Tweet
Tweet

Slide 1

Slide 1 text

Ines Montani Explosion What the history of the web can teach us about the future of AI

Slide 2

Slide 2 text

I started with the web…

Slide 3

Slide 3 text

ines.io/blog/how-i-started-coding I started with the web…

Slide 4

Slide 4 text

ines.io/blog/how-i-started-coding My Blog 2006 - 2009 I started with the web…

Slide 5

Slide 5 text

ines.io/blog/how-i-started-coding My Blog 2006 - 2009 I CSS I started with the web…

Slide 6

Slide 6 text

displaCy ines.io/blog/how-i-started-coding My Blog 2006 - 2009 I CSS I started with the web…

Slide 7

Slide 7 text

displaCy ines.io/blog/how-i-started-coding I NLP … and still love it today! My Blog 2006 - 2009 I CSS I started with the web…

Slide 8

Slide 8 text

Open-source library for industrial-strength natural language processing spacy.io spaCy 320m+ DOWNLOADS

Slide 9

Slide 9 text

Open-source library for industrial-strength natural language processing spacy.io spaCy ChatGPT can write spaCy code! 320m+ DOWNLOADS

Slide 10

Slide 10 text

Prodigy Modern scriptable annotation tool for machine learning developers prodigy.ai 900+ COMPANIES 10k+ USERS

Slide 11

Slide 11 text

Prodigy Modern scriptable annotation tool for machine learning developers prodigy.ai 900+ COMPANIES 10k+ USERS Alex Smith Developer Kim Miller Analyst GPT-4 API

Slide 12

Slide 12 text

0 % 20 % 40 % 60 % Data Analysis Web Dev Machine Learning DevOps Parsing / Scraping Python Developers Survey 2023 Python usage

Slide 13

Slide 13 text

0 % 20 % 40 % 60 % Data Analysis Web Dev Machine Learning DevOps Parsing / Scraping Python Developers Survey 2023 Python usage Web Development Machine Learning Data Analysis Research Education DevOps Data Engineering 0 % 25 % Python Developers Survey 2023 What do you use Python for the most?

Slide 14

Slide 14 text

0 % 20 % 40 % 60 % Data Analysis Web Dev Machine Learning DevOps Parsing / Scraping Python Developers Survey 2023 Python usage Web Development Machine Learning Data Analysis Research Education DevOps Data Engineering 0 % 25 % Python Developers Survey 2023 What do you use Python for the most? Python wins for AI because it’s a general- purpose language

Slide 15

Slide 15 text

Big Tech is reinventing the wheel and the road spacy.fyi/ltwc Reinvent the wheel, but don’t reinvent the road!

Slide 16

Slide 16 text

Big Tech is reinventing the wheel and the road Google most-used search engine with market share of over 80% spacy.fyi/ltwc Reinvent the wheel, but don’t reinvent the road!

Slide 17

Slide 17 text

Big Tech is reinventing the wheel and the road Google most-used search engine with market share of over 80% Chrome / Chromium most popular browser and engine powering Microsoft Edge, Opera, Arc and more spacy.fyi/ltwc Reinvent the wheel, but don’t reinvent the road!

Slide 18

Slide 18 text

Big Tech is reinventing the wheel and the road Google most-used search engine with market share of over 80% Chrome / Chromium most popular browser and engine powering Microsoft Edge, Opera, Arc and more Gemini (formerly Bard) generative AI model and chat bot integrated into search spacy.fyi/ltwc Reinvent the wheel, but don’t reinvent the road!

Slide 19

Slide 19 text

Big Tech is reinventing the wheel and the road Accelerated Mobile Pages open-source framework for faster mobile browsing, encouraged for best search results Google most-used search engine with market share of over 80% Chrome / Chromium most popular browser and engine powering Microsoft Edge, Opera, Arc and more Gemini (formerly Bard) generative AI model and chat bot integrated into search spacy.fyi/ltwc Reinvent the wheel, but don’t reinvent the road!

Slide 20

Slide 20 text

Big Tech is reinventing the wheel and the road Accelerated Mobile Pages open-source framework for faster mobile browsing, encouraged for best search results Google most-used search engine with market share of over 80% Chrome / Chromium most popular browser and engine powering Microsoft Edge, Opera, Arc and more Gemini (formerly Bard) generative AI model and chat bot integrated into search Google Ads search advertising generating $200b+ revenue per year spacy.fyi/ltwc Reinvent the wheel, but don’t reinvent the road!

Slide 21

Slide 21 text

No content

Slide 22

Slide 22 text

Will Big Tech monopolize AI?

Slide 23

Slide 23 text

Will Big Tech monopolize AI? What about open- source software?

Slide 24

Slide 24 text

Are we still going to need developers? Will Big Tech monopolize AI? What about open- source software?

Slide 25

Slide 25 text

Are we still going to need developers? Will Big Tech monopolize AI? What about open- source software? Will AI developers be replaced by AI?

Slide 26

Slide 26 text

Are we still going to need developers? Will we become dependent on API providers? Will Big Tech monopolize AI? What about open- source software? Will AI developers be replaced by AI?

Slide 27

Slide 27 text

Are we still going to need developers? Will we become dependent on API providers? Will Big Tech monopolize AI? How do we keep our data private? What about open- source software? Will AI developers be replaced by AI?

Slide 28

Slide 28 text

Are we still going to need developers? What can we learn from the past? Will we become dependent on API providers? Will Big Tech monopolize AI? How do we keep our data private? What about open- source software? Will AI developers be replaced by AI?

Slide 29

Slide 29 text

Don’t confuse raising the floor with raising 
 the ceiling.

Slide 30

Slide 30 text

ceiling floor

Slide 31

Slide 31 text

ceiling floor local store sets up website without requiring help from a web developer 🛍 +10% customers and revenue

Slide 32

Slide 32 text

ceiling floor local store sets up website without requiring help from a web developer 🛍 +10% customers and revenue todo list app adds automated translation feature using an API 🤖 +5% international user growth

Slide 33

Slide 33 text

ceiling floor streaming service makes web player 1ms faster 📺 +3% time spent in app local store sets up website without requiring help from a web developer 🛍 +10% customers and revenue todo list app adds automated translation feature using an API 🤖 +5% international user growth

Slide 34

Slide 34 text

ceiling floor hotel booking platform improves recommendation system 💸 +0.5% click-through rate, +$1m revenue streaming service makes web player 1ms faster 📺 +3% time spent in app local store sets up website without requiring help from a web developer 🛍 +10% customers and revenue todo list app adds automated translation feature using an API 🤖 +5% international user growth

Slide 35

Slide 35 text

ceiling floor hotel booking platform improves recommendation system 💸 +0.5% click-through rate, +$1m revenue streaming service makes web player 1ms faster 📺 +3% time spent in app local store sets up website without requiring help from a web developer 🛍 +10% customers and revenue todo list app adds automated translation feature using an API 🤖 +5% international user growth high adoption

Slide 36

Slide 36 text

ceiling floor hotel booking platform improves recommendation system 💸 +0.5% click-through rate, +$1m revenue streaming service makes web player 1ms faster 📺 +3% time spent in app local store sets up website without requiring help from a web developer 🛍 +10% customers and revenue todo list app adds automated translation feature using an API 🤖 +5% international user growth high value high adoption

Slide 37

Slide 37 text

ceiling floor

Slide 38

Slide 38 text

ceiling floor website builders

Slide 39

Slide 39 text

ceiling floor website builders frameworks

Slide 40

Slide 40 text

ceiling floor website builders browsers frameworks

Slide 41

Slide 41 text

ceiling floor website builders browsers web standards frameworks

Slide 42

Slide 42 text

ceiling floor website builders browsers web standards AI features frameworks

Slide 43

Slide 43 text

ceiling floor website builders browsers web standards AI features frameworks frameworks

Slide 44

Slide 44 text

ceiling floor website builders browsers web standards AI features frameworks frameworks environment

Slide 45

Slide 45 text

ceiling floor website builders browsers web standards AI features frameworks frameworks environment techniques

Slide 46

Slide 46 text

High-value use cases are worth development effort.

Slide 47

Slide 47 text

real-time commodities trading insights by extracting structured attributes explosion.ai/blog/sp-global-commodities S&P Global CASE STUDY 99% F-SCORE 6mb MODEL SIZE 16k+ WORDS / S

Slide 48

Slide 48 text

real-time commodities trading insights by extracting structured attributes high-security environment explosion.ai/blog/sp-global-commodities S&P Global CASE STUDY 99% F-SCORE 6mb MODEL SIZE 16k+ WORDS / S

Slide 49

Slide 49 text

real-time commodities trading insights by extracting structured attributes high-security environment used LLM during annotation explosion.ai/blog/sp-global-commodities S&P Global CASE STUDY 99% F-SCORE 6mb MODEL SIZE 16k+ WORDS / S

Slide 50

Slide 50 text

real-time commodities trading insights by extracting structured attributes high-security environment used LLM during annotation 10× data development speedup with humans and model in the loop explosion.ai/blog/sp-global-commodities S&P Global CASE STUDY 99% F-SCORE 6mb MODEL SIZE 16k+ WORDS / S

Slide 51

Slide 51 text

real-time commodities trading insights by extracting structured attributes high-security environment used LLM during annotation 10× data development speedup with humans and model in the loop 8+ market pipelines in production explosion.ai/blog/sp-global-commodities S&P Global CASE STUDY 99% F-SCORE 6mb MODEL SIZE 16k+ WORDS / S

Slide 52

Slide 52 text

real-time commodities trading insights by extracting structured attributes high-security environment used LLM during annotation 10× data development speedup with humans and model in the loop 8+ market pipelines in production explosion.ai/blog/sp-global-commodities S&P Global CASE STUDY 99% F-SCORE 6mb MODEL SIZE 16k+ WORDS / S

Slide 53

Slide 53 text

ceiling floor 🔮 Human-in-the-loop distillation explosion.ai/blog/human-in-the-loop-distillation 🔮 LLM

Slide 54

Slide 54 text

ceiling floor 🔮 Human-in-the-loop distillation explosion.ai/blog/human-in-the-loop-distillation continuous evaluation baseline 🔮 LLM

Slide 55

Slide 55 text

ceiling floor 🔮 Human-in-the-loop distillation explosion.ai/blog/human-in-the-loop-distillation continuous evaluation baseline 🔮 LLM prompting

Slide 56

Slide 56 text

ceiling floor 🔮 Human-in-the-loop distillation explosion.ai/blog/human-in-the-loop-distillation continuous evaluation baseline 🔮 LLM prompting Prodigy

Slide 57

Slide 57 text

ceiling floor 🔮 Human-in-the-loop distillation explosion.ai/blog/human-in-the-loop-distillation continuous evaluation baseline 🔮 LLM prompting 📦 component transfer learning Prodigy

Slide 58

Slide 58 text

ceiling floor 🔮 Human-in-the-loop distillation explosion.ai/blog/human-in-the-loop-distillation continuous evaluation baseline 🔮 LLM prompting 📦 component transfer learning distilled model Prodigy

Slide 59

Slide 59 text

ceiling floor 🔮 Human-in-the-loop distillation explosion.ai/blog/human-in-the-loop-distillation continuous evaluation baseline 🔮 LLM prompting 📦 component transfer learning distilled model Prodigy deploy 🚀

Slide 60

Slide 60 text

The development process lets you take back control.

Slide 61

Slide 61 text

Development workflows over time

Slide 62

Slide 62 text

static pages Development workflows over time

Slide 63

Slide 63 text

static pages dynamic pages Development workflows over time

Slide 64

Slide 64 text

static pages dynamic pages static pages Development workflows over time

Slide 65

Slide 65 text

static pages dynamic pages static pages Development workflows over time compile static data at build time

Slide 66

Slide 66 text

static pages custom models dynamic pages static pages Development workflows over time compile static data at build time

Slide 67

Slide 67 text

static pages custom models dynamic pages static pages pretrained models Development workflows over time compile static data at build time

Slide 68

Slide 68 text

static pages custom models dynamic pages static pages custom models pretrained models Development workflows over time compile static data at build time

Slide 69

Slide 69 text

static pages custom models dynamic pages static pages custom models distill models into smaller, faster and private components pretrained models Development workflows over time compile static data at build time

Slide 70

Slide 70 text

AI products are more than just a model spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 71

Slide 71 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 72

Slide 72 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT most important di ff erentiation is product, not just technology spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 73

Slide 73 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT UI / UX marketing customization most important di ff erentiation is product, not just technology spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 74

Slide 74 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT UI / UX marketing customization most important di ff erentiation is product, not just technology swappable components based on research, impacts are quantifiable spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 75

Slide 75 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT UI / UX marketing customization speed accuracy latency cost most important di ff erentiation is product, not just technology swappable components based on research, impacts are quantifiable spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 76

Slide 76 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT UI / UX marketing customization speed accuracy latency cost But what about the data? most important di ff erentiation is product, not just technology swappable components based on research, impacts are quantifiable spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 77

Slide 77 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT UI / UX marketing customization speed accuracy latency cost But what about the data? User data is an advantage for product, not the foundation for machine-facing tasks. most important di ff erentiation is product, not just technology swappable components based on research, impacts are quantifiable spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 78

Slide 78 text

AI products are more than just a model machine-facing models GPT-4 human-facing systems ChatGPT UI / UX marketing customization speed accuracy latency cost But what about the data? User data is an advantage for product, not the foundation for machine-facing tasks. You don’t need specific data to gain general knowledge. most important di ff erentiation is product, not just technology swappable components based on research, impacts are quantifiable spacy.fyi/ai-revolution How open-source beats economies of scale, even for LLMs

Slide 79

Slide 79 text

🔮 👩💻 ⚠ 🎁

Slide 80

Slide 80 text

🔮 Developers Don’t confuse raising the floor with raising the ceiling. 👩💻 ⚠ 🎁

Slide 81

Slide 81 text

🔮 Developers Don’t confuse raising the floor with raising the ceiling. 👩💻 ⚠ 🎁 high-value use cases are worth development e ort

Slide 82

Slide 82 text

Models Take back control in the development process. 🔮 Developers Don’t confuse raising the floor with raising the ceiling. 👩💻 ⚠ 🎁 high-value use cases are worth development e ort

Slide 83

Slide 83 text

compile smaller, faster and private models at development time Models Take back control in the development process. 🔮 Developers Don’t confuse raising the floor with raising the ceiling. 👩💻 ⚠ 🎁 high-value use cases are worth development e ort

Slide 84

Slide 84 text

compile smaller, faster and private models at development time Interoperability is the opposite of monopoly. Open Source Models Take back control in the development process. 🔮 Developers Don’t confuse raising the floor with raising the ceiling. 👩💻 ⚠ 🎁 high-value use cases are worth development e ort

Slide 85

Slide 85 text

compile smaller, faster and private models at development time LLMs can be one part of a product or process, and swapped Interoperability is the opposite of monopoly. Open Source Models Take back control in the development process. 🔮 Developers Don’t confuse raising the floor with raising the ceiling. 👩💻 ⚠ 🎁 high-value use cases are worth development e ort

Slide 86

Slide 86 text

compile smaller, faster and private models at development time LLMs can be one part of a product or process, and swapped Interoperability is the opposite of monopoly. Open Source Regulation should focus on products and actions, not components. Regulation Models Take back control in the development process. 🔮 Developers Don’t confuse raising the floor with raising the ceiling. 👩💻 ⚠ 🎁 high-value use cases are worth development e ort

Slide 87

Slide 87 text

Thank you! LinkedIn Explosion spaCy Prodigy Twitter Mastodon Bluesky explosion.ai spacy.io prodigy.ai @_inesmontani @[email protected] @inesmontani.bsky.social