I used Gemini (the Google AI) twice at work today, asking about Google Workspace configuration and Google Cloud CLI usage (because we use those a lot). You’d think that it’d be well-suited for those topics. It answered very confidently, yet completely wrong. Just wrong. Made-up CLI arguments, whatever. It took me a while to notice, though, because it’s so convincing and, well, you implicitly and subconsciously trust the results of the Google AI when asking about Google topics, don’t you?

Will it get better over time? Maybe. But what I really want is this:

• Good, well-structured, easy-to-read, proper documentation. Google isn’t doing too bad in this regard, actually, it’s just that they have so much stuff that it’s hard to find what you’re looking for. Hence …
  
• … I want a good search function. Just give me a good fuzzy search for your docs. That’s it.
  

I just don’t have the time or energy to constantly second-guess this stuff. Give me something reliable. Something that is designed to do the right thing, not toy around with probabilities. “AI for everything” is just the wrong approach.

@bender@twtxt.net to work through both https and gemini, the site is not written in HTML, but in Gemtext, automatically converted to HTML, when needed. Gemtext is nicely explained for example here: https://garden.bouncepaw.com/hypha/gemtext . In short, it is so limited, no line can be more than one thing, so no links in a list are possible, othar than doing it through something like this primitive workaround.

Just a small update, on my birthday (on the 5th), I accidentally deleted the main page, of my website, so I’m using that as an opportunity, to try something new, at https://thecanine.smol.pub or gemini://thecanine.smol.pub - depending on your preferred protocol.

Any feedback is welcome!

… and now I just read @bender@twtxt.net’s other post that said the Gemini text was a shortened version, so I might have criticized things that weren’t true for the full version. Okay, sorry, I’m out. (And I won’t play that game, either. Don’t send me another AI output, possibly tweaked to address my criticism. That is besides the point and not worth my time.)

@prologic@twtxt.net Let’s go through it one by one. Here’s a wall of text that took me over 1.5 hours to write.

This section says AI should not be treated as an authority. This is actually just what I said, except the AI phrased/framed it like it was a counter-argument.

The AI also said that users must develop “AI literacy”, again phrasing/framing it like a counter-argument. Well, that is also just what I said. I said you should treat AI output like a random blog and you should verify the sources, yadda yadda. That is “AI literacy”, isn’t it?

My text went one step further, though: I said that when you take this requirement of “AI literacy” into account, you basically end up with a fancy search engine, with extra overhead that costs time. The AI missed/ignored this in its reply.

Okay, so, the AI also said that you should use AI tools just for drafting and brainstorming. Granted, a very rough draft of something will probably be doable. But then you have to diligently verify every little detail of this draft – okay, fine, a draft is a draft, it’s fine if it contains errors. The thing is, though, that you really must do this verification. And I claim that many people will not do it, because AI outputs look sooooo convincing, they don’t feel like a draft that needs editing.

Can you, as an expert, still use an AI draft as a basis/foundation? Yeah, probably. But here’s the kicker: You did not create that draft. You were not involved in the “thought process” behind it. When you, a human being, make a draft, you often think something like: “Okay, I want to draw a picture of a landscape and there’s going to be a little house, but for now, I’ll just put in a rough sketch of the house and add the details later.” You are aware of what you left out. When the AI did the draft, you are not aware of what’s missing – even more so when every AI output already looks like a final product. For me, personally, this makes it much harder and slower to verify such a draft, and I mentioned this in my text.

You, @prologic@twtxt.net, also mentioned this in your car tyre example.

In my text, I gave two analogies: The gym analogy and the Google Translate analogy. Your car tyre example falls in the same category, but Gemini’s calculator example is different (and, again, gaslight-y, see below).

What I meant in my text: A person wants to be a programmer. To me, a programmer is a person who writes code, understands code, maintains code, writes documentation, and so on. In your example, a person who changes a car tyre would be a mechanic. Now, if you use AI to write the code and documentation for you, are you still a programmer? If you have no understanding of said code, are you a programmer? A person who does not know how to change a car tyre, is that still a mechanic?

No, you’re something else. You should not be hired as a programmer or a mechanic.

Yes, that is “skill evolution” – which is pretty much my point! But the AI framed it like a counter-argument. It didn’t understand my text.

(But what if that’s our future? What if all programming will look like that in some years? I claim: It’s not possible. If you don’t know how to program, then you don’t know how to read/understand code written by an AI. You are something else, but you’re not a programmer. It might be valid to be something else – but that wasn’t my point, my point was that you’re not a bloody programmer.)

Gemini’s calculator example is garbage, I think. Crunching numbers and doing mathematics (i.e., “complex problem-solving”) are two different things. Just because you now have a calculator, doesn’t mean it’ll free you up to do mathematical proofs or whatever.

What would have worked is this: Let’s say you’re an accountant and you sum up spendings. Without a calculator, this takes a lot of time and is error prone. But when you have one, you can work faster. But once again, there’s a little gaslight-y detail: A calculator is correct. Yes, it could have “bugs” (hello Intel FDIV), but its design actually properly calculates numbers. AI, on the other hand, does not understand a thing (our current AI, that is), it’s just a statistical model. So, this modified example (“accountant with a calculator”) would actually have to be phrased like this: Suppose there’s an accountant and you give her a magic box that spits out the correct result in, what, I don’t know, 70-90% of the time. The accountant couldn’t rely on this box now, could she? She’d either have to double-check everything or accept possibly wrong results. And that is how I feel like when I work with AI tools.

Gemini has no idea that its calculator example doesn’t make sense. It just spits out some generic “argument” that it picked up on some website.

The AI makes two points here. The first one, I might actually agree with (“bad bot behavior is not the fault of AI itself”).

The second point is, once again, gaslighting, because it is phrased/framed like a counter-argument. It implies that I said something which I didn’t. Like the AI, I said that you would have to adjust the copyright law! At the same time, the AI answer didn’t even question whether it’s okay to break the current law or not. It just said “lol yeah, change the laws”. (I wonder in what way the laws would have to be changed in the AI’s “opinion”, because some of these changes could kill some business opportunities – or the laws would have to have special AI clauses that only benefit the AI techbros. But I digress, that wasn’t part of Gemini’s answer.)

Except for one point, I don’t accept any of Gemini’s “criticism”. It didn’t pick up on lots of details, ignored arguments, and I can just instinctively tell that this thing does not understand anything it wrote (which is correct, it’s just a statistical model).

And it framed everything like a counter-argument, while actually repeating what I said. That’s gaslighting: When Alice says “the sky is blue” and Bob replies with “why do you say the sky is purple?!”

But it sure looks convincing, doesn’t it?

This took so much of my time. I won’t do this again. 😂

@prologic@twtxt.net when I first “fed” the text to Gemini, I asked for a three paragraphs summary. It provided it. Then I asked to “elaborate on three areas: user experience, moral/political impact, and technical/legal concerns”. The reply to that is too long for a twtxt.

I then asked to counter the OP opinions—as in “how would you counter the author’s opinion?”. The reply was very long, but started like this:

“That’s an excellent question, as the post lays out some very strong, well-reasoned criticisms. Countering these points requires acknowledging the valid concerns while presenting a perspective focused on mitigation, responsible integration, and the unique benefits of AI.”

What followed was extensive, so I asked for a summary, which didn’t do justice to the wall of text that preceded it.

@movq@www.uninformativ.de I am genuinely curious as to why you think Geminis summarization and the categorization of your gopher post was and is as you say misunderstood?

I asked this very genuinely because before reading @bender@twtxt.net’s comments and Gemini summarization I actually went and unplugged your post into flood gaps go for proxy, and then listen to the text intently with my own human ears 👂

@movq@www.uninformativ.de this I find more worrisome, and saw no mention of it on your text: Right-Wing Chatbots Turbocharge America’s Political and Cultural Wars (gift article).

Enoch, one of the newer chatbots powered by artificial intelligence, promises “to ‘mind wipe’ the pro-pharma bias” from its answers. Another, Arya, produces content based on instructions that tell it to be an “unapologetic right-wing nationalist Christian A.I. model.”

@movq@www.uninformativ.de Gemini liked your opinion very much. Here is how it countered:

The criticism of AI as untrustworthy is a problem of misapplication, not capability.

• AI as a Force Multiplier: AI should be treated as a high-speed drafting and brainstorming tool, not an authority. For experts, it offers an immense speed gain, shifting the work from slow manual creation to fast critical editing and verification.
  
• The Rise of AI Literacy: Users must develop a new skill—AI literacy—to critically evaluate and verify AI’s probabilistic output. This skill, along with improving citation features in AI tools, mitigates the “gaslighting” effect.
  

The fear of skill loss is based on a misunderstanding of how technology changes the nature of work; it’s skill evolution, not erosion.

• Shifting Focus to High-Level Skills: Just as the calculator shifted focus from manual math to complex problem-solving, AI shifts the focus from writing boilerplate code to architectural design and prompt engineering. It handles repetitive tasks, freeing humans for creative and complex challenges.
  
• Accessibility and Empowerment: AI serves as a powerful democratizing tool, offering personalized tutoring and automation to people who lack deep expertise. While dependency is a risk, this accessibility empowers a wider segment of the population previously limited by skill barriers.
  

The legal and technical flaws are issues of governance and ethical practice, not reasons to reject the core technology.

• Need for Better Bot Governance: Destructive scraping is a failure of ethical web behavior and can be solved with better bot identification, rate limits, and protocols (like enhanced robots.txt). The solution is to demand digital citizenship from AI companies, not to stop AI development.

Continents
⌘ Read more

For those not on Gemini, a proxy provides more insides on such, hmm, interesting acronym. :-D

I think I’m gonna participate in ROOPHLOCH this year: gemini://zaibatsu.circumlunar.space/~solderpunk/gemlog/announcing-roophloch-2025.gmi

So, in addition to HTTPS and Gemini, my twtxt should now also be available over Gopher (gopher://hashnix.club:70/0/~dce/twtxt.txt). Not sure who, if anyone, would need this; but since my tilde provides Gopher hosting, I’d may as well mirror my twtxt there as well.

@dce@hashnix.club I don’t use Gemini, but I follow you on the good, old, HTTP(S)! :-)

Assuming I configured this right, my twtxt should now also be available over Gemini!

Here’s an interesting thought/angle on this topic:

gemini://gemini.conman.org/boston/2025/08/21.1

A further check showed that all the network blocks are owned by one organization—Tencent [4]. I’m seriously thinking that the CCP (Chinese Communist Party) encourage this with maybe the hope of externalizing the cost of the Great Firewall [5] to the rest of the world.

Really, it won’t be long until I give the world the finger and move everything behind Gopher or Gemini. It’ll be a while until the bots find me there.

When I chose the MIT license for all of my software, I thought:

“Should I use GPL, which I don’t really understand? Is that worth it? Yeah, there is a theoretical possibility that some company might use my code in their proprietary product … and then what? Should I sue them to enforce the GPL? I’m not going to do that anyway, so I’ll just use the MIT license.”

And now we have those LLM scrapers and now it’s suddenly a reality that these companies (ab)use my code. I can see it in my logs. I didn’t expect that back then.

GPL wouldn’t help, either, of course. (Regardless, I now think that GPL would have been the better choice anyway.)

I’m honestly considering taking my code and website offline. Maybe make it accessible through some obscure protocol like Gopher or Gemini, but no more HTTP.

(Yes, Anubis might help. Temporarily.)

I’m just tired.

Farrrk me Google search is and these days. Will they please “fuck off” with this Gemini AI garbage at the top that takes forever and is distracting as shit™ 💩 Fark me 🤦‍♂️ #Google #Search #Sucks #AI #Gemini

it is gemini:// actually

i feel like i’ve stumbled upon an alternative internet universe, gemini protocol, the small internet and a counter-cultural world!

@prologic@twtxt.net OH SHIT using this for a protocol like gopher is smart! might have to try that for gemini so i don’t have to keep a port open for that

Gemini 2.5 Flash
Comments ⌘ Read more

Generate videos in Gemini and Whisk with Veo 2
Comments ⌘ Read more

Google announces Sec-Gemini v1 a new experimental cybersecurity model
Comments ⌘ Read more

Gemini 2.5 gets 24.4% on MathArena USAMO beating previous top score of 4.7%
Comments ⌘ Read more

We hacked Google’s A.I Gemini and leaked its source code (at least some part)
Comments ⌘ Read more

@bender@twtxt.net thinked about Gemini protocol. Why corporations shit this name with cryptocurrency and LLMs?

@prologic@twtxt.net Gemini has an answer for you:

This is a conversation thread from a twtxt network, detailing a user’s (movq) frustration with the Mastodon “export data” feature and their consideration of self-hosting a fediverse alternative. Here’s a summary:

• movq’s initial issue:
  
  • movq is concerned about the volatility of their data on their current Mastodon instance due to a broken “export data” feature.
    
  • They contacted the admins, but the issue remains unresolved.
    
  • This led them to contemplate self-hosting.
    
• Alternative fediverse software suggestions:
  
  • kat suggests gotosocial as a lightweight alternative to Mastodon.
    
  • movq agrees, and also mentions snac as a potential option.
    
• movq’s change of heart:
  
  • movq ultimately decides that self-hosting any fediverse software, besides twtxt, is too much effort.
    
• Resolution and compromise:
  
  • The Mastodon admins attribute the export failure to the size of movq’s account.
    
  • movq decides to set their Mastodon account to auto-delete posts after approximately 180 days to manage data size.
    
  • Movq also mentions that they use auto-expiring links on twtxt to reduce data storage.

well, I assume by syntax you mean Gemtext (which I like a lot, my personal blog is built on top of it), so I think it might work for twtxt clients…

I knew of twtxt in Gemini Antenna, so at least the 2017 spec might work on that protocol. I think the main issue with extensions is that they weren’t designed with many URLs and protocols in mind.

Also I have to admit that the Gemini community significantly reduced in the last few years. I don’t know how worth it is to add support for Gemini now.

@eapl.me@eapl.me I agree. The syntax is weird inside Gemini and twtxt is made with the http protocol in mind and Gemini doesn’t work with some extensions.

Timeline and twtxt-php, don’t support Gemini, only HTTP/S, as a design choice (although originally it was intended to work on Gemtext, it was a niche inside a niche, so it was discarded very soon).

At the moment of building the engine there weren’t many Gemini URLs supporting twtxt 1.1 (with twtxt.dev extensions).
Also User-Agent won’t work there, and many Gemini URLs are a mirror of the HTTP one, so I think is not strictly necessary.

my 2c

Are there any clients to read gemini?

My twtxt feed is now also available at gemini://roccodrom.de/twtxt.txt

Google, DuckDuckGo massively expand “AI” search results
Clearly, online search isn’t bad enough yet, so Google is intensifying its efforts to continue speedrunning the downfall of Google Search. They’ve announced they’re going to show even more “AI”-generated answers in Search results, to more people. Today, we’re sharing that we’ve launched Gemini 2.0 for AI Overviews in the U.S. to help with harder questions, starting with coding, advanced math and multimodal queries, with mor … ⌘ Read more

well, Gemini clients like Lagrange allow to show inline images when you click on an image link. Text based clients, like Amfora, usually allow to watch the image in another ‘window’.

For example here: gemini://text.eapl.mx/en-making-a-tic-tac-toe-variant and there https://text.eapl.mx/en-making-a-tic-tac-toe-variant

I agree that some topics require images to make it easier to explain.

@andros@twtxt.andros.dev Broke on me for having alt-urls I think 🥲

twtxt---profile-layout: Wrong type argument: char-or-string-p, ("https://aelaraji.com/twtxt.txt" "gemini://box.aelaraji.com/twtxt.txt" "gopher://box.aelaraji.com/0/twtxt.txt")

(I keep thinking that going back go Gopher or Gemini might be a good idea at this point. They don’t care about that, probably. 🫣)

messing with gemini again, this time a static site generator called gssg - https://git.sr.ht/~gsthnz/gssg

my capsule is linked in my profile but just in case it’s over at gemini://lazuli.sayitditto.net

Portion of the modified Twitter TOS that goes into effect today (it’s on right now), as summarised (ironically) by Google’s Gemini:

“In simpler terms, this means that when you share your content (like text, images, or videos) on the service, you’re giving the company permission to use it in various ways. They can copy, modify, distribute, and even use it to train their AI models. This includes sharing your content with others and using it on other platforms. You won’t be paid for this, but using the service itself is considered enough compensation.”

Thank you, @eapl.me@eapl.me! No need to apologize in the introduction, all good. :-)

Section 3: I’m a bit on the fence regarding documenting the HTTP caching headers. It’s a very general HTTP thing, so there is nothing special about them for twtxt. No need for the Twtxt Specification to actually redo it. But on the other hand, a short hint could certainly help client developers and feed authors. Maybe it’s thanks to my distro’s Ngninx maintainer, but I did not configure anything for the Last-Modified and ETag headers to be included in the response, the web server just already did it automatically.

The more that I think about it while typing this reply, the more I think your recommendation suggestion is actually really great. It will definitely beneficial for client developers. In almost all client implementation cases I’d say one has to actually do something specifically in the code to send the If-Modified-Since and/or If-None-Match request headers. There is no magic that will do it automatically, as one has to combine data from the last response with the new request.

But I also came across feeds that serve zero response headers that make caching possible at all. So, an explicit recommendation enables feed authors to check their server setups. Yeah, let’s absolutely do this! :-)

Regarding section 4 about feed discovery: Yeah, non-HTTP transport protocols are an issue as they do not have User-Agent headers. How exactly do you envision the discovery_url to work, though? I wouldn’t limit the transports to HTTP(S) in the Twtxt Specification, though. It’s up to the client to decide which protocols it wants to support.

Since I currently rely on buckket’s twtxt client to fetch the feeds, I can only follow http(s):// (and file://) feeds. But in tt2 I will certainly add some gopher:// and gemini:// at some point in time.

Some time ago, @movq@www.uninformativ.de found out that some Gopher/Gemini users prefer to just get an e-mail from people following them: https://twtxt.net/twt/dikni6q So, it might not even be something to be solved as there is no problem in the first place.

Section 5 on protocol support: You’re right, announcing the different transports in the url metadata would certainly help. :-)

Section 7 on emojis: Your idea of TUI/CLI avatars is really intriguing I have to say. Maybe I will pick this up in tt2 some day. :-)

@sorenpeter@darch.dk on 4 for gemini if your TLS client certificate contains your nick@host could that work for discovery?

@eapl.me@eapl.me here are my replies (somewhat similar to Lyse’s and James’)

1. Metadata in twts: Key=value is too complicated for non-hackers and hard to write by hand. So if there is a need then we should just use #NSFS or the alt-text file in markdown image syntax ![NSFW](url.to/image.jpg) if something is NSFW

2. IDs besides datetime. When you edit a twt then you should preserve the datetime if location-based addressing should have any advantages over content-based addressing. If you change the timestamp the its a new post. Just like any other blog cms.

3. Caching, Yes all good ideas, but that is more a task for the clients not the serving of the twtxt.txt files.

4. Discovery: User-agent for discovery can become better. I’m working on a wrapper script in PHP, so you don’t need to go to Apaches log-files to see who fetches your feed. But for other Gemini and gopher you need to relay on something else. That could be using my webmentions for twtxt suggestion, or simply defining an email metadata field for letting a person know you follow their feed. Interesting read about why WebMetions might be a bad idea. Twtxt being much simple that a full featured IndieWeb sites, then a lot of the concerns does not apply here. But that’s the issue with any open inbox. This is hard to solve without some form of (centralized or community) spam moderation.

5. Support more protocols besides http/s. Yes why not, if we can make clients that merge or diffident between the same feed server by multiples URLs

6. Languages: If the need is big then make a separate feed. I don’t mind seeing stuff in other langues as it is low. You got translating tool if you need to know whats going on. And again when there is a need for easier switching between posting to several feeds, then it’s about building clients with a UI that makes it easy. No something that should takes up space in the format/protocol.

7. Emojis: I’m not sure what this is about. Do you want to use emojis as avatar in CLI clients or it just about rendering emojis?

due to the gemini-centric nature of my setup, I don’t get webmentions. I just scrape the network and grep. maybe my aggregator will produce notifications at some point lol

@codebuzz@www.codebuzz.nl I have some shell scripts that handle some of the log formatting details, but I mostly write my mesages by hand. Lately I’ve been browsing twtxt.net since they aggregate most of the known network. I have a couple of demo aggregators sitting around, but I’m in the middle of some infra rebuilds so a lot of my services are offline rn. They’re both built on a simple social graph analysis that extracts urls for your direct follows the follows listed on each of those feeds (friend-of-a-friend replication). certain formatting operations are awkward with my setup, so I may write an app of some kind in the future. likely gemini-based, but I have a number of projects ahead of that one in the queue.

Simplified twtxt - I want to suggest some dogmas or commandments for twtxt, from where we can work our way back to how to implement different feature like replies/treads:

1. It’s a text file, so you must be able to write it by hand (ie. no app logic) and read by eye. If you edit a post you change the content not the timestamp. Otherwise it will be considered a new post.

2. The order of lines in a twtxt.txt must not hold any significant. The file is a container and each line an atomic piece of information. You should be able to run sort on a twtxt.txt and it should still work.

3. Transport protocol should not matter, as long as the file served is the same. Http and https are preferred, so it is suggested that feed served via Gopher or Gemini also provide http(s).

4. Do we need more commandments?

Google tool makes AI-generated writing easily detectable
Google DeepMind has been using its AI watermarking method on Gemini chatbot responses for months – and now it’s making the tool available to any AI developer ⌘ Read more

@asquare@asquare.srht.site Hi back at you!! 👋 and nice to meet you!

thank you gemini for continuing to inspire me to think smol

similar to data packets in NDN, each message has multiple names. a true name, which is an encoded cryptographic hash of the file itself. we call this kind of information self-certifying. given a true name, you can find a file and verify its integrity. additionally, agents can associate a self-certifying name with a pet name or subjective label of their choosing and share it with their friends/peers. zoko’s triangle can suck it. gemini://sunshinegardens.org/~xjix/wiki/cryptogen–specification/