Eric's Render Notes in MX

155 render20220816:

1 11:30 am

241 8/16/22 2 starting at 11:55am - delayed 25 min 2 Howard Rheingold 3 Alice Albrecht 3 Jerry Michalski 3 John Borthwick 3 Linus Lee 2 Davey Morse 3 Gordon Brander 3 Christopher Pedregal 3 Esther Dyson 3 Daniel Doyon 3 Tristan Homsi 2 How do we define tools for thinking and why do they matter? 3 Inflection Points for Tools For Thinking 2 Demo: Protocol Design for tools for thinking 2 The History and Future of Software as Tools For Thinking 3 Tools for Thinking Product Demos 2 Leveraging AI and ML in building new Tools For Thinking 3 Idea Dimensionality and Representing Semantic Meaning 3 Closing 2 betaworkss.fabulous-project.webflow.io 3 Building Bicycles For The Mind 2 going to add all notes under this topic, link later, maybe make a Brain 2 but that would be a lot of manual linking unless I put each not under the respective presentation topic 3 and we have presentations with more than one participant 3 the out.txt report currently does not go more than 1 level down 3 ok the livestream is beginning now 2 the announcer reminded me of the Doors song that has "the movie is about to begin..." 2 inflection points 2 interested in Esther's concept of dimensions and if related to Ted Nelson's idea 3 supercharge this category in the future 2 Plexus - collaborative thought sharing 3 scribes are in the back so people can see them 3 Neel Baronia - Betaworks 2 Backing ideas they find interesting - Think Camp - accelerator 3 betaworks.com/camp 3 follow-on events over next few weeks 2 John and Jerry - new ice cream flavor? 2 John Borthwick, CEO Betaworks, Jerry Michalski 3 interesting start of conference - getting people to talk before you start 2 tools to organize,scaffold,make meaning of info in a functional way 2 information = formation of the mind - we are mass consumers not thinkers 3 organize, construct beliefs, how i think about the world 2 covid forced dev of graph dbs, ml, large tuned datasets, web3 open protocols 3 sharing data across multiple apps - who will win? 2 semantic spatial work - towards AR/VR extension of mobile device 3 how different people think 3 initially thought of them as memorization tools - reverted to paper 2 associative function is more important than memory function 3 let things simmer in your brain - taking breaks after scaffolding 3 better interfaces, tooling, use of tech 2 different tools for different folks (Different Strokes - history of phrase) 3 jazz hands up, down, eh 2 Release 1.0 newsletter - link mgt, mind mapping, then met the Brain 12/97 2 linked each session for today. Like dioramas out of toothpicks. 3 Jerry's brain is art 3 Hypercard could have been the web 3 Echo, bubble charts 2 aggregated memory - adds to categories over time 3 mapping tools for thinking - to cluster of thoughts, color coded 2 Jerry the lone wikipedian - how to create social memory that shares data 3 lessons from my brain 2 lesson:We are an amnesic society - consumerism 3 (check out Jaron Lanier in Jerry's brain) 2 better recall - refreshes paths in head (dynamic ram refresh!) 3 embers of interest - makes links on the fly and can tell story while navigating 3 how to collaborate across siloed datasets? build a bonfire in the middle 2 civizational implications (note Ted Nelson's opinion) 3 shared brain, open content, journalists, policy makers - need the data 3 better decisions as a society - we are in 5 crises right now 2 cinema - recorded theater, pan, jump, montage - new vocabulary. we are stuck in web 1.0 - IP overprotection, need to liberate info between us, instrument it to help us make decisions 2 how we got here - why we're stuck with metaphors (file folders) - Ted hates hierarchy 2 Daniel Doyon, Tristan Homsi, John Borthwick 2 Readwise - 5.5 years ago, from Stripe, Superhuman. Reading & spaced repitition 2 helps you read better, retain more, be a more productive reader/writer 3 use case: highlights, forgetting, never going back. resurface them 2 integration across multiple tools, and out to notetaking: notion, obsidion, roam 3 in middle of consumption and productivity 2 I see connections with Mx if we import data and link to existing base 2 graph db impact, future changes (5 yr) - macro level - evernote focused on writing, then paradigm shift - notion 2015-2019 - block based architecture - infinitely nested + multiplayer, then wave of innovations built upon it 2 Roam - first mainstream graph db, backlinks and transclusion, daily notes page - 2021 - then apps followed them but no big innovation/inflection point since - lot of work in progress 3 Enterprise Use Cases were not addressed 2 ^evernote - also not focused on core application 3 funding models - companies raised lots of money, other models 2 zero barriers to entry - hits industries differently - new models 3 user count is how market evaluates it? 2 first 2 years - weren't sure - started charging monthly fee - saas ramp of death until they could support a team 3 web 3, daos. enterprise saas is proven, else you are consumer 2 people hoping other business models emerge. web 2 was attention and saas 2 a substack link was mentioned that sounded interesting 2 experimenting with writing tools - themed reviews - subset of annotations for review - make a book, essay, paragraphs, highlight - incrementally improve writing - then put in note taking app, sequence it, repeat or fill in gaps - process being developed 2 roles - draft and editor - dont want them interfering too early 2 AI can't be the product 3 AI to enhance/augment products - still need to build the core product 3 blank page is a broken model - our thoughts live in different silos 2 ML for prompting - overprompting - you get fascinated by the machine - not focused on thinking - hard problem 2 spaced repetition has gone mainstream - still early adopter - need to find real world use cases for self improvement - help consumers improve their lives 2 audio - podcast rise, so much - very challenging to move it in - transcribe, workflow 3 useful for capture - spaced repition? 2 books is slow growing, audiobooks grow 15-20% per year, TTS adoption - people love it - helps consume high latency low density info 2 audio to block/highlight - early attempts - ux issues - not seamless - if it continues to improve will be an important role (otter.ai)? 3 ML for TTS, STT - its a bit off if not using latest tools. will be free and everywhere in future. 2 audiobook dominated by one company - closed format, publishing industry, drm 3 innovation on podcast and tts side 2 hey siri take a quote - os level trapping, amazon trapping, silos 3 social highlighting app - ran into silo issues that killed it 2 goldmine of underutilized kindle highlights - then got pulled into tools for thought - people thought they would do everything in one app - but cooling down - people taking portfolio approach to their toolkit 2 twitter discourse - was just interest, now flame wars 2 enterprise setting - change language to single player , multi player 3 need breakthrough - why reading is not just a single player activity 3 main interest is helping people read better and self improvement 2 consumer saas is high - support burden, churn 3 listening to podcasts vs reading - are people reading slower? 2 when listening your attention span is split - lower retention 2 if you listen and read at same time - immersion reading - flow state - even if only 15 to 20 minutes 2 attention span - Walden Ponding not a long term solution - propose fight tech with tech - immersion reading on your phone 3 plumbing between tools - how does that get better? they are hackers 2 open standards would be great - hypothes.is working on open web annotation spec - so far very web focused - can go back to Xanadu but in practice we deal with so many sources - annotations from so many sources - specs not flexible enough. 3 ^key point 2 knowledge blocks- twitter threads have influenced it. It needs to be transportable. 3 "We need the right core abstractions before we start building it" - Tristan 2 turn everything into a block? didn't feel good. Reading in Roam - metaphor - blocks are fluid - now they have a phase change metaphor. Turn fluid thoughts into a coherent text to transfer your ideas to someone else 2 Reading app - take solid and decompose it into knowledge blocks and blobs - solid liquid gas - medium of exchange is an annotation 3 twitter thread saving feature - read when you want to read a 100 tweet thread. Want the blocks compiled into a document that they can annotate. 2 Readwise allows jumping from tool to tool. 3 Customizing the ingestion engine - metadata. Readwise for Slack? 2 Roam users - insisted on formatting features - gave them a templating language. low floor high ceiling approach 3 like a stylesheet? 2 Use cases in education? adaptations, adoption by students? Never had success in academic environment - requires intrinsic motivation. 3 Spaced repitition is for long term retention - market is auto-didacts 2 Subconscious - decntralized protocol for thoughts 2 max 60 subthoughts in mx? 2 bigger than an app - more internet scale. We had it for 40 years. 3 tools needed to cope with info 2 can we think together as a species? 2 apps trap thoughts in SaaS silos - same origin security model baked in assumptions 2 hard to forge connections - needs business deals - no way to scale properly 2 Gordon drew upon Xanadu dream - heavy lift - but tech breakthroughs - it is adjacent possible - Noosphere - worldwide knowledge graph - permissionless - decentralized over IPFS - prioritizes user ownership, multiplayer at a low level, belongs to everyone - like http 2 local first file sync, change history, cross-app transclusions and backlinks, user-owned data backlink, self-sovereign social graph, lots of copies keep stuff safe - metaphor is a data backpack 3 you own what you put into the graph 2 more resiliant foundation - keeps info safe 2 libp2p does heavy lifting - layer on 3 things - human friendly naming, versioning, signing. 2 to link @gordon/paths to memos - deference to content address 3 can take followers and content from app to app 3 memos are a generic data structure. 2 generic protocol can support new structures (VR) - just an envelope with header metadata, pointer to content on IPFS, and ptr to prev version of content - links stay alive, and can link to previous versions 2 security model - crypto - like a movie ticket 2 self-sovereign - wallet of your keys 2 memos get gossipped around by web servers that talk to ipfs - maps handles to your public key, and paths to content addresses - following model 3 syncs with mobile devices - just talk to web server over REST 2 QR code to their discord - open source project 3 Noosphere - Apache/MIT license 3 todo: compare Noosphere to Hypercore protocol 2 (the todo: above may not be found by my higher level todo: list)

482 render20220816: PART 2 OF 2

(Note: The numbers on the left represent the number of links each thought has. Every thought is linked to the date and time, so you can ignore the 2's and 3's).

2 resumes at 2pm EST
3 Howard Rheingold interview
3 peeragogy
3 early 1980s - contract writing at PARC - 1997 article SciAm Alan Kay - microelectronics and the personal computer
3 1982 Time person of the year = PC - seemed it was Steve Jobs and Bill Gates - but they stood on shoulders of many giants
3 PC didn't come from DOD, IBM - individuals who used institutions to create tools to extend their ability to think.
3 book: Tools For Thought
3 book: The Man From The Future - about von Neumann
3 internet not used for nuclear contingency - transformed by the people into communications tech
3 Lichleiter - psycho-acoustics - batch processing - he didn't like the thought workflow - wanted to directly communicate with computer
3 wanted to think more quickly - 1960 article - Man Computer Symbiosis
2 Sputnik 1957 led to Arpa - Info Processing Techniques Office - Lichleiter headed up the office, he knew Doug Engelbart
2 Doug didn't have the right framework for thinking about an integrated toolkit for thought.  AHI paper - he broke down what he meant by augmenting human intellect - how would automating free our brains to think.
3 1968 demo - depended on Lichleiter's funding of Doug's research.
2 Integrated system of humans using language artifacts, methodology, and training - the artifacts had become millions or billions more powerful
2 Doug worked with psychologist - James Fathoman (psychadelic microdosing)
3 Tools to build better tools - orgs could do same thing - teams that race ahead and try new tools
2 What didn't catch on from Doug's demo: A work = service/product, B work = improve ability to create the product, C work = improving ability to improve - required team - skunkworks - try new things then introduced good tech to B work people.
2 Difficult to upgrade the way an org does things - easier to create your own
3 Today - innovation factories - your app can do big things
3 App ecosystem - Doug wanted integrated toolkit (like Smalltalk?)
2 ITE = Integrated Thinking Environment
3 Relationships that need to be made more clear to Tools For Thought community: Knowledge, Thinking, and Learning tools
2 Organizing - way of using visualization of what I know to think
3 Jerry views brain as photoshop for ideas - is this worth remembering?
2 Jerry sees the connections (James Burke influence) - side by side in the same tool.  If it was a DB search but he couldn't annotate or link in to it would not work - he needs the flexibility
2 Brain externalizes Jerry's memory.  PARC had blue journals - ex: advent of GUI - underlined a phrase - putting items on screen creates visual cache for your memory - 7 +/- 2 theory
3 visually designated relationships - we can think at a higher level of abstraction
2 when you see items of knowledge laid out visually and try to use it to show network of connections - most important - central nodes - you look at the framework of thought - when you animate it  you traverse entire trees of relationships - (the tweening is important to our brains)
2 3x5 cards - indexed stacks of notes - proximity was important
2 can speed up and broaden - in ways not possible with physical objects - (zzstructure?)
3 autocad - didn't have expressive capacity for complex things until a new software came out
2 taxonomoy/ontology needed - we need to draw the boundary between notetaking, writing, backlinks, semantic tagging, visualizations
3 we need more conversation about how the tools work - need more interop
2 brain does not do database lookup - manual maintenance of list of representatives
3 Howard's classes explore the tools - what did he learn?
2 People who use certain tools find things to do with them that authors did not have in mind.
3 Packet switching was the tech for bombproof communication
2 Arpanet adopted packet switching.  Researchers wanted to process data on one computer on another computer.  Then researchers communicated about special interests.  Lichleiter and Bob Taylor ran the project - video with John Markoff at UT Austin - mentioned in a newsletter by Rheingold
2 recognized emergence of new comm medium - "Intergalactic Network" - paper: The computer as a communication device - (did this influence Star Trek?)
2 We need developers to get feedback from users - HLAM/T
2 more agency you give to learners (incl responsibility for teaching each other) then role of expert isn't as useful as a group learning together - colearning
3 Jerry mentioned one of Howard's books then shifted to topic of crap detection
2 Howard thinks it's a wicked problem.  Pre-google search engines for HW.  Books had editor, publisher, librarian as gatekeepers who verified authority of the text.  Now you can ask a question and get a million answers but you need to evaluate the credibility.
2 Thought bad info was a problem of literacy.  Microtargeting sophisticated misinformation computationally - it is ahead of education.
3 Medical and political disinformation makes the problem wicked
2 behavioral profiling -> microtargeting
3 social and algorithmic filtering needed - curate list of trusted experts - algorithmic way of using social network to filter materials - yet to be done - assumes you understand and care enough and know enough people - but family bias
3 sift - to check veracity of webpage - get off it - search on name of author - look at reviews
3 huge business in misinforming people
2 The Well - Experts on the Well - brilliant answers
3 We forgot how well we can rely on humans for disambiguation
2 ques about political info - self reinforcing loops for your political view (selective exposure after a purchase?) How can we educate for influencing?
2 Confirmation bias - we're more attuned to evidence that supports what we already believe.  Also emotional attachment to a tribal belief.  Outside of rational fact-based system.  James Fishkin - deliberative polling
2 group of people with different opinions - poll them on beliefs on a topic - experts talk and debate and take questions - audience talks with each other - then poll again - shows audience opinions shift more to the factual
3 human prejudices to learning outside our comfort zone - but we can learn to work around that
3 strong connections
3 individual and group psychology - we're not disembodied brains
2 The Well exists, experts are there.  If I knew answer didn't cost me anything to give it.  If I had a question I'd get answers.  Return of 10x
2 1990s - why would people give answers to strangers?  mark smith - social capital, knowledge capital, and communion *** key point
3 dynamics of social aspect of knowledge sharing - not tools just for individuals
3 you have a community that you turn to
2 Blair Newman scribbled his posts on the Well before suicide.  People reconstructed his posts from their logs.  Jerry found out his mom knew Blair's mom and he printed the article - this showed that something big and important was happening
3 Place for authority figures without going back to gatekeeping?
2 authority of the text - how much can you trust the info? (the bible?) or info from a particular person
3 College prof is the authority on the subject?
2 Social media issues and literacy - learners were more engaged if thrown a question and tried to figure it out for themselves - who do they listen to as authorities?
3 Authority conferred by other people (psych model of this?)
3 questioning authority - much more important now without the traditional gatekeepers - we need more community idea of authority (but what about mass formation psychosis?)
2 Down with primitivism - critique is a letter saying Don't Read This
2 superconducting medium of communication - business model rewards addiction and extremism
2 our brain operates on emotion - how should it play into 2nd brain development? we need to undestand our biases.  The Media Equation book - standard psych tests and substituted computer or cartoon rendition or computer voice - found people attribute human qualities to things computers do (attribution theory) - we haven't evolved to distinguish them!
2 people fall into trap - our weaknesses can lead us to ways to work around them.
3 teaching critical thinking - people will question teachers and parents - people don't like that!  Elementary school - considers it a communist plot
2 (get kindle version of Tools For Thought)
2 Spaciality and Tactility - how important are they?  to what extent will we be held back?  MIT - puts computational ability in physical objects.  Book recommend: The Extended Mind.  Lot of research shows teachers who use gestures - their students learn more effectively (connect to manipulative cards for music education)
2 3 dimensions - realm to explore- (haptics?)
3 computer can slow you down too- ask you to reflect and track it over time
2 when you hear something different and interesting - you have a physical reaction - we should study this more
3 Annie Murphy Paul - people who are aware/not aware of heartbeat.  Aware - made more lucrative financial decisions!
2 Psychoactive experience with pioneers?  Got access to Alto and 5mb disk.  Moving a paragraph without retyping a whole page - it was blissful.  Convinced him this was not just a tool for manipulating info - it was for thinking better.  Frees up space for high level.  Similar to abstraction in computer programming.
2 Late 1980s - teenagers psychadelics - LERYL list for their chats.  Google is building an AI
2 If you can give attributes to physically appearing objects in simulated space you'd have a more sophisticated cache, and ability to manipulate relationships to different concepts.  VR AR XR - feels like it hasn't advanced
2 thebetterverse.org - what would it look like? (my dzez videos)
3 (but jerry doesn't like text in 3d)
2 Product Demos - 3pm
3 Davey - Plexus
2 What it takes for stuff here to have an impact out there?
2 non-coders, non tech people - what apps will look like?
3 2018 Williams College - notes tool like Roam - worked for him but not friends
2 it used special grammar to connect thoughts
3 then he tried using AI to connect thoughts - friends thought it was cooler but didn't see use in their lives
2 problems such as misdiagnosis, not sure of next step
3 connections between people - extended community
2 auto connection between multiple people's thoughts!
3 launching shared brain for knowledge tech community
2 render.plexus.earth
3 get started page
2 Alice - ReCollect
3 AI powered creative assistant - bridge between people and machines honoring the way your mind works
2 cognitive neuroscience, ML
3 augment memory to build extended mind
2 people dont have access to their ideas and thoughts when they are ready to use them - it is an evolutionary problem
3 we don't know where the info is - we can't reason about it effectively
2 Remembering to forget - what will my future self want? we're overwhelmed, anxious about missing info
3 digital breadcrumbs but we need to remember what we recorded and need to tag it - too many breadcrumbs problem
3 piles
2 no explicit tagging or organizing - just bring info in for you
3 info isn't lost but can we find it?  Finding is hard!
3 past self is a jerk in terms of tagging
2 more natural human language - full sentences
3 Play out loud - focus on what you do best - ideas, connections, creation
2 web app, web extension - to get info in when you visit a source often
3 in recollect you ideate and create
3 left side = rich text editor.  right = playground view
2 infinite canvas - cards - white are notes you made.  yellow = recalled
3 backend ML models connect info - similar to how your mind connects it - (training?)
2 to recall info - highlight any sentence and click recall - connections made at the sentence level
2 autoconnect ideas - expanded view - tabs you haven't read or don't remember - peruse results and choose which to keep - there is also contextual info
2 idea board has context- then can play with your yellow cards, and recall more
2 want to get ideas out quickly, reorder thoughts.
3 word and character count for people who want to do quick tweets
3 synthesizes ideas
2 Jerry's demo of The Brain - unique features
2 events - who met, who new, etc
2 actors under their movies, authors above their books - connects books and movies
3 puts director to left using a jump thought
2 director to movie - Don't look up - to climate change
3 insights from using the brain - UI it gets right
2 adding a thought to a news article
2 drag URL to brain, copy past title and author, author connected to NY Times reporters
3 contextual association - self created meaning - (it's an alphabet soup)
2 fast text search is critical
2 show an opinion - links opinions he really cares about - then thesis: SNIP
2 (I wanted a subset of Jerry's brain to explore SNIP)
2 siblings are children of a common parent
2 default color is white. yellow for collections of things. purple for opinions.
3 wants idea connections with other people's brains
2 Sentence Gradients - Linus
3 (I can feel the pressure of the people giving demos)
2 thought experiments - window into interesting kind of future
3 future of how we think and communicate ideas
2 how we work with knowedge and info - record, share, operate, disagree, use for evidence
3 how can language evolve?  how can it improve?
2 spoken natural languages - shortcomings - not tuned for human visual perception system.  not good at looking at a page of text.  written words may look similar but say different things and the converse
3 language is a discontinuous interface - can't operate on the words
2 other kinds of info with better representations - color - but also more quantified representatives (RGB, CYMK)
3 (interesting thoughts)
2 start with text editor. select it, graphic interface on right.  Crop of sentences and meaning close to it.  Anchor points below it.  Unrelated sentences below it.  Interpolate the dot towards the other sentences - changes the sentence below.
3 encoded sentences into numerical continuous representations.
3 lot of structure in the idea space - embedding - from a language model
3 semantic structure - make X axis like a spectrum and place the dot where you think it goes.
2 can help editors revise writing.
3 mechanical aids finely controllable - transformations on thoughts - movement in semantic space
2 far future - notations less about explicit ideas, ex: long division using arabic numerals - the notation encodes ideas - like Algebra
2 !fascinating! - Engelbart quote: tools that actually allow us to find solutions that we didn't think were possible - thoughts we couldn't have thought before
3 Tristan and Daniel
2 (my mind connected those 2 with a pair of video podcasters I watched earlier)
3 Superhuman - email client
3 Superhuman of Instapaper
3 keyboard shortcuts
2 Manifesto - users built complex workflow for reading
3 PDFs - where knowledge goes to die - Jerry
3 your articles - save using browser extension, open in tab
2 cross platform, local first - read offline
3 keyboard based reading - H to highlight, hide sidebars, take notes, marginalia
3 supports ePubs
3 forward emails
2 The Feed - all your feeds - process - read later
3 twitter threads emerged as a new form of blog post
3 highlight across tweets - into your notetaking app
2 customization - views are dynamic folders
3 ex; short reads - to organize, and create your own workflow
3 successive refinement of your query by looking at number of search results
2 save the view - it is available in left sidebar
3 inbox, archive - help power users manage reading workflow
2 full text search - will combine with filtering eventually
3 Do you trust the parsing?  Benchmark of top 200 articles - they are past it
2 can highlight images
2 xpath, highlight tags, sync across webpage - months of work
3 they archive the html locally
2 one app is actually 5 apps in one - private beta ramping up
3 on mobile when you tap a link it saves to your account.  links will parse out forward and back links.
2 Panel discussion: Alice, Linus, Christopher
3 ML trends - large pretrained language modeled
2 2018 - large models, transfer learning - small groups can harness power of big models
3 PhD not required anymore - better tools
2 GPT3 - extremely large scale training - improves models at a predictable rate
3 large models generalize well - pretrained then fine tune it
2 high potential applications of these tech?
2 people focused on prompting currently.  but downside is skill needed and discontinuous performance
3 language understanding at a high level now - natural language possible
2 interface abstractions can be built - files of text can be auto categorized and grouped assuming computer understands your text
3 computers can now work with meaning - language models are one component
2 conceptual understanding still not clear
3 natural language commands - our expectations vs what it does and how it coordinates the task
2 we need more human in the loop and questioning - de-anthropomorphosize it
3 unsolved problems - individual concept formation over time - we can model that and find ways to use combo of models to pull the info out (Hofstadter?)
2 we need other advancements, combination of models put together by humans - it's a component system
2 limitations today: labelled datasets - fine tuning required - deep learning, unsupervised - may change over time
2 lessons learned from deep learning models to apply to unsupervised models
3 generative models - look at results and you ask why - causes friction - will change over time- probing methods
2 hurdle in deeply understanding how the models work, how it was trained, what would happen if you swap in a new model - they are not interoperable yet
3 controllabilty - as a human I would expect this output - if I don't like it how would I steer it? (Cory Doctorow's Walkaway)
2 how to guide model towards important things like avoiding biases
3 companies need to put disclaimers on their products that use models
2 what tech stack? what will change? competition?
3 Alice advised companies in 2018 how to build ML.  Lot of considerations for small companies
2 Can you find an off-the-shelf model to customize, stack, and run on infrastructure you can afford?
2 For every data transformation - design system to be computationally feasible - (several days using GPUs)
3 approvals needed in corps for data use
2 easier to reuse models - its about how you put the things together - parameters - how you break down the problem and find the appropriate models to put together
3 model cannot be your product
2 user interfaces can differentiate your product, not necessary how much compute you use
3 The demo cannot be easily tranferred to someone else.  Can't retrain on the fly
3 What can you do one time?  What can be done realtime?
2 Model size, data size, cost tradeoff
3 Cheaper ways to get specific tasks done - eek performance out of smaller models
2 Try to solve problem without big models - they also are slow
3 Startup - bets on where you put your attention - eat cost now?
2 ex: Lyft was upside down until times changed
3 cheap compute but models growing - (Hamming reference?)
2 models of interaction.  text completion, collaboration with ai, where headed?
2 sometimes best interface is features that are invisible - use fact that models understand language - behind a button - but search results better, also photo editing - search for filters/styles with natural language.
2 lot of products learn from our usage data and change the experience over time - can build invisibility in the interaction
3 autocorrect/spell check - if it gets better by training on your information
3 change text based on what you actually know
2 we never had something speak to us that is not a human
3 need to clarify that you are talking to an agent/bot
3 assistant metaphor, collaborrator metaphor, extended mind - it is part of you
2 it should feel like you
3 if the interface you expose is text, people don't like a lot of text
3 might be better ways to expose info - categories, icons
2 interface design is often the limiting factor
2 humans good at navigating spaces.  make info heavy tasks like search into navigating.  structure projected - people decide which branches to go down (ex: text adventure game)
2 summary of an essay - what if model highlights what it thinks is important - heatmap of relevancy and confidence - helps humans - navigation/filtering tasks
3 new UI paradigms to unleash the power of models - look at group community practice
2 use other systems that are good at pulling the gist of info - once we extract things we lose the spatial navigation - rely on other cognitive mechanisms that are not so slow *** key point
2 models need to decide what is important first - relevancy - extract/summarization - don't need to just rely on machines - human highlighting and annotation to give the machine context for relevancy
2 (think about Nelson/Engelbart ideas for algorithmic paragraph summarization in views)
2 if AI sounds smart we perceive it as an authority.  we learn to use programs by clicking around and observing - problem can be exacerbated - false sense of trust - interface design - when to add grains of salt (salt shaker icon?)
2 AI ethics - if system can change on its own, what's the worst thing that can happen?  impact on people - think about what is the training data?  people building the models - still have a diversity problem - especially needed for base models.  Underlying architecture - how to stave off the worst that can happen (toxic chatbots)
3 capital intensive models used as foundation - creates a scale effect.  any small bias or unexpected influence can filter their way down
2 prompts have been input/output based.  gestures on phone to give more info about our meaning.  how to get there?
3 low effort put into DALL-E - reward/punishment training.
3 passive learning - how long do we wait for it?
2 get more info from the human - biological data - change in intonation - emphasis
2 DALL-E is an experiment, entertaining if weird results.  For more serious models - threshhold is higher.
2 discontinuity in output - if people understood that you could guardrail the model against undesired output - how we understand concepts is not in there.
3 scaling with more data should taper off at some point.  we should be getting smarter about what can be left out.  imprint concepts on the output of the models using knowledge and facts.
2 text is self-similar. AI can color text. Notion/Roam docs look the same at a distance.  Large language model - symbolic representation of a paragraph - to see from a far distance.  Higher level understanding of document structure during training.  Interface design - repeated structure should be refactored out.
3 could you replace a paragraph with a DALL-E glyph side-by-side?
2 how tools expand our intellectual abilities - what tool amplifies is diminished in human.  Mechanical clock - you lose intrinsic sense of time.  Google Maps lose navigate streets or sense of orientation.  How do we know the tools are not making us dumber?  What we outsourced - offload thought, less memorization.  People also rely on each other to offload.  Creativity is implicit in intelligence - bring that out.  We are overloaded now.
3 (these people are thinking on a higher level than I am - mapping the space would be helpful)
3 Esther Dyson - interdimensionality
2 (music - sounds 60s trippy)
2 Esther's reactions
2 investor in abridge.ai - to summarize and bring out main points in dr patient conversations - U. Penn using it - insurance company can lower premiums because record of conversation
3 investor in hypothes.is
2 wellville is non-profit
2 at 8 years old got off school bus and had ocular migrane - lost ability to construct a 3d world - couldn't create the mental model
2 ability of brain to construct dimensions from 2d input
3 could you construct a 4d world plus time?  could we think multidimensionality?
3 quadrant charts - how you can move
3 search about topics/things vs actions and changes
2 language understanding needed for temporal dimension
3 fundamental dimensions? canonical set?  Linus started with it.  Time - dependencies, sequence, references to other objects
2 positional meaning
3 a block is like a PDF - you want to point to things inside the block - structures
2 AI not considering effect of wind on a house - structure as well as spatial representation - architecture of structure - like a sailboat
3 where are people strong or weak in a job - what could you remove without changing something - 3 legged dog
2 nugget size - holons of info - how do we zoom in and out of them or turn them into legos or modularize into an argument (pros, cons) - not formal but no general language for it
2 law - in theory is structured and logical.  but lot of inherent contradictions/loopholes.  language usually obfuscates - spectrum: don't steal --- 89 pages of what you don't understand - be concretely general
3 Lexon - legal language based on solidity - readable language for smart contracts - lexon.tech - ex: lex escrow contract
2 is tech making us dumber?  another angle - as you get older you forget things? (you're not losing your memory - lots of things in there - harder to find) - you pay less attention to things like people's badges vs learning names - we don't garbage collect our brain - we have open loops - get them out of your head into a reliable system that will feed you the next task
2 if you don't look at it or you're not good at finding things in it
2 psychology - parents were divorced- problem if called mom by stepmoms name - so she's risk averse for learning names - people are complicated - she'd use an AR system - name use feels forced to her
2 Esther's trail going forward- we're still talking about one person's brain - David Waltz - words ptrs to shared understanding.  AI don't have it
3 AI did not learn - it was taught.
2 Shared understanding among people working on a task.  Slack is one part.  Multidimensional issue.  Jeff Jonas - Las Vegas - tool for casinos that would lose license if let in a known felon.  DB grew to 600million because of duplicates and multiple addresses.  It got cleaned up.  Design/implementation - every query turned into data added to DB
2 you want dedupe logic in a shared understanding of some problem.
3 how do you merge after a 2 week vacation?  dynamics of info sharing structure
3 fact denial, bad actors - more messy
2 DNA vs Metabolism - the code, your ID.  Who produced this news?  Source?  obfuscated.  Look at it like a PET scan - looks at the metabolism - what in the body is absorbing large amounts of sugar
2 for information - it is the money flow.  misinfocon - this stuff makes a lot of money
2 promotion of false views - not a natural metabolism - it is manipulation - can it be detected as a perturbation in the statistical flow?
3 The metabolic presence of thinking in an organization?  When there is a distorition - someone talks too much or pushing something.
2 People who promote themselves beyond  - unusual behavior - pushing ideas to get promoted - where are the abonormalities (ML outlier detection)
2 Who are the people nobody talks to?  Some are metastatic, some are genuine creativity?
3 but you don't want to be surveilling people.  amazing what you can find by looking at data.  email analysis - categorization
2 relationship of trust and thinking in these systems - creative thinking course - takes a month to develop trust among participants - cognitive load that people will say something people judge them for
2 what happens to doctors when you introduce tools?  has to be trust in the tools, institution that provides them, collects your data.  low diversity in the room
3 vaccine story - trust is a function of individual's past experience
2 AI Neuroscience - in the same folder.  How people's minds work, how AI should work.  Trust is a different dimension - but it is a human one.  How much evidence do you need to trust a source?
2 Howard learned to have the kids design the curriculum.  They present to each other - trusting students to do that.
2 We've been designing institutions out of mistrust.  Flip it around - give people back their autonomy.  Hidden curriculum of school - Gatto - Was teaching kids the wrong messages.
3 He broke all the rules of teaching and that made him win teaching awards.
2 We lost faith in humans - it is baked in now.  It is comfortable when we flip it.
2 theme around language - sanitizing the interpretation to create something meaningful equally across multiple people/experiences -ex: legal contract
2 lawyers get paid to make sure each party is interpreting contract the same way.  if it falls apart, judge interprets the contract.  when you strip interpretation out of the equation you get dogma and inflexible belief systems
3 maybe it is better to lean into interpretations?  Linus model - create different explanations of the same thing and find the one that resonates with them - quantifying intention in the communication
2 dimension where something happens slowly, various atmospheres (my mind heard "slowly changing dimension" from data warehousing)
2 abstraction of "mother"
3 purpose of legal document is to figure out range of possibilities it can apply to, and exceptions
3 A well regulated militia - nobody can agree on what it means
2 fight over the meaning of the language is intensely important
3 trust is in context with power structure and how resiliant the system should be
2 high trust - you have consequences, including social capital
3 how do people compromise in the right way?
2 assuming more trust is better - read the architectures of mistrust - Shared Spaces - those affordance we think raise safety decrease safety
2 redesigned intersections - depends on the domain - architecture is a big deal
3 be trustworthy - be transparent
3 evidence of trustworthiness needed (but what about Whuffie?)
2 self disclosure - the more real you are and share your motivations
3 tell people what you're willing to get in a bargain.
2 the block- atomic unit of thought - it is not sufficient.  Atoms - H2O - you don't know if the molecule is wet - need to look at how the molecules interact
2 what's interesting is what connects the blocks - ice crystals vs smooth flow vs gas distance
3 address book - order is not inherently meaningful.  map has arrangement that adds the meaning and helps you find things.
2 Zeotech - auto transcribes, transfer, segments, zoom calls - does for video what google maps did for an address book
2 homophone problem - their, there - I can't bear this bear of a question barely!
3 autosuggest seems to understand it
2 critical mass, inflection points - they bring like-minded people together to talk about it - future programs - will reach out.  Camp program.
2 Think Camp - accelerator in September.
3 Language is our original tool for thinking
2 Think about it as a starting point of the journey - Jerry will host weekly podcast/zoom with interviews of builders of tools for thinking
3 Expand the big tent
3 Lot needs to be created - we need to map these topics to see them