HomeMy HighlightsDiscover
Sign up
Kazuki

Kazuki

@kazuki

Ask AI Clone

Cofounder of Glasp. I collect ideas and stories worth sharing 📚

San Francisco, CA

Joined Oct 9, 2020

1072

Following

5832

Followers

1.47k

pages

13.59k

highlights

171.84k

views
M
W
F
Mar
Apr
May
All/Weekly/Yearly
Total Days:
Total Weeks:
823-Days 📚
154-Weeks 📚
Tags
Domains
AI 159
Learning 93
Founder 84
Knowledge 84
Growth 82
Startup 74
Product Development 71
Founding Story 71
Reading 67
Psychology 58
Human Behavior 58
Self-improvement 55
Life Lessons 54
Glasp 46
Note-taking 46
Mindset 44
Startup Idea 38
PMF 35
Knowledge Management 34
LLMs 32
Economics 32
Consumer App 31
Mission 31
Purpose 30
Community 28
Writing 28
Search Engine 27
Metrics 27
Curator Economy 26
Highlight 25
Thinking 25
Advice 24
Curation 24
VC 23
Productivity 21
Culture 21
Creator Economy 20
Crypto 20
Marketing 20
Motivation 20
Tools for Thought 19
Creativity 18
Business Model 18
Web3.0 17
Decision-Making 17
PM 17
YC 16
Fundraise 16
Social Token 15
Education 15
UX 15
PG 15
Investment 14
Quotes 14
Design 13
Social 13
Habit 12
Innovation 12
Leadership 12
Philosophy 11
Twitter 11
Branding 11
NFT 11
Memory 11
Future of Work 11
Network Effect 11
Equity 10
Finance 10
Strategy 9
ML 9
Social Media 9
User Acquisition 9
Career 9
a16z 9
Cognitive Science 8
Investor 8
Co-founder 8
Retention 8
Media 8
SEO 7
TED Talk 7
Collect 7
Blockchain 7
DAO 7
Gamification 7
Annotation 7
CAC 7
Publication 7
Neuroscience 6
Content Marketing 6
Moat 6
Chrome Extension 6
Viral 6
500ish 5
Accelerator 5
Human History 5
Copyright 5
Social Annotation 5
Ideation 5
Marginalia 5
HR 5
Sociology 5
Storytelling 5
Pitch 4
Mental Model 4
Gen Z 4
Consumer Trends 4
Edtech 4
Steve Jobs 4
Personal Development 4
Burnout 4
Communication 4
AI Agents 4
Product Hunt 4
Legal 4
Agency 4
Tokenomics 4
Marketplace 4
Technology 3
Vector Database 3
Workflow 3
Health 3
Book 3
Hiring 3
Organization 3
Ethereum 3
GTM 3
Market Research 3
Engagement 3
Algorithm 3
Advisor 2
Autonomous Vehicles 2
Problem Solving 2
Wellness 2
Remote 2
Collaboration 2
Core Action 2
Intelligence 2
Meditation 2
Elon Musk 2
Onboarding 2
AGI 2
Defensibility 2
Pre-seed 2
Corporate Governance 2
Analytics 2
Kudos 2
Personalization 2
Life 2
Luck 2
Newsletter 2
Content Discovery 2
Audio 2
Market Size 2
Pitch Deck 2
Relationship 2
Growth Hacking 2
Cold Email 2
BtoB 2
Email Marketing 2
Coding 1
Warren Buffett 1
Literacy 1
Debate 1
Cognitive Bias 1
Vision 1
Push Notification 1
Public Policy 1
Negotiation 1
Hearing 1
M&A 1
Visa 1
CTO 1
Teaching 1
HCI 1
Discussion 1
Prompt 1
Pivot 1
Information 1
Curiosity 1
Longevity 1
Salary 1
Time Management 1
AI Safety 1
Security 1
Ecosystem 1
Nature 1
Digital Garden 1
IP 1
International Growth 1
Typography 1
Biography 1
Commerce 1
Child Mortality 1
Reading List 1
Language Learning 1
Design Thinking 1
Glasp Talk 1
Interpretability 1
Quantum Computing 1
Child Development 1
Freemium 1
Swift 1
Meme 1
Cultural Insights 1
Text Embeddings 1
History 1
Word of Mouth 1
Insurance 1
Governance 1
BtoC 1
Independence 1
Climate Change 1
Naval 1
Digital Legacy 1
Show All

Atomic Graph

Highlights
Favorite
Kindle
Video
Bookmarks
Hatches
Posts

The first secret of great design | Tony Fadell - YouTube

www.youtube.com/watch?v=9uOMectkCCs

Design
UX

Nov 10, 2022

111

A trillion dollar opportunity

talia.gold/2022/10/31/a-trillion-dollar-opportunity/

Search Engine
Knowledge
Social

Nov 9, 2022

85

Easy Come, Easy Go: Understanding the Universal Law of Money & Success

hardfork.substack.com/p/easy-come-easy-go-understanding-the

Life Lessons
Founder

Nov 9, 2022

61

What to Watch in AI | The Generalist

www.generalist.com/briefing/what-to-watch-in-ai

AI

Nov 9, 2022

51

Great advice on writing from Tim Urban

brianbeckcom.medium.com/great-advice-on-writing-from-tim-urban-e601053173cd

Writing

Nov 4, 2022

61

Sharing Notes

learningaloud.com/blog/2022/10/30/sharing-notes/

Knowledge Management
Glasp
Social Annotation

Nov 4, 2022

51

TikTok’s Future: Learnings From China

a16z.com/2022/09/21/what-china-can-teach-us-about-the-future-of-tiktok-and-video-search/

Consumer Trends
Social
Search Engine

Nov 4, 2022

7

How to Raise Money

paulgraham.com/fr.html

Fundraise
Startup

Nov 2, 2022

5417

Limiting Beliefs: Invert, Invert, Invert to Get to Bigger & Better Life

hardfork.substack.com/p/limiting-beliefs-invert-invert-invert

Self-improvement
Life Lessons

Nov 2, 2022

42

AI-powered Psychoanalysis: Journaling in Mem with author Tim Klein

newsletter.mem.ai/p/ai-powered-psychoanalysis-journaling

Purpose
Life Lessons
Decision-Making

Oct 29, 2022

153

Solitude and Leadership - The American Scholar

theamericanscholar.org/solitude-and-leadership/

Leadership
Thinking
Mindset

Oct 27, 2022

224

The Death of Intellectual Curiosity | Sven Schnieders

svenschnieders.github.io/curiosity/

Learning
Self-improvement
Education

Oct 27, 2022

155

What is a good activation rate

www.lennysnewsletter.com/p/what-is-a-good-activation-rate

Metrics
Product Development

Oct 26, 2022

71

Humanloop partners with Stability AI to build the first open-source InstructGPT

humanloop.com/blog/stability-ai-partnership

AI

Oct 25, 2022

4

How Lex Happened

every.to/divinations/how-lex-happened

Life Lessons
Startup Idea

Oct 22, 2022

4

Build a website in 30 seconds, with AI | Ben's Bites

bensbites.beehiiv.com/p/build-website-30-seconds-ai

AI
Glasp

Oct 21, 2022

11

AI: Startup Vs Incumbent Value

blog.eladgil.com/2022/10/ai-startup-vs-incumbent-value.html

AI
Startup
Economics

Oct 21, 2022

185

6 Powerful Note-Taking Tools to Activate Your Mind & Connect Ideas

medium.com/swlh/6-powerful-note-taking-tools-to-activate-your-mind-connect-ideas-548214069c5b

Note-taking
Glasp
Learning

Oct 20, 2022

94

Spending on Artificial Intelligence Solutions Will Double in the United States by 2025, According to a New IDC Spending Guide

www.idc.com/getdoc.jsp?containerId=prUS48958822

AI

Oct 18, 2022

3

5 Productivity Apps Hyped Up Right Now

medium.com/keep-productive/5-productivity-apps-hyped-up-right-now-44610dcc788a

Glasp
Productivity

Oct 18, 2022

11

Aligning Language Models to Follow Instructions

openai.com/blog/instruction-following/

AI
ML

Oct 18, 2022

101

Examining Emergent Abilities in Large Language Models

hai.stanford.edu/news/examining-emergent-abilities-large-language-models

AI

Oct 16, 2022

4

DAOs and The Iron Law of Oligarchy

eriktorenberg.substack.com/p/daos-and-the-iron-law-of-oligarchy

Web3.0
DAO
Governance

Oct 16, 2022

153

Designing a better thinking-writing medium | thesephist.com

thesephist.com/posts/medium/

Thinking
Tools for Thought
Writing

Oct 13, 2022

101

How We Misunderstand Anxiety and Miss Out on Its Benefits - Neuroscience News

neurosciencenews.com/anxiety-dopamine-21390/

Psychology
Mindset
Neuroscience

Oct 12, 2022

121

Glasp vs. Matter: Which Highlighter App Should You Use?

www.hongkiat.com/blog/glasp-vs-matter/

Highlight
Reading
Glasp

Oct 12, 2022

81

About

podcast.ai/about

AI
Creativity

Oct 12, 2022

3

🚀 Is AI art ethical?

www.readaccelerated.com/p/is-ai-art-ethical

AI
Creativity

Oct 10, 2022

72

Making “Freemium” Work

hbr.org/2014/05/making-freemium-work

Business Model
Freemium

Oct 8, 2022

8

How to Model Viral Growth: The Hybrid Model | LinkedIn

www.linkedin.com/pulse/20121002124206-18876785-how-to-model-viral-growth-the-hybrid-model/

Growth
User Acquisition

Oct 8, 2022

4

Funds that lead pre-seeds

outsetcapital.com/writing/posts/lead-preseeds

VC
Pre-seed
Fundraise

Oct 7, 2022

2

The TikTokization of Everything

digitalnative.substack.com/p/the-tiktokization-of-everything

AI
Consumer Trends
Startup Idea

Oct 6, 2022

5

You and Your Research

www.cs.virginia.edu/~robins/YouAndYourResearch.html

Life Lessons
Philosophy
Thinking

Oct 6, 2022

368

The idea maze

cdixon.org/2013/08/04/the-idea-maze

Founder
Advice
Startup Idea
Mindset

Oct 4, 2022

84

lecture5-market-wireframing-design.pdf

spark-public.s3.amazonaws.com/startup/lecture_slides/lecture5-market-wireframing-design.pdf

Startup
Startup Idea
Founder

Oct 4, 2022

172

Generative AI: A Creative New World

www.sequoiacap.com/article/generative-ai-a-creative-new-world/

AI

Sep 30, 2022

11

Introductions and the “forward intro email”

also.roybahat.com/introductions-and-the-forward-intro-email-14e2827716a1

Investment
Fundraise

Sep 29, 2022

9

The Infinite Article

every.to/divinations/the-infinite-article

AI

Sep 29, 2022

93

Eigenquestions for the AI Red Wedding

lspace.swyx.io/p/eigenquestions-for-the-ai-red-wedding

AI

Sep 28, 2022

61

67

Aligning Language Models to Follow Instructions

URL
https://openai.com/blog/instruction-following/
Tag
AI
ML

Highlights & Notes

GPT-3 is trained to predict the next word on a large dataset of Internet text, rather than to safely perform the language task that the user wants. In other words, these models aren’t aligned with their users.

To make our models safer, more helpful, and more aligned, we use an existing technique called reinforcement learning from human feedback (RLHF).

The resulting InstructGPT models are much better at following instructions than GPT-3. They also make up facts less often, and show small decreases in toxic output generation.

Our labelers prefer outputs from our 1.3B InstructGPT model over outputs from a 175B GPT-3 model, despite having more than 100x fewer parameters.

Some of our previous research in this direction found that we can reduce harmful outputs by fine-tuning on a small curated dataset of human demonstrations.

User
Interesting. By using curated information on Glasp, we could provide better outputs.

We find that InstructGPT models are significantly preferred on prompts submitted to both the InstructGPT and GPT-3 models on the API.

We also conduct human evaluations on our API prompt distribution, and find that InstructGPT makes up facts (“hallucinates”) less often, and generates more appropriate outputs.

Despite making significant progress, our InstructGPT models are far from fully aligned or fully safe; they still generate toxic or biased outputs, make up facts, and generate sexual and violent content without explicit prompting.

they may become more susceptible to misuse if instructed to produce unsafe outputs. Solving this requires our models to refuse certain instructions; doing this reliably is an important open research problem

Right now, InstructGPT is trained to follow instructions in English; thus, it is biased towards the cultural values of English-speaking people. We are conducting research into understanding the differences and disagreements between labelers’ preferences so we can condition our models on the values of more specific populations.