#🧵Opine#💬Communication#🎨Workshop#Twitter#📚Series

🌲Opine is my name for a theoretical twitter-like social media product that tries to innovate on the medium with more humane principles, primarily using AI. This is an experimental space where I can explore these ideas, some of which may be contradictory.

Key Links

Opine Design Opine Research Questions Opine Workshop Opine Principles Opine Governance AI Content Moderation Report Card Reaction System ⭐️ Desired Emotion Activity Review

Evergreen Notes

Worldwide commons

We must all learn how to participate in a worldwide commons A worldwide commons requires etiquette baked into the medium to function properly

Users

Unreasonable people have an outsized impact on discourse online Normal people don’t post online Interactions with people outside our own bubble may increase polarization Cohorting users by development stage may improve social media The greater the differences between people, the more constraints are required for successful interactions Reflecting our recent activity back to us can build self-awareness

Product

Social cohesion may not be good to design for New social media ideas are difficult to test Social media provides emotional roads without traffic signs The Reply Box is one of social media’s most enduring emotional traps

Goals

Build off-ramps and circuit breakers Build off-ramps for bad emotions Favor quality of engagement over quantity Favor reduced negative emotion over increased positive emotion Focus on good faith usage Assume everyone on social media is flying on instruments

Insights

Social media’s weaknesses can become strengths The solution to bad speech is probably not just more speech Lies spread faster than the truth

Market

Platforms want to connect us to people we don’t know because they control those connections

Stream

When Product Markets Become Collective Traps: The Case of Social Media (2023, Becker Friedman Institute via JonHaidt)

• Large shares of consumers use Instagram and TikTok out of a fear of missing out rather than genuine interest and, as a result, are worse off than if the platforms did not exist in the first place.
• Seems like Race to the Bottom / Moloch

2023-12-29 Clarification Scaffolds

Goal: Improve reply ranking accuracy & visibility Object: a reply, which needs to be classified for usefulness to better rank it Who: a feature for the first people to read the reply Method:

1. Triggers when AI has idea of how to classify a reply (e.g. off-topic), but isn’t sure.
2. For first few people to read reply, AI adds Clarification Scaffold below the reply, asking clarifying question (e.g. “is this off-topic?“)
3. If sufficient number of first viewers say ‘yes’, post is considered off-topic, and whatever ranking of usefulness that is gets attached to that.
4. Possible: replyer is notified their reply is considered off-topic so they know for the future.

Inspiration: This HN reply (2023) essentially saying “I didn’t see this post about X (testing performance of computers) talk about my pet issue Y (building on network vs on computer)“. Clearly off-topic.

2023-12-29 Qualitative Performance

Goal: provide better feedback to users on their posts, …

1. without psychological downsides of displaying feedback numerically, such as compulsive re-engagement and metrics chasing
2. without needing to see duplicative content
3. normalizing according to relative popularity, taking into consideration who liked it and their Importance in this context
4. minimize unpopularity blindspot of assuming many positive responses mean poster’s viewpoint is broadly supported (e.g. 10k likes on a post that 3mm people see, of whom 2.9mm disapprove but have little mechanism to indicate without engaging, aka ‘everyone likes my awful point because nobody is engaging in debating the other side’) Who: for people who have lots of followers Object: the feedback attached to user’s post Method:
5. AI collects up/downvotes and contents of replies
6. AI summarizes the points that people respond with
7. AI attempts to summarize why the post performed well or poorly (see ideas in Clarification Scaffolds)
8. (extra, if it also fully filters replies) Highlight specific replies that may be worth engaging in, either due to interesting questions, friends responses, notable people, etc.

Example:

• A few people agree with you on this, while quite a number disagree
• Your post gained widespread agreement among people who interact on these issues, along with interest.

Inspiration: This is an alternative solution to the above Clarification Scaffolds, to just use more basic UI of up/downvotes as hints that users give to the AI, which the AI uses to inform the poster/replyer how their message is landing. This would involve AI analysis. In this case, the AI might tell the replyer “your post doesn’t seem to be received well. My best guess is that what you brought up is not the primary topic of the conversation, which is the performance of a computer.” This would require cultural shift of meaning of up/down votes to be hinting.

2023-12-29 Importance in this context

Some way of rating users according to how important their views are in a given post’s context. E.g. if popular responses come from people who are outsiders, that should be weighted less than if they’re experts on the post’s context.

2023-12-24 Familiarity of interfaces, community v scene

From visakanv (2023) a scene is not a community. oftentimes this distinction doesn’t really matter but sometimes it really does… expecting a scene to behave like a community, treat you like a community, etc is a recipe for disappointment. scenes are closer to geographic (or psychographic) regions, communities are closer to families. they can overlap but they are not the same thing… vagueness in categorizing is useful when it comes to experimentation, but it’ll bite you in the ass when you expect, say, kinship from strangers. which isn’t to say strangers can’t be kind. strangers saved my life. but navigating this requires a lot of finesse that ppl dont have

The Twitter reply box is designed for close familiarity, such as that of families and small communities. It doesn’t work for a scene. So what is the appropriate reply box for a scene?

Related: The Reply Box is one of social media’s most enduring emotional traps Reading: How to build and maintain communities (2016, visakanv)

2023-12-23 Correct Priors / Reality Alignment

• Currently on social media we are fed a stream of under-scrutinized claims and misconceptions that have out-competed others.
  • We are slaves to them, not the other way around (📕 The Beginning of Infinity)
• It is only by chance that initial misconceptions of ours are corrected.
• We need something that makes this happen on purpose. It should function as follows.
• App like rewind.ai tracks any news information you see, specifically information that’s likely to be something that forms a belief.
  • This tracking doesn’t even need to be a heavy lift on each edge device, as people are often seeing the same news items.
• App checks on what the truth is of those on occasion.
  • This oracle is a separate system, perhaps a
• If corrections are in order, app surfaces those.
• App shows you provenance of that information, and offers you the ability to filter that content in the future.
  • App may also automatically do so, and remember that this source has provided poor information in the future (Reputation Systems)

2023-12-14

New York to Target Social Media Firms That Charge for Interface via triketora - 🤖 bill summary

2023-12-14 Whats the point of microblogging

• Used to be breaking news
• But as journalism diminishes, its also just how ppl get info
• Lots of people doing things and wanting to talk about it
• Those people prefer informal expression
• Consumers prefer short dopamine hits
• Trailhead and dive deep is a good model
• Variance is a bad model, except when surfacing unusual connections between disparate subjects
  • We’re moving from a wire-down to a wire-across model; more than just depth v breadth, or top-down v bottom-up, this model is about how we construct understanding. Do we follow a single thread for a long time, or do we sample a broader space and look for the connections? Touching elephant in different ways.
  • Unusual connections could be a Building GPTs idea
  • Wire-across may be riskier from the perspective of Connection Sickness

2023-12-14 Adversarial Interoperability

How one could build Opine features without building a whole social network:

• Build a chromium app with puppeteer to scrape twitter etc in headless mode, which only hits twitter URLs (not a browser). Scrape both via DOM and if that’s shut down, via GPTvision scraping.
• Do things like summarize the day’s news, reformat tweet layouts, add extra twitter features and context, etc.
• Aggregate Bluesky, Mastodon, Threads, etc, so they’re all under one roof.
• Analyze user social networks and cache data on posts/accounts to reduce summarization costs and latency
• Notifications don’t matter as much because it’s about having a healthy relationship with it and coming to it when you need it. System would be aware you entered a conversation with someone, so if you had a lot of open threads it’d periodically send notifications to see if you wanted to check on them, probably with a trailing off based on when you interacted.
• Call it ‘Contra’ short for contraception and contraband and whatever else

2023-12-13 Expanding the timescale of social media

• Lots of too-good-to-be-true fabulists on social media that we’re exposed to (example). This is the downside of instant access to information - its mostly wrong.
• What if we had a ‘check back on this later’ ability, so that if it turned out it was false, we’d find out? This might correct our takeaways.
• This could also summarize things you checked back later versus ones you haven’t, to show you your ability to spot bullshit.
• These would appear in your feed.

2023-12-10

Defining a social medium through the conceptual, the spatial, and the temporal.

• Temporal - Reply Timers
• Spatial - Microblogging vs longer, rooms
• Conceptual - idea

2023-12-08

Starting to prep for Greg’s Building GPTs

Toxic comments are associated with reduced volunteer activity on Wikipedia via HN (2023) - they rate toxicity

2023-12-07

Formal Accusation - Use LLM to detect when user is making an accusation that someone has done something bad. This system would ask the user to confirm what they were accusing the person of, and that it would go on the user’s record they were making an accusation and others could browse all their accusations. Scoping this more it could ask what they felt the appropriate punishment was as well. The idea here being to make accusations more formal, in order to match the seriousness of the action to the awareness and friction in the UI. Overall it would be to discourage their use except in infrequent cases.

2023-11-27

Contrasting Opinion - when showing a post with a particular opinion, also provide users with past posts that have a different opinion. Goal: keep reality in balance, see Connection Sickness

2023-11-25

I need to get more systematic about Opine feature ideas if I want to publish stuff:

• Standardize how features are presented
  • What principles they match
  • Evergreen notes related to them (above)
  • Where they fit in Message Flows
  • What other features they rely on, precede, or follow, if any
  • Research questions and goals
  • What additional definition they need
• Look at how Product Wiki did it a bit
• Build airtable of all the possible interventions and when they could come up and what problem(s) they solve for what users
• Add#🧵Opine to whatever features I want to publish
• Get domain (pos: opine.social, opine.fun, opine.xyz, opine.ai)

Imported most of My Twitter Product Vision (2022)

2023-11-24

Conversation Rooms - when replying, OP can initiate a reply room to respond to a reply of their message. This room is default private, which allows messages of greater meaning to be shared. If both participants agree at the end the reply room can be made public. OP may invite other users to the breakout, and other users may request to be let in. Goal: improved meaning.

See: Figma Twitter New Reply Composer (with Reply Privately) Private Replies Breakout Replies / rooms

2023-11-17

What if I had a ‘social media builder’ site that let users turn on/off various features in the Opine Message Flow. They’d then get to click around and see what the experience was like in various states.

2023-11-13

📕 Whole Brain Living - When we feel fear the learning & memory circuitry in our brain turn off.

Better said / related pines

• When a pine has been stated better by another, or stated slightly differently
• Rather than the up/down of context, this relies on a left/right of viewing similar pines
• Uses AI to judge similarity, as it is basically a Compression task across pines
• Basically pre-search but better

2023-11-12

Various ideas which relate a bit to Reply Timers and Slowing Down Social Media (Draft):

Reaction System - selecting a reaction instead of just liking a pine. This is a prerequisite for a number of features.

Time-on-read

• Decrease reply timer if user has spent time reading the replies to the tweet, versus jumping right in

Distance sizing

• Socially distant people have smaller messages (font size) which require holding on to ‘get closer to’

Second chance interstitials

• Optional choice to select imagery, messages, and capabilities that may give you a second chance work best to un-trigger you.
  • Think of your kids, reputation, workplace rules (‘will this pine get me fired?‘)
  • Delay sending for 10 minutes and if I still want to send it I can press ‘send’

Co-browsing

• Schedule times to browse the same timeline together. Be each other’s emotional regulators.
• What if each person traded off taking browsing actions. Scroll for 10 seconds and either tap a tweet or the scrolling is given over to the friend. Then if the friend saw a pine they liked in the prior 10 second window, they can scroll back and tap it. Else they can scroll for 10 seconds and the hand-off happens again.
• Reading a pine or an article can be done on each person’s separate time (if they read at different speeds), or be done jointly in hand-off form
• All the while they people can talk to each other
• Feed is selected to be something that overlaps both people’s interests by default

Similar Replies - lets say 3 people say something similar. If someone is replying with another one, in the Response Submission step they get the similar reply prompt showing the similar reply that got the most likes.

See my 3 step check and balance between elites and proles that i drew in notability i think; judiciary and people fuck what was it

They could come in and write a specific type of community note that explains why it isn’t a strawman.

2023-11-08

Article summaries - before replying on an article, the article summary is broken down into a set of claims, and user has to click through and acknowledge each claim. They may also add recations to the claims themselves, like 🤔.

2023-11-06

Reputation in different groups could be reviewed by peers but without seeing who it was, only seeing past contributions to community. That way it’s judging solely by contribution not identity.

🤖 Simulation Testing - thinking about how to think about a system like this to identify a stable community. I think it is to look at the system sans user inputted language. Like what if the system only had an AI write responses between people, and those messages were random, so it could be horrible things or nonsensical or emotional. Wait, maybe this is actually a simulation of AI agents, who are impacted by different messages, and must learn to use the non-language features (like mea culpa, regret, etc) to bring the system back in balance.

Regret - if you got into an argument w someone you can come to it later and explicitly indicate that you regret saying how that turned out. Basically a low friction path to mending fences vs lingering. “I acknowledge your regret” might be a method of response (like forgiven was for mistake).

Reply Timer (assuming I change that to Slow Lanes) - reply timers are a way to put a pine in a queue for replying to later. They let us catch our breath, consider what we’re going to say, and engage our type 2 slow thinking system.

2023-11-03

Frames - a Context Mode kind of feature that suggests a user may have a very different value system and their tweet is the product of that value system.

Bad at everything - cat pictures, stuff that makes us think, dispassionate news, MTG said what on the house floor, some very different person’s opinion… these are all very different things.

A universe of options - the simplicity of the short message suggests a simplicity of it’s container, the social network itself. But this doesn’t have to be the case! There’s a universe of options, of possible containers that create, shape, contextualize, and spread these simple messages. The only one anyone has ever really been focused on is the feed algorithm. And while that’s important, boy is that just scratching the surface.

2023-11-01

Intentional Activity - a high-willpower feature, intentions are essentially higher-level containers or frames that we can reason about for any type of activity we want to engage in on the app. Intentions let us be intentional about the specific way we want to experience the app at a given moment. Each intention functions a little bit differently in terms of both feed algorithm and rules for interaction. These include:

• 🔵 connect with my community - identifies people by the community they’re in, shows you the community connection of people in the replies, lets you see how you’ve interacted with those people in the past (Shared History).
  • Interest filtering, basically say ‘i’m interested in the community of people talking about obsidian and tools for thought, and then the system finds among those you follow the community of people that do that. If you don’t follow many suggested people will appear.
  • This filters by both person and topic, so you don’t get the pines on unrelated topics of people you like/follow.
  • You can tune what you see by + or - the people and posts you see that you like or dislike, which helps assemble the community feed experience you want. This is far softer than muting/blocking, and far simpler than making lists.
  • When browsing anywhere, it will show you them in pines and replies.
  • You have to pick a community you intend to interact with, and the whole experience and content revolves around that.
  • Perhaps actions that bring you closer to someone, like giving praise, offering to help, etc?
• ⚪️ connect across divides - opt-in to be a connector type that represents their differing views, can select what topic they differ on and you can find others that differ.
  • If you fight here, you’re being a dick and can lose privilege.
• 🟢 see all perspectives on a given pine - relatively unfiltered set of replies, rather than summary of replies (below)
• 🪙 sharpen - sharpen your ability to spot different types of content (Literacy Prompt, That’s Bait)
• 🔴 debate truths - formal kialo and community notes-like structure
• surf the canopy of the tallest pines - best of content
• ⚫️ find hidden pinecones - content that is interesting and new, at much lower engagement
• 🟤 rabbit hole - find a topic to go deep on by poking around a wide variety of pines that are specifically in topics that have rabbit hole qualities. (alt name: trailhead). In this mode, you can ask an AI questions about a topic, and it finds related pines and resources about the topic. You can save rabbit holes to return to later, which are roughly like bookmarks but go in a special ‘rabbit hole’ list, and within this it remembers where you last were and what resources you’ve checked out so you can pick right back up.
• 🟠 context - aka see the forest from the trees, gives you bigger picture & context of the pine. Lets you see the trends happening and explore what might be influencing them, both opine-wide and any given pine (Context Mode)
• 🟡 post - posting is its own intention, and the only one that you see the button to be able to post. This may have a ‘post crafter’ help that lets you speak into it or write, and then it reflects back a condensed version of what you’re saying and has you confirm or pick among the claims or ideas you’re putting out. This is a bit of a pre-post quality filter.
• 🟣 learn - shifts feed to learning about new things

Some features of intentions:

• Intention history - you can see the last modes you were in and what you did, so you can get some level of perspective on the usage you have. When you pick up the device it may show that list so you can pick back up
• Intention picker - appears at the top as a sort of filter on the feed and mode for alternative interaction. This is a colored bar at the top saying what it is and with a picker for other experiences.
• Intention timing - each intention can have time limits per day, session, week, etc.

Summarized replies - AI summarizer of what people are saying in replies, along with any given ones that are especially important (pinned by original piner). Tap the summary to unfold to see all of them. This is default UX.

Branching - leaning fully into this metaphor, what if the core text UI was a branching tree, at least in Kialo-mode. The seed was at the bottom, and replies were above it, and they branched in a bunch of directions, and the ‘winners’ became visible, and we could scroll navigate as well as sideways navigate. This is an intentionally constructive process, of people building off other people. And original poster can invite only certain people in, they effectively manage the bonsai by trimming the branches they don’t like or feeding the ones they do. That way they can have several spikey branches.

Forest - seeing forest from trees, this is Context Mode

Cones - something to do with this as koans.

Explaining Why

I’ve gotten a lot of value out of Twitter over the years, and I’ve also found it to be a pretty awful place. But mostly, I’ve been disappointed by it’s lack of innovation. I believe there’s so much that can still be done with short form social media.

2023-09-19 Unpopular Opinion

Unpopular opinion - ‘this is an unpopular opinion’ label on posts that contain opinions that may seem broadly shared, but wind up being very niche. Social media has a way of shaping our perceptions to make unpopular opinions seem far more prevalent, creating a Perception Gap. What if this unpopular opinion mode would show based on how many people are talking about something whether it is something that lots of others express as their opinion as well, vs silent majority or whatever. This could be part of Context Mode.

2023-09-19

Mastodon extensions - I wonder whether its possible to create a mastodon server that has specific extensions such that I could implement Opine Workshop ideas like De-Escalating Social Media. From convo with People > 2023-09-19 Orwa Diraneyya

2023-09-18 Whats the game plan for social?

Media products in the way they are right now I wonder what’s the game plan? Do we really expect human civilization to communicate this way in perpetuity? Do we really think this is a sustainable system? What is the theory behind why there’s so much noise? Is it just netiquette or not knowing people not knowing how to speak?

Twitter-like products in my view are primitive, they are the social media equivalent of cave paintings. Their designs reflect very little understanding of human motivations and behavior and the complexities of social systems. 

“They’re popular, therefore they’re correct!” is the most likely retort to any challenge to them.

2023-07-13 Regrettable Posts

Written on the way back from Crane

Regrettable posts - posts or replies that hit a certain score of regrettable - or users whose posts are likely to create later regret - may have their posts withheld until a later point (say 5 minutes later, or in less time sensitive communications an hour or more later) at which point they may review them and decide whether to post them. This may also be implemented at an OS level for any text inputs that are communications, holding in a common queue. Users may receive later push notifications to review and choose to send or not. During the review process it may require reading the post again to send it (instituting a timer based on content length) while offering a ‘discard / do not send’ button at anytime.

This could further be augmented at the community level. Users who frequently post shitty things may be required to of through the regret queue, whereas those who show they can control their impulses and add value in their posts may not be required.

2008

From Ideas on thinkpad (2008) Earned expansion - someone writes a summary of what they want to talk about, people upvote, and if it gets enough upvotes the writer gets notice they may write more to elaborate.