With the release of various interesting text generation tools, I’m starting an experiment this and next month.

I will be posting computer generated text, prompted by my own current interests, to a separate blog and Mastodon account. For two months I will explore how such generated texts may create interaction or not with and between people, and how that feels.

There are several things that interest me.

I currently experience generated texts as often bland, as flat planes of text not hinting at any richness of experience of the author lying behind it. The texts are fully self contained, don’t acknowledge a world outside of it, let alone incorporate facets of that world within itself. In a previous posting I dubbed it an absence of ‘proof of work’.

Looking at human agency and social media dynamics, asymmetries often take agency away. It is many orders of magnitude easier to (auto)post disinformation or troll than it is for individuals to guard and defend against. Generated texts seem to introduce new asymmetries: it is much cheaper to generate reams of text and share them, than it is in terms of attention and reading for an individual person to determine if they are actually engaging with someone and intentionally expressed meaning, or are confronted with a type of output where only the prompt that created it held human intention.

If we interact with a generated text by ourselves, does that convey meaning or learning? If annotation is conversation, what does annotating generated texts mean to us? If multiple annotators interact with eachother, does new meaning emerge, does meaning shift?

Can computer generated texts be useful or meaningful objects of sociality?

Right after I came up with this, my Mastodon timeline passed me this post by Jeff Jarvis, which seems to be a good example of things to explore:


I posted this imperfect answer from GPTchat and now folks are arguing with it.

Jeff Jarvis

My computer generated counterpart in this experiment is Artslyz Not (which is me and my name, having stepped through the looking glass). Artslyz Not has a blog, and a Mastodon account. Two computer generated images show us working together and posing together for an avatar.


The generated image of a person and a humanoid robot writing texts


The generated avatar image for the Mastodon account

Bookmarked Agency Made Me Do It by Mike Travers

This looks like an interesting site to explore and follow (though there is no feed). First in terms of the topic, agency. I’m very interested myself in the role of technology in agency, specifically networked agency which is located in the same spot where a lot of our everyday complexity lives. Second in terms of set-up. Mike Travers left his old blog behind to create this new site, generated from his Logseq notes, which is “more like an open notebook project. Parts of it are essay-like but other parts are collections of rough notes or pointers to content that doesn’t exist yet. The two parts are somewhat intertwingled”. I’m interested in that intertwingling to shape this space here differently in similar ways, although unlike Travers with existing content maintained. Something that shows the trees and the forest at the same time, as I said about it earlier.

Agency Made Me Do It, an evolving hypertext document which is trying to be some combination of personal wiki and replacement for my old blog. … I’ve been circling around the topic of agency for a few decades now. I wrote a dissertation on how metaphors of agency are baked into computers, programming languages, and the technical language engineers use to talk about them. … I’m using “agency” as kind of a magic word to open up the contested terrain where physical causality and the mental intersect. … We are all forced to be practitioners of agency, forced to construct ourselves as agents…

Mike Travers

Bookmarked Using GPT-3 to augment human intelligence: Learning through open-ended conversations with large language models by Henrik Olof Karlsson

Wow, this essay comes with a bunch of examples of using the GPT-3 language model in such fascinating ways. Have it stage a discussion between two famous innovators and duke it out over a fundamental question, run your ideas by an impersonation of Steve Jobs, use it to first explore a new domain to you (while being aware that GPT-3 will likely confabulate a bunch of nonsense). Just wow.
Some immediate points:

  • Karlsson talks about prompt engineering, to make the model spit out what you want more closely. Prompt design is an important feature in large scale listening, to tap into a rich interpreted stream of narrated experiences. I can do prompt design to get people to share their experiences, and it would be fascinating to try that experience out on GPT-3.
  • He mentions Matt Webbs 2020 post about prompting, quoting “it’s down to the human user to interview GPT-3“. This morning I’ve started reading Luhmann’s Communicating with Slip Boxes with a view to annotation. Luhmann talks about the need for his notes collection to be thematically open ended, and the factual status or not of information to be a result of the moment of communication. GPT-3 is trained with the internet, and it hallucinates. Now here we are communicating with it, interviewing it, to elicit new thoughts, ideas and perspectives, similar to what Luhmann evocatively describes as communication with his notes. That GPT-3 results can be totally bogus is much less relevant as it’s the interaction that leads to new notions within yourself, and you’re not after using GPT-3s output as fact or as a finished result.
  • Are all of us building notes collections, especially those mimicking Luhmann as if it was the originator of such systems of note taking, actually better off learning to prompt and interrogate GPT-3?
  • Karlsson writes about treating GPT-3 as an interface to the internet, which allows using GPT-3 as a research assistant. In a much more specific way than he describes this is what the tool Elicit I just mentioned here does based on GPT-3 too. You give Elicit your research question as a prompt and it will come up with relevant papers that may help answer it.

On first reading this is like opening a treasure trove, albeit a boobytrapped one. Need to go through this in much more detail and follow up on sources and associations.

Some people already do most of their learning by prompting GPT-3 to write custom-made essays about things they are trying to understand. I’ve talked to people who prompt GPT-3 to give them legal advice and diagnose their illnesses. I’ve talked to men who let their five-year-olds hang out with GPT-3, treating it as an eternally patient uncle, answering questions, while dad gets on with work.

Henrik Olof Karlsson

Bookmarked Elicit.org

A while ago I mentioned Research Rabbit here as a tool to find research papers, based on the ones already in my collection (e.g. through syncing with Zotero). Last week I created an account at Elicit. It’s a natural language processing based algorithm to find relevant papers for you based on a specific research question you give it to work with (although it can also take your own collection of papers as a starting point). My first attempt after creating an account yielded very interesting suggestions. Will certainly try this out more, as a tool assisting literature review.

I found Elicit because Maggie Appleton’s feed told me she’s joining the company, Ought, that created Elicit.

Elicit is a research assistant using language models like GPT-3 to automate parts of researchers’ workflows. Currently, the main workflow in Elicit is Literature Review. If you ask a question, Elicit will show relevant papers and summaries of key information about those papers in an easy-to-use table.

Elicit FAQ

In reply to Call for Model Examples of Zettelkasten Output by Chris Aldrich

Even while on hiatus I obviously cannot ignore Chris Aldrich’s call for examples of output creation systems and the actual output created with Zettelkasten style note card systems. For two reasons. One is that I fully agree with him that having such examples publicly visible is important. The other is that I recognise his observations about the singular focus on system design and tweaking often being a timesink precluding outputs (with the loudest voices often being utterly silent on output).

Here’s a first list of outputs from my system, without the receipts though as I’m writing this away from home with limited tools. After the list I’ll make a few general observations as well.

  • I have created 2 or 3 slide decks for client internal and conference presentations from my conceptual notes. First searching for notes on the topic, and the contextual factors of where the slide deck will be used. Then gathering the findings in what I call an ’emergent outline’ (Ahrens calls them speculative outlines). Or perhaps I already have an overview of sorts in the form of an ‘elephant path’ (a map of content, or annotated topical index) which normally help me navigate.
  • I have written blogposts directly from my notes. This is now easier than before, since earlier this year I created a way of publishing to this site from my internal notes. This allows me to write in a note, linking internally or including, all within the notes environment and then push the result out to the website.
  • I created some new personal insights from new connections within my notes. Not sure if that counts towards Chris’ definition of outputs. This results in new notes where the edge, i.e. the newly found link between two notions, gets expressed as a note in its own right. The first such connection (between my notions of Maker Households and Networked Agency) happened when I was about 35 notes ‘in’.
  • For a recent panel at a conference I collated my talking points from my notes
  • I use my notes a lot in work conversations, pulling up concepts as needed. I used to do this to pull up facts and earlier meeting notes with the same participants. Now I also use this to provide richer input into the conversations themselves, including pointing to sources and references. This emerged during the many video calls in the pandemic lockdowns, where it was easy to pull up additional material on one of my screens. Now that I have more meetings in person again, I find I still do this automatically. Whatever material I mention I also link in my own meeting notes. This has been remarked upon by conversation partners as a valuable thing.
  • I have some elephant paths I regard as output in their own right. One currently important to me is the Practices elephant path. It gives an overview of things I want to approach as a practice (which I place somewhere on the spectrum between habit/routine on one end and literacy (in the Rheingoldian sense of skill plus community) on the other end. Practices are the sweet spot to me for (groups of) knowledge workers to implement fields of theory in their own daily work
  • I maintain a client website directly from my notes on EU digital and data legislation. I have conceptual notes for all the regulations involved and maintain summaries alongside them. Those summary notes are automatically synced to GitHub and then published on Github pages as well as the client’s own domain. These same summaries also serve as outline and text for my frequent presentations on this subject, where the slidedeck is kept up to date from the notes that I am certain are always up to date because they are the notes I work with daily.

Some other observations:

What constitutes output? The ‘Luhmann had 90k notes and wrote 70 books’ mantra makes for a rather daunting benchmark to be compared against. I propose we count outputs that have utility to its creator. For me then there are two types of outputs from my notes. A group that is the result of better project tracking, allowing me to pick up where I previously left of, which is a valuable ratcheting effect. Me building my own micropub tools resulted from such ratcheting in 15 minute increments. This group of outputs results from notes, but not the conceptual notes of my ‘Garden of the Forking Paths’ (ie my Zettelkasten style collection). The other group results from re-using and re-arranging the material in my ‘Garden of Forking Paths’ and the example outputs listed above follow from it. In a sense all my work is an output of my notes and my experience, and my tools have always been aiding in my work. Yet there is a qualitative difference.

I have used notes based PKM for over two decades, and in hindsight it was mostly focused on reporting conversations, project stuff, conversations with myself, and many many examples of things I thought relevant. Those I would tag extensively, and I think most of those historic tags would now be their own conceptual notes, expressing the communality of the tagged examples and material, or expressing the link/edge between two or three of the tagged source notes as a notion.

Many of my conceptual notes (now 1000+) and ideas plus non-conceptual atomic notes (another 500 or so) stem from ‘atomising’ my archive of blogposts, and my presentations of the last 10-15 years. Many notes are thus created from earlier outputs themselves.

I recognise what Stephen Downes remarked, that creating the notes is the valuable part towards pattern recognition, and making output needs further gathering of new material. In part this is because adding things to my notes is aiding memory. Once it’s noted it’s no longer novel, and in that sense looses part of the surprisal (informational worth) that led to its creation in the first place. If outputs in my own mind need to be novel, then my notes are limited in value. (This goes back to earlier conversations of the 90% is crap heuristic which I see as feeding impostor syndrom. Outputs imo highly connected to impostor syndrom.

I don’t think I have actual established processes for outputs yet, I’d like to, and I don’t yet feel outputs created suggest as-effective-as-can-be processes yet. Maybe that is because I have not been really tracking such outputs and how I created them. I have become better at starting anything with interrogating my notes first, and putting them together, before starting exploration further afield. Often I find I already have some useful things, which gives a headstart in exploring anything new: there’s something to connect new findings to.

I do not think my current notes could yield something along the lines of a book, other than the nonsense kind of a single idea padded out with anecdotes. I also feel the method of information collection isn’t good enough to base any work on academically. This goes back to the earlier remark as to what qualifies as output of good enough quality.

Robin Sloan has proposed a protocol, Spring ’83, that serves publisher’s content like a magazine stand. You see a board of cards, where cards get replaced whenever its publisher releases a new one. He aims to ditch the timeline experience it seems, partly considering form and content as pieces of the same expression, as well as a way to maintain space for voices that do not express themselves every other minute but way more infrequently.

A Beijing news stand with spread out mags competing for your attention. Image by Peter Ashlock, license CC BY

Others in my feedreader have commented on it in the past days and it gets me thinking. Not in any structured way yet. No idea yet therefore what I think about this in a form I can narrate, but some associations come to mind.

I do like the notion of small cards. Makes me think of Hugh’s Gaping Void back-of-a-business-card drawings, and of tiny zines made as a folded single sheet chapbook. The set limit creates friction for creativity to feed on. Yet, the built in size limit, when putting more of them together on a ‘board’ may well mean the same drawbacks as in Twitter, aiming for the highest attention grabbing value. Magazines in a kiosk do the same thing after all, using the cover to try and lure you into reading them. Look at that image above. Does that make a board of cards just a collection of adverts for your attention? Reading Maya’s annotations, there too the scarcity mindset a board of such cards might introduce is raised. Are there other ways to thread such cards?

The focus on p2p distribution, and on making it easy to put out there, chimes with me in terms of networked agency and in terms of low thresholds for such agency.

The notion where softer voices have the same claim to space as louder ones (i.e. more frequently posting ones) I appreciate a lot. Kicks Condor in his Fraidycat feedreader provides neat sparklines indicating frequency of posting, and allocates every single author the same space by displaying their last few postings regardless of timelines. That points back as well to my use of social distance (not the pandemic kind!) as a method to order presentation of feeds I follow, in a person focused way, and less a timeline. I follow people’s expressions, not blogs as publications. It also makes me cringe at the use of the word publisher in Robin Sloan’s explanation.

À propos following people, Maya also mentions how she likes to see friction between different strands of her online expression (e.g. blogposts, and Mastodon messages). Such different strands have different qualities to them, and having them in one place, like an IndieWeb enabled site may put them too closely or too obviously together. The notion of friction is important I think when getting to know someone online in more detail by following more of their online traces. I follow people, and for a good number I follow multiple traces (photos, posts, tweets e.g.). Combining those traces needs friction I feel, getting to know someone better from their expressions needs a certain effort. That’s about me having something at stake in building interaction. Blogs are distributed conversations to me and you need to invest your presence in such conversations. Connecting with others should be extremely easy in terms of being able to connect, but certainly not effortless in terms of time spent on the actual connecting. Way back when (2006), Lilia and I had conversations about this, and it’s still relevant now. My site purposefully introduces friction to readers: casual visitors see only a fraction of the postings, some content is only shared through RSS and not findable in the site, some content is both not listed nor shared through feeds etc. All the fragments are still in the same place, mine, though, and not farmed out to various silos to create the same effect of deliberate fragmentation. It means I’ve greatly reduced the friction for me as author using IndieWeb, not eroded the needed friction for readers. Someone who puts in the effort will be able to gather all my traces in their reader.

Tracy Durnell has some remarks, and compares Spring ’83 to IndieWeb efforts and discusses the visual aspects. Her suggestion showing a blogroll as cards, not as a list, is a good one I think, perhaps showing the last three postings the Fraidycat way? I’ve seen others do it as a river of news, but that once more provides additional amplification to the loudest authors.
Louis Potok takes a first look under the hood.