NEWLY UPDATED WITH VIDEO! Your Metaverse Is Too Small: My Keynote Presentation to the 2026 Virtual Worlds Best Practices in Education (VWBPE) Conference

Marie Vans (left) introduces me (right) as a keynote speaker at the 19th annual Virtual Worlds Best Practices in Education conference, held in the pioneering virtual world of Second Life (screen capture from the video at the end of this blog post)

As some of you might know, a while back I was asked by my librarian colleague (and fellow Second Life aficianado) Marie Vans if I would be willing to be one of the three keynote speakers during the 2026 Virtual Worlds Best Practices in Education (VWBPE) Conference, which is held every year in the pioneering virtual world of Second Life. I said yes (of course!). I never turn down an opportunity to give a presentation in my beloved Second Life. (I was asked to speak at the 2021 Virtual Ability Mental Health Symposium, giving a presentation on the topic of this blogpost on acedia during the pandemic, and in 2024, I gave a presentation on virtual world building in Second Life, in Second Life, to a graduate class in virtual world building and design, which was team-taught by a computer science professor and an interior design professor.) So, as you can see, this was not my first rodeo. 😉

The title of my presentation, which I gave as the keynote speech of the conference on its second day, Friday, March 20th, was Your Metaverse Is Too Small: How the Biases and Preconceptions of Virtual Worlds Hinder Their Use in Education. I was inspired by a keynote address at last November’s IMMERSIVE X conference by Andy Fidel, who titled her talk, held on the ENGAGE social VR platform: The State of the Metaverse in 2026. My talk therefore consisted of three parts as follows:

  • Quotes from Andy Fidel’s talk which I found inspiring and wanted to share (7 slides);
  • My initial, general observations about the metaverse (4 slides); and finally
  • A section titled Your Metaverse Is Too Small, where I discuss various ways our biases and preconceptions about virtual worlds and social VR/AR actually hinder their effective use in educational settings (with reference to Andy’s comments; 14 slides).

Yes, it’s a lot to cover in 45 minutes, but I did it! I’m just going to share my slides as-is, without any Creative Commons-type license this time around, since a lot of it is referring back to Andy Fidel’s ideas, which I found so inspirational in the first place. And yes, while the topics of these 14 slides in part 3 all sprang out of one particularly fevered brain dump of my ideas one evening, rather than relying on GenAI, I do freely admit that I fed my entire blog into a Google NotebookLM and asked it questions in order to create the content based on thirteen (yes, 13!) different ways that, quote: “your metaverse is too small!

Your metaverse is too small because…

  • it has too steep a learning curve for new users
  • your platform has a poor fit-to-purpose
  • it lacks accessibility features (e.g. speech-to-text for the Deaf/HoH community)
  • it is poorly designed and/or Quality Assurance tested, and it causes VR sickness/nausea (more common among women than men)
  • it is soulless/designed by committee (hello Meta Horizon Worlds and Workrooms! Proof positive that you cannot will metaverse platforms into existence by executive fiat and the spending of billions of dollars.)
  • it requires a VR/AR/XR headset (I used two slides to discuss this controversial take; see below for more detail)
  • it relies on cryptocurrencies, NFTs, or some other form of blockchain (do I really have to explain this at this point?)
  • it has poor (or non-existent) safety and trust features and policies
  • it focuses on the product rather than the community
  • it has user data privacy issues/is based on surveillance capitalism (once again, hello Meta Horizon Worlds and Workrooms!)
  • you fail to market it properly (or, in many cases, fail to market it at all)
  • it is unfriendly to different cultures and subcultures (e.g. trans people, furries, etc.)
  • (another controversial one, explained further below) it refuses adult content

Now, before all you social VR adherents rise up with torches and pitchforks and tar and feather me for even daring to say “your metaverse is too small because it requires users to have a VR/AR/XR headset,” here are the two slides, plus speaker notes:

Your metaverse is too small because it requires you to use a virtual reality headset.

It is not a surprise that many of the most popular social VR platforms (e.g. Rec Room, VRChat) also allow for non-VR users to participate. The VR headset market still has not taken off. Even the best-selling Meta Quest line of wireless virtual reality headsets (which make up an estimated 70% of the global VR headset market) has sold only approximately 30 million units around the world, and many of those devices land up collecting dust after the initial novelty of the product wears off. Apple’s Vision Pro, launched to enormous fanfare, does not publish sales figures, but industry reporters said that the company shipped approximately 390,000 units in 2024 and approximately 90,000 units in 2025. VR hardware remains bulky, heavy, and uncomfortable for extended wear. Nausea continues to affect a significant proportion of users. A VR headset isolates the wearer from their physical surroundings and from the facial expressions of the people around them—an anti-social device, in practice, even when its purpose is socialization via social VR.

The PC-tethered VR headset market—the high-end, high-fidelity segment that many early social VR platforms had built toward—proved especially stagnant. The dream of millions of consumers owning gaming-grade PCs with tethered Oculus Rift or HTC Vive headsets never materialized. Even the shift to standalone headsets like the Meta Quest series failed to generate the consumer mass-market that had been anticipated. Sansar is perhaps the most instructive case study in the danger of building a platform around assumed headset adoption. Developed by Linden Lab—the company behind Second Life—Sansar was announced in 2014 and launched in beta in 2017, timed almost precisely to coincide with what seemed like the dawning of the VR era following Facebook’s acquisition of Oculus. But the bet on tethered PCVR headsets never paid off. In March 2020, Linden Lab sold Sansar to a little-known startup called Wookey and pivoted to focus on live music events and virtual concerts, attempting to find a more sustainable niche. That hasn’t worked, either. Sansar still exists, but it is only being kept alive by volunteers at this point.

Research on technology adoption consistently shows that devices requiring behavioral change—for example, for VR, wearing something on your face, isolating yourself from physical surroundings—face much higher adoption friction than technologies that integrate into existing habits. The iPhone and Android phones succeeded partly because they fit into already-established phone-carrying behavior. VR headsets require building a new behavior from scratch.

The failure of the last metaverse hype cycle does not mean that immersive technology has no future. What failed was the specific prediction that millions of people would soon be spending significant time in virtual worlds accessed primarily through VR headsets; that this would create platform-scale opportunities comparable to social media or mobile devices.

Don’t hate me for speaking facts. No VR/AR headset (even the Meta Quest line of headsets) has taken off in the way that iPhones/Android phones and tablets and smartwatches have. In particular, the developers of those platforms who bet the farm on widespread adoption of high-end tethered PCVR headsets (hello, Sansar and High Fidelity!) lost that bet badly; Sansar is essentially moribund, and High Fidelity is now closed (although it does live on in its successor social VR platforms Vircadia and Overte, which were based on HiFi’s open-source codebase, but are also not heavily used). This failure is one of the reasons why Second Life is still going strong (or strong enough) to endure and still be profitable for Linden Lab, for over 22 years now.

And speaking of SL…

I want to make one thing very clear: in some educational applications of the metaverse (especially those intended for children and teenagers, i.e. K-12 education), a ban on adult content is absolutely necessary.

However. As my speaker notes for this last slide in my presentation state:

However, in any institution of higher learning (e.g. a university). you will find faculty, staff, and students teaching about, learning about, and doing research on topics which may include controversial or adult topics. I have argued that one of the most significant strategic errors a metaverse platform can make is the outright refusal to host adult content (or do some other sort of heavy-handed sanitization of adult content, like imposing baked-on underwear on the base male and female adult avatars).

In my blog, I’ve pointed out that for some successful virtual worlds, adult communities are not just a niche—they are the economic and social engine that keeps the lights on. I have frequently cited Second Life as the prime example of a platform that understands the value of adult content. On my blog, I’ve noted that the adult-rated regions of Second Life generate a good portion of the platform’s revenue through land tier fees and the sale of virtual goods (clothing, skins, animations). In contrast, I wrote about Sansar’s early decision to strictly moderate content and its struggle to establish a clear policy on adult material. I argued that by trying to keep the platform “brand-safe” for corporate partners, they essentially “cut off their nose to spite their face,” alienating a potential demographic of creators and consumers who were ready to spend money on higher-fidelity adult experiences. And the corporate clients never came anyway!!

I believe that the ability to explore one’s identity—including its sexual or adult aspects—is fundamental to the metaverse experience. For example, both Second Life and VRChat tend to attract the trans community, giving them a way to experiment with how they represent themselves in a way that might be difficult or impossible to do in real life (particularly at a time when trans people are increasingly under attack in certain jurisdictions). Platforms that ban adult content often end up banning people by extension. If a platform’s moderation is too aggressive, it can lead to the marginalization of subcultures (like the furry community or the trans community) who use virtual worlds as a safe space for exploration. This aligns with Andy’s focus on “presence” and “feeling seen”. Andy argues that gathering spaces should be “smaller, weirder,” and more human. I have argued that by refusing to host adult content, platforms are choosing “corporate safety” over “human authenticity.” They are creating “noise” for brands rather than “spaces that matter” to real people.

One of my core arguments is that you cannot impose a culture on a virtual world; the users bring the culture with them. I’ve pointed out that in almost every successful social VR platform (like VRChat), “NSFW” content and communities exist regardless of official policies. Trying to ban these things is like trying to stop the tide with a broom. Platforms that fight their own communities on this issue usually lose the “heart and soul” that Andy Fidel says is required for a space to be successful. Andy speaks about “architecting belonging” and building spaces like cities. A real-life city has red-light districts, gay bathhouses, private clubs, and adult stores. By refusing to allow these “niche micro-communities” to exist, platform owners are failing to be the architects of a real society and are instead acting as corporate landlords of a sanitized shopping mall.

Okay, enough ranting. Here’s my slide presentation, which you can download to read the rest of my slides and my speaking notes:


Please note: while Philip Rosedale’s keynote speech on the first day of the Virtual Worlds Best Practices in Education conference on Thursday, March 19th has already been uploaded to the VWBPE YouTube channel, mine has not yet been uploaded to view. When it is, I will update this blogpost with a link to the video of my talk.

UPDATE March 30th, 2026: The video of my presentation in Second Life has now been uploaded to the VWBPE YouTube channel! Here it is (and I haven’t even watched it myself yet):

UPDATED! Generative AI Update, March 2026: My Updated Presentation on Artificial Intelligence and GenAI, Plus My First Thoughts on the Claude Add-In for PowerPoint, and Yet Another Head-to-Head Comparison Between Claude, Gemini, and ChatGPT

I am (as you can clearly tell by this absurdly long blogpost title) trying to do three related things here. If you want, you can skip to the very end, where there will be an executive summary, where I have some thoughts to share about (waves hands) all this.

First, I wanted to share an updated version of the original slide presentation on artificial intelligence and generative AI, which I shared in a December 2025 blogpost. I used to think that keeping track of the many metaverse platforms I blog about was a task similar to herding cats, but let me tell you, it was a breeze compared to trying to stay abreast of all the rapidly changing and accelerating developments in generative AI!

Keeping on top of developments in generative AI is like herding cats, where the cats are multiplying and mutating!
One of the updated comparison charts in my PowerPoint slide deck (see link below to download)

Below is my updated PowerPoint slide presentation, complete with my speaker notes, for you to download and use as you wish, with some stipulations. I am using the Creative Commons licence CC BY-NC-SA 4.0, which gives the following rights and restrictions):

Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International

This license requires that reusers give credit to the creator. It allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, for noncommercial purposes only. If others modify or adapt the material, they must license the modified material under identical terms.

BY: Credit must be given to you, the creator.

NC: Only noncommercial use of your work is permitted. Noncommercial means not primarily intended for or directed towards commercial advantage or monetary compensation.

SA: Adaptations must be shared under the same terms.

(The tool I used to determine the appropriate Creative Commons licence can be found here: https://creativecommons.org/chooser/.)

So, with all that said, here is my PowerPoint presentation (please click on the text link or the black Download button under the picture, not the picture itself):


NEW: Claude a just released add-ins for Microsoft Office

Second, today I installed a brand-new add-in from Anthropic’s Claude GenAI tool, which is supposed to work with Microsoft PowerPoint. This is an initial review of a very beta product.

And I have an actual real-world use case against which I will be trying out this new add-in: the design of an actual keynote presentation which I will be giving in a couple of weeks. (I am also using it in the third section, but in a different test of all three of ChatGPT, Claude, and Gemini.)

Now, before I get into this, I should explain that I have tried in the past with all three GenAI tools on which I currently have paid accounts (OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini) to create a PowerPoint slide deck presentation design—only to get highly disappointing and completely unusable results back. So I was not expecting much here, particularly as this is a still a research beta version of the PowerPoint add-in.

My initial prompt to the Claude add-in to Microsoft PowerPoint was:

Please create a new PowerPoint slide presentation design with the title of the presentation being: “Your Metaverse Is Too Small: How the Biases and Preconceptions of Virtual Worlds Hinder Their Use in Education.” The theme of the talk is educational uses of virtual worlds, social VR, and the metaverse in general. I want to have some nice background images to use in some of my slides, as well as a visually pleasing title slide. I’d prefer blue as a colour in the slide deck theme, thanks!

And Claude chugged away on my request, keeping my posted on what it was doing:

And it even prompted me to be sure I wanted to delete the Claude add-in help slide!

The set-up for the title slide took a long, loooong time, much longer than I would taken to click on the Designer button in the PowerPoint toolbar and just select one of the default options, and a colour scheme. Eventually, I just gave up on waiting and went off to work on another task, leaving Claude to beaver away. After fifteen minutes, I realized that I still had to explicitly okay the clearing of the original slide design (inset Homer Simpson “D’oh!), which I did, so that the work could continue.

If I could summarize the result in one word, it would be: meh (again, shout-out to The Simpsons):

I mean, I could easily do better than this myself. And two dots do not make, as I asked for, “some nice background images to use in some of my slides, as well as a visually pleasing title slide.” Here’s my section title slide:

Again, extremely underwhelming, and frankly, not an improvement at all over my previous failed attempts to generate a PowerPoint slide presentation design using any of the GenAI tools (Claude, ChatGPT, or Gemini). Mind you, I have deliberately stayed away from using the image-generation tools in these products; I can spot a GenAI-produced image from a mile away by this point, having been playing around with these tools, off and on, since they first came out in 2022.

Claude continued to generate all the standard versions of PowerPoint slides in this theme, ending with a final slide that, I must confess, I kind of liked the look of (although, again, I would have preferred some sort of background image):

This is where the process got interesting, as I finally decided to stop having to manually okay each individual step, and just gave Claude carte blanche to do whatever it felt was best. (I mean, the worst that could happen was that it come up with something I hated so much that I threw it away and started over.)

Claude was still working away while I took my lunch break, giving feedback along the lines of “Build stunning title slide design.” 🙄 (I’ll be the judge of what’s considered stunning, Claude. Calm the fuck down.)

Here’s the final result, my “stunning” title slide (insert RuPaul’s Drag Race shade death rattle):

The addition of three pieces of clip art in the upper right corner of the slide, plus a few more bubbles/dots. So, yes, this is, once again, a complete fail. I will probably still use this as a basic slide design, but obviously I will be locating and using my own images to illustrate it. This is now the second new tool in a week (first Claude Cowork and now Claude PowerPoint add-in) which has utterly failed at the tasks given it. I am not impressed.


Third, and finally, thank God, I had much better luck was in issuing all three general-purpose GenAI tools the exact same text prompt, a technique I had used before here (and one which I found very useful in comparing and contrasting the responses):

I am writing a keynote presentation on the mistakes companies make when creating, designing, and marketing the following product category: virtual worlds, social VR/AR, and metaverse platforms in general. Please give me a list of failed or shut down metaverse platforms, along with reasons why they might have failed. Please cite both academic and industry sources of information in your answer.

In all cases, I used the latest models as specified in Ethan Mollick’s latest AI Guide:

  • ChatGPT’s GPT 5.2 Thinking with the Extended Thinking option;
  • Claude Opus 4.2 Extended Thinking with the Research option; and
  • Gemini 3 Thinking with the Deep Research option.

Unlike the last comparison, I’m not going to go into great detail on the results (because I will be using some of these results, once they are double-checked against more authoritative sources, in an actual keynote presentation I will be delivering later this month). Instead, I will my general overall impression of each report (and all three did provide a detailed report with citations).

Please note that I deliberately left it up to the specific GenAI tool to define what “failed” or “shut down” means, how far back and how thoroughly to search for failed platforms, and what metaverse platforms to include or exclude from its final report. As always, I find the differences between the reports to be an interesting way to compare and contrast the results, so below I will give some basic statistics:

GenAI Tool# Failed Platforms ListedTime Range of Failed Platforms# Citations in Final Report
ChatGPT152003 to 202623
Claude13(start dates not given) to 2023/”effectively failed, still limping along”30
Gemini92009 to 2024 (but some platforms had no timeline information given)33

While ChatGPT was the most thorough in listing failed metaverse platforms, and seems to have gone the furthest back in time (including There.com, which launched back in 2003!), it also had the fewest number of citations, and most of them were historical, platform-related announcements (e.g. a 2020 announcement of the shutdown of the then-social-VR platform High Fidelity by its CEO) rather than peer-reviewed academic journal articles (although there were a couple of those, too). While Claude had more citations, a review of those showed mostly blogs and news websites, with fewer references to actual academic research papers (probably because much of that content is locked behind academic publisher paywalls, although there were still quite a few academic references to free sources such as ResearchGate and PubMed Central/PMC; see the Claude report image below for one section which did focus on academic sources). Of the three, Gemini’s 33 citations used included the most resources which I would consider academic, from a good range of different publishers (as well as more informal websites). Interestingly, Gemini also included a list of resources which it looked at, but chose not to include in the final report, something which neither ChatGPT nor Claude offered! I thought that was particularly valuable, in case something else caught my eye to follow up on. Gemini for the win here.

Gemini was also notable for the strong, overarching narrative structure to its report, something which I had also noticed in previous queries using this GenAI tool. Gemini has clearly been trained well in telling a cohesive story! However, Claude was also notable for listing, in a separate section of its report, what it called “cross-cutting failure themes” in its 13 examined metaverse failures (which is definitely a phrase I will be stealing for my final keynote presentation!). By comparison, the final report from ChatGPT, while thorough, was jargon-heavy, poorly-formatted, and seemed to lack the final polish of its competitors. For example, there were three separate sections titled “failure themes and comparative analysis,” “theme-to-platform mapping,” (?!??) and “top 10 failures by primary cause.” It was, in my opinion, the poorest of the three reports generated, just in terms of sheer (lack of) organization and narrative. Again, Gemini for the win!

Gemini’s report had a strong, overarching narrative structure—something which I have noticed seems to be a particular strength of this GenAI tool, a sort of final overall polish to the text that ChatGPT, in particular, was lacking in its report (see below).
Claude’s report had a summary section titled “cross-cutting failure themes,” which I am definitely stealing for my keynote presentation!
Compared to the Gemini report, the ChatGPT report was jargon-heavy and poorly-formatted.

EXECUTIVE SUMMARY: So, here are my final thoughts.

  • It is getting harder and harder (in fact, almost a full-time job) to keep on top of what is fast becoming an arms race between the top three general-purpose generative AI tools (ChatGPT, Claude, Gemini), not to mention an ever-growing legion of more narrowly-focused applications, which might be better at certain specific tasks, such as writing programming code or generating music.
  • While Claude seems to be good at putting new agentic (e.g. Claude Cowork) and add-in tools (Claude for PowerPoint) into the hands of its users first, my personal experience with these new tools has been very disappointing, even comically bad. However, Claude’s chatbot interface works well for generating detailed answers with citations (although slightly edged out by Gemini).
  • I am impressed by Gemini’s consistent ability to create a strong narrative structure within its generated reports, something in which ChatGPT in particular is noticeably lacking. It also came first in a key metric: actual citations to academic literature, not just freely-accessible websites (blogs and news articles).
  • If I were forced to rank the three GenAI tools by just this one head-to-head-to-head comparison (i.e. the third part of my blogpost), I would rank them as follows:
    • 1st: Google Gemini.
    • 2nd: Anthropic Claude.
    • 3rd. OpenAI ChatGPT.
  • Again, when these GenAI tools work, they work well (sometimes very well!), but they they fail, they fail spectacularly. Which, in my mind, is another reason why it is good to put these tools to the test regularly, and use them in real-life situations, so that you can learn what they are good and bad at!

My Presentation on Social VR in Higher Education

My half-hour presentation on virtual reality in higher education (with an emphasis on social VR), which I gave to my university’s Senate Committee on Academic Computing, was well received. As promised, here is a copy of my PowerPoint slides from yesterday’s presentation (60 slides, 191MB in size; it’s so large because I included a number of animated GIFs). I would have liked to embed my slides into this blogpost, but the various ways I tried unfortunately failed to work. You will have to download the slides and run them on your own computer!

Some of the examples I used in this presentation (and, in particular, the Barriers and Challenges section) I took from a new report by the Immersive Learning Research Network (iLRN), their State of XR and Immersive Learning Outlook Report 2021. I’d like to thank the writers of this 90-page report for their work! They made my job much easier.

I had been sorely tempted to set this so that you had to join my Patreon in order to get it, which would have cost you at least $1.00, but in the end I decided that it would be seen by more people if I made it free to download. However, I do reserve the right at a future point to start doing this for some of my content! From the bottom of my heart, I would like to thank my Patreon patrons; their support means the world to me! I am going to rack my brains to see if I can come up with new perks for you!

I Will Be Giving a Presentation on Social VR at the Virtual Germany Meetup in ENGAGE on April 28th, 2020

Mark your calendars! I have been invited to give a presentation to a Germany-wide VR Meetup in ENGAGE on Tuesday, April 28th, 20:00 CET (11:00 a.m. PST). The event is called Virtual Germany, and it is described as:

In cooperation with the VR communities from Berlin, Bremen, Cologne, Erfurt, Frankfurt, Hamburg, Hannover, Leipzig and Munich, we are planning the first virtual VR and AR Meetup in Germany.

Look forward to exciting lectures on the topic of social VR and various guests from all over Germany and around the world.

One of the many people involved with Virtual Germany is the illustrious XaosPrincess, who is well-known for her work in High Fidelity, and whom I have written about before here, here, and here. She tells me:

“Virtual Germany” is organized as democratic collaboration of many German real-life Meetup groups (most of Germany’s states are represented). Our host in ENGAGE is Rolf Kruse, while the whole organization is a combined effort.

The topic of my 20-minute presentation is “Overview of the Most Important Social VR Platforms”. The talk will be in English. (I do speak German, but I do not consider myself fluent in the language.)

This will be a semi-public, RSVP event in ENGAGE, which will be advertised in each separate German VR Meetup group. Xaos tells me:

The attendees will be asked to register with the ENGAGE event to be admitted (and to give us an overview on the numbers we can expect – probably/hopefully 100+ visitors minimum). We’re still preparing the final ENGAGE event link and will post it as soon as possible 🙂

We’re going to have four rooms:
– One main room with live talks and presentations (Moon)
– Two backup rooms for additional visitors with live YouTube-Streams of the talks in the main room (Mars & Earth)
– One chill and networking room without any streams (a high rise apartment)

The talks will also be streamed live on YouTube.

If you do not have the ENGAGE software yet, you can download and install it for free here. ENGAGE works on desktop or laptop computers running Windows 10 and a wide variety of VR headsets (here’s a list of supported devices). When there is an event listing on the ENGAGE events calendar, I will update this blogpost with a link to that.

See you on April 28th! I am quite looking forward to this.