This is a blogpost (I guess? idk what to call it) that sort of tries to say that LLMs maybe aren't the worst thing ever created. If you are just tired of the infinite discourse about "ai", I...
This is a blogpost (I guess? idk what to call it) that sort of tries to say that LLMs maybe aren't the worst thing ever created. If you are just tired of the infinite discourse about "ai", I completely understand not wanting to see anything more related to it, so to be honest just have a nice day and don't read it. I wrote it after arguing with people about LLMs too much online, which caused me to want to instead organize my thoughts about LLMs in some more productive way. It's probably way too long for the not very revolutionary points it brings. I am not a good writer, but I tried my best for quite a lot of hours to finish it, it probably still has a lot of issues but I could probably keep on rewriting it forever and maybe even just making it worse over time.
The article is as I said too long, but I feel like yapping about something related to this topic a bit more here, I am sorry. I sort of also thought that writing this article is something to do to make me feel like I did more for Timasomo than just make a Ludum Dare game in one weekend. And I am mentioning that because Ludum Dare also reminded me of being conflicted about LLMs too. Basically, the way LD is setup is that to get ratings on your own game you have to rate games other people made. That means I've been playing a lot of random games. Most games are basically the same as all the LDs I did before, some incredible, some... not so incredible. That is completely normal, and every LD I did was like that.
But then there were also a couple that were noticeably "ai" assisted. One didn't even bother to mention it anywhere, which just really annoyed me and made me refuse to play it. According to LD rules you should at minimum opt out of categories you didn't make by yourself, so if you didn't put substantial effort into making the art, or the music, you should opt out of the audio or graphics categories. But this person didn't care at all. And I just don't know what it is, is it that these people are actually just jackasses and always were jackasses? I've seen some people say that it really is "ai" itself that causes this. Anyway, that is alright, I saw 1 game fully slop vibed out with no respect for anyone out of many games, whatever.
The game that actually made me feel really conflicted is a game that I just played. It was honestly quite pretty, I liked quite a lot about it. It had a really good mood. The author of it disclosed that they used "ai" to create the engine for the game. I am ok with that. But where I started being confused is when I saw this middot character (this thing -> "·") in way too many places in the UI of the game. I feel like this thing is such an obvious tell of LLM written text and UIs. I feel like it's much more of a tell than for example em-dashes ever were, I definitely used to see some people use em-dashes normally even before LLMs were a thing. Meanwhile this middot thing I literally never saw anyone use in normal text, literally maybe like ever. Now it's everywhere in every vibecoded UI that exists. But the thing is that the UI and text in the game didn't really feel like stereotypical vibecoded stuff. But seeing that middot thing and also knowing that the author already disclosed they used LLMs elsewhere just made me start thinking about if other stuff might be LLM generated, or even just affected by LLMs. The only character in the game had a name that I feel like is basically what LLMs always pick by default as their very first choice when you ask them to suggest a name of a certain kind. Is it still perfectly plausible that the dev just picked that name? Yes it's definitely possible. In fact I think it's much more likely than that they asked a LLM to help them choose a name. But it just makes me really annoyed at myself that I am thinking about all this. I hate that I am by myself sort of going on this witch hunt in my head when I am playing this random game that I even feel like is one of the better ones I played in this game jam. The creator of the game really could have even used the middots, its not like it isn't a unicode character that exists for a long time, I am guessing someone even here on Tildes will tell me that they used to use it in normal text. And even if they did ask a LLM to help them pick out a name for a character, or even in general write the text, would it be that bad? I just don't know. I really don't know. I also don't really have any point where I am going with this.
I'm a frequent user of em dashes (as well as semicolons, etc.), and I've never used that dot symbol. I've rarely ever even seen it used, to the extent that I didn't recognize it at first and...
I'm a frequent user of em dashes (as well as semicolons, etc.), and I've never used that dot symbol. I've rarely ever even seen it used, to the extent that I didn't recognize it at first and wondered if you were talking about bullet points. But then I remembered that Word uses that dot to represent spaces when you set it to markdown mode.
I wonder why LLMs like to latch onto certain obscure things so much.
Part of my day job involves editing text written by SMEs. Recently, some of the SMEs I work with have started using LLMs to help write the text, and I can very easily tell which parts the LLMs have written and which parts they wrote themselves. I still have to heavily edit both kinds of text; I'm just correcting different problems — more typos and grammatical errors with the human text (quick and easy to spot and fix), more structural issues with the generated text (more challenging to fix, especially since I'm not an expert on the subject matter myself). I hope it saves them time, because it sure doesn't save me any.
I've been using LLMs to help me edit, and I've found they also do much better editing human-written text than LLM-generated text. With editing human text, an LLM can halve the time it takes me to get through the work, but with editing LLM text, I can't tell that there's any time savings to be had. LLMs mostly don't notice the issues I'm seeing, and I have to hand-hold them through the process.
Did you write this? I enjoyed it quite a bit and the points all resonate with my own lived experience. I've sadly become the "AI guy" in most of my social circles since I work at a tech company...
Did you write this?
I enjoyed it quite a bit and the points all resonate with my own lived experience. I've sadly become the "AI guy" in most of my social circles since I work at a tech company and have heavily embraced LLMs because I have to in order to survive in my career field. Plus I'll be honest, its a new technology, and the reason I got into this field is because I like learning about and using new technology. I'm far from a fanboy of the dominant players in the space or the way these things are created & run. But also - that doesn't mean they're not an incredibly useful tool when applied correctly.
It seems impossible to have a nuanced conversation about this topic since the (objectively bad) side effects and ownership model of LLMs make people so emotional. I like how you laid out each emotional reaction and then responded to it in a thoughtful and pragmatic way.
Your Photoshop example resonates with me. We've seen these immorally created tools used to immoral ends, but they don't have to be. Hopefully going forward we'll start to see the pendulum swing towards these tools empowering people to throw off the power structures that led to their immoral creation in the first place.
I use middots all the time in web design. I've always loved them as separator characters over pipes or hyphens. They feel much cleaner — almost elegant. I've never seen an LLM use one before,...
I use middots all the time in web design. I've always loved them as separator characters over pipes or hyphens. They feel much cleaner — almost elegant.
I've never seen an LLM use one before, though, and I don't even know how you'd use them in conversational text. I consider the middot a typographic tool, not a grammatical one.
My kid's school is bringing back handwritten essays and computer labs. You get your lab time to do research and what not, but you'll need to take notes and remember it in order to get it out in a...
In a magical perfect world there'd be enough teachers to be able to pay enough attention to every student to make sure they are doing the work, but we are nowhere near that world.
My kid's school is bringing back handwritten essays and computer labs. You get your lab time to do research and what not, but you'll need to take notes and remember it in order to get it out in a coherent way.
This is much cheaper than buying every kid a Chromebook and then all the software to use it.
Edit: A completely seperate thought, but also don't want to make 5 top-level posts as they come up.
The amoral grounds one in particular is tricky, as I think a lot of the stuff they are doing is the culmination of many of our other moral problems. A talisman of so many other things wrong with society balled up in one, which will then also feed back into the other things that it leverages. Climate change being the single biggest problem. Without a hard demand for 110% capacity carbon-free power production being built onsite to power the datacenter and the community it was thrust upon, it's just undoing much of the progress that has been made. No, building onsite natural gas plants and buying carbon offsets does not count.
This is a blogpost (I guess? idk what to call it) that sort of tries to say that LLMs maybe aren't the worst thing ever created. If you are just tired of the infinite discourse about "ai", I completely understand not wanting to see anything more related to it, so to be honest just have a nice day and don't read it. I wrote it after arguing with people about LLMs too much online, which caused me to want to instead organize my thoughts about LLMs in some more productive way. It's probably way too long for the not very revolutionary points it brings. I am not a good writer, but I tried my best for quite a lot of hours to finish it, it probably still has a lot of issues but I could probably keep on rewriting it forever and maybe even just making it worse over time.
The article is as I said too long, but I feel like yapping about something related to this topic a bit more here, I am sorry. I sort of also thought that writing this article is something to do to make me feel like I did more for Timasomo than just make a Ludum Dare game in one weekend. And I am mentioning that because Ludum Dare also reminded me of being conflicted about LLMs too. Basically, the way LD is setup is that to get ratings on your own game you have to rate games other people made. That means I've been playing a lot of random games. Most games are basically the same as all the LDs I did before, some incredible, some... not so incredible. That is completely normal, and every LD I did was like that.
But then there were also a couple that were noticeably "ai" assisted. One didn't even bother to mention it anywhere, which just really annoyed me and made me refuse to play it. According to LD rules you should at minimum opt out of categories you didn't make by yourself, so if you didn't put substantial effort into making the art, or the music, you should opt out of the audio or graphics categories. But this person didn't care at all. And I just don't know what it is, is it that these people are actually just jackasses and always were jackasses? I've seen some people say that it really is "ai" itself that causes this. Anyway, that is alright, I saw 1 game fully slop vibed out with no respect for anyone out of many games, whatever.
The game that actually made me feel really conflicted is a game that I just played. It was honestly quite pretty, I liked quite a lot about it. It had a really good mood. The author of it disclosed that they used "ai" to create the engine for the game. I am ok with that. But where I started being confused is when I saw this middot character (this thing -> "·") in way too many places in the UI of the game. I feel like this thing is such an obvious tell of LLM written text and UIs. I feel like it's much more of a tell than for example em-dashes ever were, I definitely used to see some people use em-dashes normally even before LLMs were a thing. Meanwhile this middot thing I literally never saw anyone use in normal text, literally maybe like ever. Now it's everywhere in every vibecoded UI that exists. But the thing is that the UI and text in the game didn't really feel like stereotypical vibecoded stuff. But seeing that middot thing and also knowing that the author already disclosed they used LLMs elsewhere just made me start thinking about if other stuff might be LLM generated, or even just affected by LLMs. The only character in the game had a name that I feel like is basically what LLMs always pick by default as their very first choice when you ask them to suggest a name of a certain kind. Is it still perfectly plausible that the dev just picked that name? Yes it's definitely possible. In fact I think it's much more likely than that they asked a LLM to help them choose a name. But it just makes me really annoyed at myself that I am thinking about all this. I hate that I am by myself sort of going on this witch hunt in my head when I am playing this random game that I even feel like is one of the better ones I played in this game jam. The creator of the game really could have even used the middots, its not like it isn't a unicode character that exists for a long time, I am guessing someone even here on Tildes will tell me that they used to use it in normal text. And even if they did ask a LLM to help them pick out a name for a character, or even in general write the text, would it be that bad? I just don't know. I really don't know. I also don't really have any point where I am going with this.
I'm a frequent user of em dashes (as well as semicolons, etc.), and I've never used that dot symbol. I've rarely ever even seen it used, to the extent that I didn't recognize it at first and wondered if you were talking about bullet points. But then I remembered that Word uses that dot to represent spaces when you set it to markdown mode.
I wonder why LLMs like to latch onto certain obscure things so much.
Part of my day job involves editing text written by SMEs. Recently, some of the SMEs I work with have started using LLMs to help write the text, and I can very easily tell which parts the LLMs have written and which parts they wrote themselves. I still have to heavily edit both kinds of text; I'm just correcting different problems — more typos and grammatical errors with the human text (quick and easy to spot and fix), more structural issues with the generated text (more challenging to fix, especially since I'm not an expert on the subject matter myself). I hope it saves them time, because it sure doesn't save me any.
I've been using LLMs to help me edit, and I've found they also do much better editing human-written text than LLM-generated text. With editing human text, an LLM can halve the time it takes me to get through the work, but with editing LLM text, I can't tell that there's any time savings to be had. LLMs mostly don't notice the issues I'm seeing, and I have to hand-hold them through the process.
Did you write this?
I enjoyed it quite a bit and the points all resonate with my own lived experience. I've sadly become the "AI guy" in most of my social circles since I work at a tech company and have heavily embraced LLMs because I have to in order to survive in my career field. Plus I'll be honest, its a new technology, and the reason I got into this field is because I like learning about and using new technology. I'm far from a fanboy of the dominant players in the space or the way these things are created & run. But also - that doesn't mean they're not an incredibly useful tool when applied correctly.
It seems impossible to have a nuanced conversation about this topic since the (objectively bad) side effects and ownership model of LLMs make people so emotional. I like how you laid out each emotional reaction and then responded to it in a thoughtful and pragmatic way.
Your Photoshop example resonates with me. We've seen these immorally created tools used to immoral ends, but they don't have to be. Hopefully going forward we'll start to see the pendulum swing towards these tools empowering people to throw off the power structures that led to their immoral creation in the first place.
I use middots all the time in web design. I've always loved them as separator characters over pipes or hyphens. They feel much cleaner — almost elegant.
I've never seen an LLM use one before, though, and I don't even know how you'd use them in conversational text. I consider the middot a typographic tool, not a grammatical one.
My kid's school is bringing back handwritten essays and computer labs. You get your lab time to do research and what not, but you'll need to take notes and remember it in order to get it out in a coherent way.
This is much cheaper than buying every kid a Chromebook and then all the software to use it.
Edit: A completely seperate thought, but also don't want to make 5 top-level posts as they come up.
The amoral grounds one in particular is tricky, as I think a lot of the stuff they are doing is the culmination of many of our other moral problems. A talisman of so many other things wrong with society balled up in one, which will then also feed back into the other things that it leverages. Climate change being the single biggest problem. Without a hard demand for 110% capacity carbon-free power production being built onsite to power the datacenter and the community it was thrust upon, it's just undoing much of the progress that has been made. No, building onsite natural gas plants and buying carbon offsets does not count.