs spend 6.3X vs non-HVUs; transport is a key differentiator
Bain · 2023 · segmentation
HVUs spend 6.3X vs non-HVUs; transport is a key differentiator
6.3X — Spending multiplier
 Funding dropped substantially from pandemic highs, including previous heavyweight DFS
Bain · 2023 · financial_analysis
Funding dropped substantially from pandemic highs, including previous heavyweight DFS
$0.2B — Private funding value
 Inaasahang magtutuloy-tuloy ang Pilipinas sa double-digit climb nito papuntang ~$35B pagdating ng 2025, at malaking bahagi nito ay dahil sa e-commerce
Bain · 2023 · market_sizing
Inaasahang magtutuloy-tuloy ang Pilipinas sa double-digit climb nito papuntang ~$35B pagdating ng 2025, at malaking bahagi nito ay dahil sa e-commerce
$35B — GMV ($B)
 Nangunguna ang Metro Manila sa digital participation, lumalaki naman ang gap sa labas ng capital
Bain · 2023 · geographic_map
Nangunguna ang Metro Manila sa digital participation, lumalaki naman ang gap sa labas ng capital
 Tinatayang magtutuloy-tuloy ang labis na paglago ng digital financial services dahil sa malaking oportunidad mula sa low base
Bain · 2023 · industry_trends
Tinatayang magtutuloy-tuloy ang labis na paglago ng digital financial services dahil sa malaking oportunidad mula sa low base
 Gumagastos ang high-value users ng 6.3X kumpara sa non-HVUs; mahalagang sector ang transportasyon
Bain · 2023 · market_landscape
Gumagastos ang high-value users ng 6.3X kumpara sa non-HVUs; mahalagang sector ang transportasyon
6.3X — Spending multiplier
 Nananatiling halos consistent ang mga investment sa nakalipas na anim na buwan
Bain · 2023 · financial_analysis
Nananatiling halos consistent ang mga investment sa nakalipas na anim na buwan
0.2 — Private funding ($B)
 Scorecard: Reviewing our predictions from 2022
AirStreetCapital · section_divider
Scorecard: Reviewing our predictions from 2022
 GPT-4 is out and it crushes every other LLM, and many humans
AirStreetCapital · industry_trends
GPT-4 is out and it crushes every other LLM, and many humans
90% — Estimated percentile
 Fueled by ChatGPT’s success, RLHF becomes MVP
AirStreetCapital · industry_trends
Fueled by ChatGPT’s success, RLHF becomes MVP
 The false promise of imitating proprietary LLMs, or how RLHF is still king
AirStreetCapital · key_messages
The false promise of imitating proprietary LLMs, or how RLHF is still king
 Even so, researchers rush to find scalable alternatives to RLHF
AirStreetCapital · industry_trends
Even so, researchers rush to find scalable alternatives to RLHF
 The GPT-4 technical report puts the nail in the coffin of SOTA LLM research...
AirStreetCapital · industry_trends
The GPT-4 technical report puts the nail in the coffin of SOTA LLM research...
 ...unless LLaMas reverse the trend
AirStreetCapital · case_study
...unless LLaMas reverse the trend
51.6 — Zero-shot accuracy
 LLaMa sets off a race of open(ish) competitive Large Language Models
AirStreetCapital · industry_trends
LLaMa sets off a race of open(ish) competitive Large Language Models
 LLaMa-2: the most generally capable and publicly accessible LLM?
AirStreetCapital · industry_trends
LLaMa-2: the most generally capable and publicly accessible LLM?
32M — Human evaluation of helpfulness
 GPT and LLaMAs win the popularity contest
AirStreetCapital · market_landscape
GPT and LLaMAs win the popularity contest
5430 — MMLU Score
 RLHF / Instruction-tuning emerges as the most trending topic since the end of 2022.
AirStreetCapital · industry_trends
RLHF / Instruction-tuning emerges as the most trending topic since the end of 2022.
 Are emergent capabilities of language models a mirage?
AirStreetCapital · diagnosis
Are emergent capabilities of language models a mirage?
92% — Model accuracy
 Context length is the new parameter count
AirStreetCapital · industry_trends
Context length is the new parameter count
 Lost in the Middle: long contexts (mostly) don't live up to the expectations
AirStreetCapital · key_takeaways
Lost in the Middle: long contexts (mostly) don't live up to the expectations
 Keeping up with high memory demands
AirStreetCapital · industry_trends
Keeping up with high memory demands
2.8x — training speedup
 Can small (with good data) rival big?
AirStreetCapital · industry_trends
Can small (with good data) rival big?
51 — Pass@1 accuracy (%) on HumanEval
 2022 Prediction: language models trained on huge amounts of data
AirStreetCapital · case_study
2022 Prediction: language models trained on huge amounts of data
13 trillion — tokens
 Are we running out of human-generated data?
AirStreetCapital · industry_trends
Are we running out of human-generated data?
2026 — Data exhaustion date
 Breaking the data ceiling: AI-generated content
AirStreetCapital · industry_trends
Breaking the data ceiling: AI-generated content
 Disentangling the real and the fake, and surfacing the real behind the fake
AirStreetCapital · industry_trends
Disentangling the real and the fake, and surfacing the real behind the fake
 Breaking the data ceiling: overtraining
AirStreetCapital · industry_trends
Breaking the data ceiling: overtraining
 Vibe check: evaluating general-purpose LLMs leaderboards and “vibes”
AirStreetCapital · market_landscape
Vibe check: evaluating general-purpose LLMs leaderboards and “vibes”
0.914 — Mean win rate
 AlphaZero is DeepMind’s gift that keeps on giving, now for low-level code optimization
AirStreetCapital · case_study
AlphaZero is DeepMind’s gift that keeps on giving, now for low-level code optimization
70% — latency
 Where are we prompting? Take a deep breath...it’s getting sophisticated
AirStreetCapital · industry_trends
Where are we prompting? Take a deep breath...it’s getting sophisticated
 Prompt engineering trial and error
AirStreetCapital · data_table
Prompt engineering trial and error
84.0% — Accuracy
 Welcome, Agent Smith: LLMs are learning to use software tools
AirStreetCapital · industry_trends
Welcome, Agent Smith: LLMs are learning to use software tools
 Open-ended learning with LLMs
AirStreetCapital · case_study
Open-ended learning with LLMs
3.3x — Number of Distinct Items
 Reasoning with language model is planning with a world model
AirStreetCapital · other
Reasoning with language model is planning with a world model
 GPT-4 out-performs RL algorithms by studying papers and reasoning
AirStreetCapital · case_study
GPT-4 out-performs RL algorithms by studying papers and reasoning
12.3 — Reward
 Vision-language models: GPT-4 wins (but API access is still limited)
AirStreetCapital · industry_trends
Vision-language models: GPT-4 wins (but API access is still limited)
27.4% — win rate
 Leveraging LLMs and world knowledge for compositional visual reasoning
AirStreetCapital · other
Leveraging LLMs and world knowledge for compositional visual reasoning
 Leveraging LLMs for autonomous driving
AirStreetCapital · case_study
Leveraging LLMs for autonomous driving
 PaLM-E: a foundation model for robotics
AirStreetCapital · case_study
PaLM-E: a foundation model for robotics
562-billion — parameter count
 Vision-language models can be fine-tuned all the way to low-level policies showing impressive performance in manipulating objects. They also retain their ability to reason about web-scale data.
AirStreetCapital · case_study
Vision-language models can be fine-tuned all the way to low-level policies showing impressive performance in manipulating objects. They also retain their ability to reason about web-scale data.
55B — parameters
 From vision-language models to low-level robot control: RoboCat
AirStreetCapital · case_study
From vision-language models to low-level robot control: RoboCat
20Hz — real-time performance
Next page →