iiinova — where your digital self keeps evolving

Geobica @geobica.com · 06/09/2026

In related news to that there's another phantom island near Iceland there in the bottom right

Geobica @geobica.com · 06/09/2026

For my case I honestly can't say having access to AI has made me more productive, in the sense that I'm not able to get done more projects that I care about. I can get infinite lines of code but fundamentally I'm limited by what feels important to me and I've been slowly rediscovering what that is.

000

Geobica @geobica.com · 06/09/2026

I think this sums up how I actually feel about the particular websites involved here

100

Geobica @geobica.com · 06/09/2026

I'm typing this because I feel a need to vent somewhere on the internet a bit of the anxiety I feel around this stuff because it does feel like stuff gets made and spread a lot faster than I'm used to, and I really do enjoy taking my time to perfect stuff, and I appreciate when other people do too.

100

Geobica @geobica.com · 06/09/2026

I was never going to finish that project and now I don't need to to see an approximation of what it might look like.

100

Geobica @geobica.com · 06/09/2026

and I get particular about being very accurate with stuff like that, having as reliable data as I can manage. That site doesn't try as hard to have scientific accuracy but it also doesn't need to to be what it is and give a bit of that sense of perspective it aims to.

100

Geobica @geobica.com · 06/09/2026

Like anyhumanever.com is an example of a site that was unmistakably vibe coded. It's also very similar to a project idea I had a few years ago that I never got around to finishing, mainly because getting accurate data of historical population distributions across different regions is difficult

100

Geobica @geobica.com · 06/09/2026

krillion is one of those projects where it's just kinda neat but it does bother me that there's this part of my brain that's become fixated on whether stuff is AI or not that I can't get to shut up, I'm sure others relate

110

Geobica @geobica.com · 06/09/2026

anyway this does tie back in to the vibe design thread I was on about earlier because I do remember seeing a few other sites with that similar red/cyan 3D filter effect that I think is particular to fable, but I have no evidence of that on hand at the moment and it doesn't matter either way

110

Geobica @geobica.com · 05/09/2026

why are there so many krillions. which one was the original lol

duckduckgo search results for "Krillion" showing:
krillion.me
playkrillion.com
krillion.lol
krillion.live
krillion.org
krillion.to

120

Geobica @geobica.com · 04/09/2026

Summary of the timeline because I learned more as I was looking around: bsky.app/profile/geob...

020

Geobica @geobica.com · 04/09/2026

the 2003 Vaux survey had 122 questions, the 2013 Katz survey had 25, mydialect.us has 122+those 26 like the peach v apricot one

000

Geobica @geobica.com · 04/09/2026

Also the 2003 survey had more precision than statewide, as did Josh Katz's survey, and so both of those pages have maps with more than statewide precision, but AI agent creating the mydialect.us page only found the statewide averages, and so had less precision, with 26 extra hallucinated questions

100

Geobica @geobica.com · 04/09/2026

2002-2003: Bert Vaux and Scott Golder run the Harvard Dialect Survey 2013: Josh Katz does a new survey using questions based on that one and uses that data for the NYTimes quiz 2026: Tim Stumbaugh and Jenil Kansara use an AI agent to create mydialect.us using the 2003 data at dialect.redlog.net

110

Geobica @geobica.com · 04/09/2026

Correction: The data is not the exact same, the way it went was:

110

Geobica @geobica.com · 04/09/2026

btw for more on how to recognize AI generated websites see this thread bsky.app/profile/geob...

0193

Geobica @geobica.com · 04/09/2026

wait no correction, the data is new data from 2013, says so on the NYT page, though the vibe coded one was using the 2003 data from dialect.redlog.net

dialect.redlog.net

020

Geobica @geobica.com · 04/09/2026

I've posted about this a few times because I find this sort of low effort vibe coded project is a real threat in how it makes bad analysis look high quality with its visuals, but there's a noticeable way these things tend to look that I keep seeing

1243

Geobica @geobica.com · 04/09/2026

also bsky.app/profile/geob...

150

Geobica @geobica.com · 04/09/2026

it's an AI generated website with fairly naive methodology, which I wrote more on in another thread basically it uses aggregated per-state data with very basic math to average it rather than the raw data which was much more granular bsky.app/profile/geob...

410033

Geobica @geobica.com · 04/09/2026

note in this that I quoted I said it was the same exact data, but later down in the thread I find that it's per state, the raw data must have been used by the NYT version but it's not easy to find online, so presumably whatever AI agent they used decided it wasn't worth it

130

Geobica @geobica.com · 04/09/2026

it's an AI generated website that uses the aggregated per-state data with very basic math to average it rather than the raw data which was much more granular bsky.app/profile/geob...

160

Geobica @geobica.com · 04/09/2026

Perhaps I could find the exact numbers in the Harvard survey later but as it is I do not trust that the data in here wasn't literally just hallucinated by the LLM.

150

Geobica @geobica.com · 04/09/2026

anyway the github page is an unfinished version of the website with LLM-generated "fill this in later" type blanks in the README github.com/jenilkansara...

181

Geobica @geobica.com · 04/09/2026

Looking at the .js of the website itself it seems that the answers are stored on a per-state basis rather than a per-city basis like the NYT one, which explains the New Jersey thing, and also implies the maps are just a gaussian kernel on the centroids of the states and...yeah that's how they did it

181

Geobica @geobica.com · 04/09/2026

Q21 I'll also note wasn't in the Harvard Dialect Survey so I'm not sure what data they used for that

010

Geobica @geobica.com · 04/09/2026

(also, why does it say New Jersey at the top but list Louisiana below? Is that a bug?)

430

Geobica @geobica.com · 04/09/2026

mydialect.us doesn't do as good a job weighing the statistics. A single question can completely change the look of the map even if it's a minor thing. Like I answered 20 questions and here's if I say "Peas" on Q21 vs "Green Peas" on Q21:

291

Geobica @geobica.com · 04/09/2026

Based on the source, it uses the same exact data as the NYT one, though with more of the questions from that original Harvard Dialect Survey. I imagine the NYT one just narrowed down to the questions which would have the strongest correlation for the final results, and based on the maps I see...

2103

Geobica @geobica.com · 04/09/2026

Q21 I'll also note wasn't in the Harvard Dialect Survey so I'm not sure what data they used for that

000

Geobica @geobica.com · 04/09/2026

(also, why does it say New Jersey at the top but list Louisiana below? Is that a bug?)

100

Geobica @geobica.com · 04/09/2026

100

Geobica @geobica.com · 29/08/2026

I've found that there's a phantom island in the middle of the Greenland Sea in the GADM database at 80.136°N, 0.018°W. No such island seems to exist irl, so maybe it was a glacier that got scanned? It was messing with some code I was running as it's right in the middle between Greenland and Svalbard

Screenshot of https://gadm.org/download_country.html with the phantom island shown

010

Geobica @geobica.com · 28/08/2026

In part this is because when you use AI it doesn't feel like you're doing any of those things, though I think it's also just how Silicon Valley types talk about their projects in general. A lot of these websites suffer from the LinkedIn voice in their prose in general.

050

Geobica @geobica.com · 28/08/2026

Another trend I forgot to mention: They almost always say they "build" the website. They didn't code it, design it, create it, or make it, it's always "built".

140

Geobica @geobica.com · 28/08/2026

A lot of these sites say "we" or "us" to imply they were made by a team, but without any actual link back to personal profiles of anybody who worked on it. I suspect almost all of these are made by one person.

150

Geobica @geobica.com · 28/08/2026

hormuzstraitmonitor.com/ is an example I've come across in the wild. This is a horribly designed website for a lot of reasons, but it's come to be the top search result for "hormuz strait status" when I look that up. It attributes its creation to "an individual developer". Nothing else to learn.

From hormuzstraitmonitor.com/about:
Who Is Behind This?

The Hormuz Strait Monitor is an independent project built and maintained by an individual developer. It is not affiliated with, funded by, or endorsed by any government, political organization, military entity, think tank, or corporation.

The project was created to fill a gap in publicly available crisis information. During the Strait of Hormuz crisis, reliable data was scattered across dozens of specialized sources — shipping databases, energy market feeds, diplomatic wires, and news outlets. This dashboard aggregates those sources into a single, accessible view for anyone who needs it.

The site has no editorial agenda. It presents data as reported by its upstream sources without political commentary or interpretation. Where AI-powered analysis is used (such as for strait status assessments and diplomatic summaries), the underlying methodology relies on real-time web data from established news and maritime sources.

150

Geobica @geobica.com · 28/08/2026

How this materializes is often these websites will pop up with no author is attributed, nobody seems to be maintaining it, and, as we've seen, they often go down pretty quickly.

150

Geobica @geobica.com · 28/08/2026

Tell #5 is the relationship between the creator of the website and the website itself. How much do they really care about it? The fact that AI speeds up development time means that people can now make stuff that looks like it has a lot of depth on the surface but which nobody really cares about.

150

Geobica @geobica.com · 28/08/2026

Ideas like the "Trust Dial" for example are clearly pretty half baked. This sort of demonstration would in the pre-AI era have taken quite a while to set up, so the whole thing feels as if it's thought out at first, but in reality it's unlikely they spent more than an hour on this.

The "Trust Dial", an interactive widget showing a "planned" feature on the twitter.now site. In several places it's implied that this has not actually been implemented or tested.

150

Geobica @geobica.com · 28/08/2026

A lot of the weirdness of the priorities in this case could just as easily be explained by it being a scam than it being Vibe Designed, but in the places where effort would be spent, the fact that the planning process wasn't slowed down to match the speed of the creation process shows.

150

Geobica @geobica.com · 28/08/2026

Instead, this landing page has a description of the concept of a Twitter reboot, because all they did was toss some thoughts at an AI and had it translate those into a webpage.

160

Geobica @geobica.com · 28/08/2026

For now, consider the twitter.now page. If you were actually trying to reboot something like Twitter, your first priority would probably be to actually host a mastodon or atproto instance and get an interested userbase, while starting work on the features that you want to make your version distinct.

150

Geobica @geobica.com · 28/08/2026

I have a lot more criticism I could give of this particular website and the way it was handled but that I think should be its own thread for later.

150

Geobica @geobica.com · 28/08/2026

But instead it just gets repeated with different wording, losing some of the original intent, because it's not really designing a website, it's converting one document format to another in a lossy way.

150

Geobica @geobica.com · 28/08/2026

This idea gets repeated twice on the about page, but without ever explaining what that means. I still have no idea what they mean by that! If this website was made by a person with a deeper understanding of the project, this would be an important thing to elaborate on, perhaps with its own page.

From the website:
10% of CT population reached via AI

From the website:
First Large-Scale Use of AI to Link Assembly Discussion with the Public

Using cutting-edge AI technology, the assembly aims to connect its deliberations with over 10% of Connecticut's population — extending the reach and legitimacy of citizen voices far beyond the assembly room.

150

Geobica @geobica.com · 28/08/2026

As far as I can tell, this about page was made by pointing an AI agent at this two page pdf ct-citizens-assembly.org/media/CT-Cit... and asking it to make a webpage advertising the plan. Note for instance this line about reaching 10% of CT's population with AI:

From the PDF:
Key Features of the Assembly
1. First large state level, state-endorsed Citizens’ Assembly in the US
2. First large scale attempt using AI tech to link assembly discussion with the larger public,
aiming to reach over 10% of the state’s population
3. Backed by State Comptroller Sean Scanlon, who has the constitutional right to push items
to the legislative agenda on financial issues

150

Geobica @geobica.com · 28/08/2026

The main effect is that when doing Vibe Design, the site can only really have as much meaningful content as the design specification or prompts you gave. I think ct-citizens-assembly.org/about is a good example of how this can go wrong.

160

Geobica @geobica.com · 28/08/2026

Tell #4 is the priorities of design: What was included in this website, in what order was it added, and how much thought was put into each part? These are questions that have somewhat intuitive answers for human designed things, but AI agents show a very different stratigraphy in their design.

170