In the 1980s, economist Robert Solow made an observation that reminded economists of today’s AI boom: “You can see the computer age everywhere but in the productivity statistics.”
An example is yesterday I used Claude a ton for updating an old static site archive, update the caddyfile, set caching, remove old bits of code, update the css on hundreds of html pages, made it so easy
Any time I want to know if a linux command exists to manipulate data, particularly with media conversion and anything involving regex/sed
analysing log files, explaining concepts, I’ve used it to build a massive python script for automating work tasks and more scripts to give me better insight into our monitoring
I’ve use Mistral to generate a wallpaper image of a forest then used https://upscayl.org/ to make it massive so it looks amazing on my 34" ultrawide
I’ve also used it to make my own selfhosted image upload site
Loads and loads and loads of discussions on health, vitamins, strength training routines etcetc
The other day I had a carbonated drink and it upset my gut which is typical (I have IBS) and it found an alternative local soft drink maker who has low sugar drinks which use monk fruit extract instead of the other artificial flavours, went on a 1.5 hour drive west and got myself some:
Those images hosted on the server and code built with claude/mistral (with supabase as the backend)
And yeah it’s WAY better for me, doubt I would have ever found it since I’ve been looking for an IBS friendly soft drink for years
I use LM Studio with different models (Qwen/Gemma/GLM/Mistral) for basic language learning and translations and tier 1 learning javascript
And so much more, claude especially in the last 6 months has kicked it up a gear while Mistral sadly does appear to be falling behind, I’m hoping they catch up soon
edit: I like how 2 people have downvoted me just for explaining what I use ai for, never change lemmy :)
Any time I want to know if a linux command exists to manipulate data
This is one of the reasons I’m very cautious of my genAI use. Most of the time I got to the stage where I thought “there should be a command / way to do this” and I asked an LLM it would tell me the intuitive interface that I expected existed. But when I tested it didn’t exist, the answer had just been creative writing
I didn’t know how but it seemed possible. I searched and didn’t quickly come up with an answer.
I asked an LLM, and it gave me a confident answer.
I checked the man for the tool and the LLM had used creative writing to create the interface I expected should exist… but it did not exist.
I don’t know how you’re swapping or merging these basic facts:
If you’re told the binary and the flag, the validation of the LLMs output is fast (either with man or executing the command)
If you have a process you’re searching for then the searching can be slow to find the combination of commands and flags (and that’s why so many people, by the sounds of it yourself included, use LLMs)
I do think a “linux tutor” is one of the better use cases for LLMs for beginners, since you can quickly validate when its recommended commands are incorrect (but you still can’t quickly validate if its description of internals is misleading). I just think it falls apart as you start requiring more specialised things, or are at a situation where “this should exist” because the LLMs habitually make things up that sound reasonable to fill in the gaps in what the tools can do. That’s not an issue for beginners / the basics especially if there are lots and lots of tutorials the LLMs are sourcing from (although, that opens ethical issues too)
Oh yeah I’m not sure how niche what you were looking for was, for me my stuff is mostly basic, eg I have a massive text file full of subnets from a routing table, I want a command to remove everything but the subnets, i want to convert png to jpg, i want to convert webp to jpg, how do i search a specific value within 10000 log lines of json etc
just stuff that I could search ecosia -> read up on how X works -> guestimate what i’m looking for with trial and error -> get result or ask claude, run command, it works
Anyone with a balanced view on AI automatically attracts downvotes from a specific crowd in here. The problem isn’t that you explained what you use AI for. They couldn’t care less. The problem is that you clearly don’t hate AI with a burning passion.
I can see that there’s clearly some great potential to have fun with a crowd like this. Just haven’t figured out a good method for it yet.
An example is yesterday I used Claude a ton for updating an old static site archive, update the caddyfile, set caching, remove old bits of code, update the css on hundreds of html pages, made it so easy
Any time I want to know if a linux command exists to manipulate data, particularly with media conversion and anything involving regex/sed
analysing log files, explaining concepts, I’ve used it to build a massive python script for automating work tasks and more scripts to give me better insight into our monitoring
I’ve use Mistral to generate a wallpaper image of a forest then used https://upscayl.org/ to make it massive so it looks amazing on my 34" ultrawide
I’ve also used it to make my own selfhosted image upload site
Loads and loads and loads of discussions on health, vitamins, strength training routines etcetc
The other day I had a carbonated drink and it upset my gut which is typical (I have IBS) and it found an alternative local soft drink maker who has low sugar drinks which use monk fruit extract instead of the other artificial flavours, went on a 1.5 hour drive west and got myself some:
https://aussie.zone/post/31756118
Those images hosted on the server and code built with claude/mistral (with supabase as the backend)
And yeah it’s WAY better for me, doubt I would have ever found it since I’ve been looking for an IBS friendly soft drink for years
I use LM Studio with different models (Qwen/Gemma/GLM/Mistral) for basic language learning and translations and tier 1 learning javascript
And so much more, claude especially in the last 6 months has kicked it up a gear while Mistral sadly does appear to be falling behind, I’m hoping they catch up soon
edit: I like how 2 people have downvoted me just for explaining what I use ai for, never change lemmy :)
This is one of the reasons I’m very cautious of my genAI use. Most of the time I got to the stage where I thought “there should be a command / way to do this” and I asked an LLM it would tell me the intuitive interface that I expected existed. But when I tested it didn’t exist, the answer had just been creative writing
i guess use whatever llm you’re using then use claude to double check its work
Ahhh… what!?
Its linux, its faster to man the command or to type it. Point is that it usefulness to bullshit ratio for me in that situation was far too low
if you can lookup an obscure command and work out the regex faster than a 10 second question to claude I don’t know what we’re discussing
I have wanted to transform things
I didn’t know how but it seemed possible. I searched and didn’t quickly come up with an answer.
I asked an LLM, and it gave me a confident answer.
I checked the man for the tool and the LLM had used creative writing to create the interface I expected should exist… but it did not exist.
I don’t know how you’re swapping or merging these basic facts:
I do think a “linux tutor” is one of the better use cases for LLMs for beginners, since you can quickly validate when its recommended commands are incorrect (but you still can’t quickly validate if its description of internals is misleading). I just think it falls apart as you start requiring more specialised things, or are at a situation where “this should exist” because the LLMs habitually make things up that sound reasonable to fill in the gaps in what the tools can do. That’s not an issue for beginners / the basics especially if there are lots and lots of tutorials the LLMs are sourcing from (although, that opens ethical issues too)
Oh yeah I’m not sure how niche what you were looking for was, for me my stuff is mostly basic, eg I have a massive text file full of subnets from a routing table, I want a command to remove everything but the subnets, i want to convert png to jpg, i want to convert webp to jpg, how do i search a specific value within 10000 log lines of json etc
just stuff that I could search ecosia -> read up on how X works -> guestimate what i’m looking for with trial and error -> get result or ask claude, run command, it works
Lemmy is indeed a very special place. :D
Anyone with a balanced view on AI automatically attracts downvotes from a specific crowd in here. The problem isn’t that you explained what you use AI for. They couldn’t care less. The problem is that you clearly don’t hate AI with a burning passion.
I can see that there’s clearly some great potential to have fun with a crowd like this. Just haven’t figured out a good method for it yet.