I’m still looking for valuable use-cases for LLMs. I’ve used them for low-stakes things such as a proof reader, a thesaurus1, some coding of bash scripts, HTML, and CSS (including for this website). But these are all things I can do myself. I trust the output of the LLM only because I have enough expertise and experience that I can verify the result. And if gives me a bunch of crap, I can come up with a solution myself.
My concern is that if/when I do find a valuable use-case that the big players in LLMs will enshittify their products around that same time. I’m going into the “AI revolution” with open eyes. I know that if I get any value from it, the endgame is to extract that value (and more) back from me.
So I installed Ollama on my M1 MacBook Air.
I’ve been comparing prompts of Ollama using the model Qwen (qwen3.5:4b) to the free plan of ChatGPT. So far the answers come out mostly the same, which is really encouraging.
Free tools tend to lag behind the cutting edge consumer tools, but I’m pleasantly surprised that Ollama is so close to commercial offerings. Oh, I’m sure it’s substantially worse by some measure that matters to people who are close to this stuff, but that’s not me. I don’t rely on LLMs for anything, so if I have access to a model that’s like ChatGPT from 2-3 years ago, that’s good enough for now.
Next I’d like to try Open WebUI, which is a web front-end for Ollama that gives an interface like the usual LLM products. It seems to offer some additional features and just uses Ollama as infrastructure.
1: Using LLMs as a thesaurus is by far the best use-case I’ve found. It plays to the fundamental strengths of the software because they are models of language based on large samples of writing. I can ask for a word based on feeling, tone, connotation, etc. I can say I want a word that is like (or not like) another word. As a model of words and their relationship to other words, it makes a very good thesaurus. ↩
I’m still looking for valuable use-cases for LLMs. I’ve used them for low-stakes things such as a proof reader, a thesaurus1, some coding of bash scripts, HTML, and CSS (including for this website). But these are all things I can do myself. I trust the output of the LLM only because I have enough expertise and experience that I can verify the result. And if gives me a bunch of crap, I can come up with a solution myself.
My concern is that if/when I do find a valuable use-case that the big players in LLMs will enshittify their products around that same time. I’m going into the “AI revolution” with open eyes. I know that if I get any value from it, the endgame is to extract that value (and more) back from me.
So I installed Ollama on my M1 MacBook Air.
I’ve been comparing prompts of Ollama using the model Qwen (qwen3.5:4b) to the free plan of ChatGPT. So far the answers come out mostly the same, which is really encouraging.
Free tools tend to lag behind the cutting edge consumer tools, but I’m pleasantly surprised that Ollama is so close to commercial offerings. Oh, I’m sure it’s substantially worse by some measure that matters to people who are close to this stuff, but that’s not me. I don’t rely on LLMs for anything, so if I have access to a model that’s like ChatGPT from 2-3 years ago, that’s good enough for now.
Next I’d like to try Open WebUI, which is a web front-end for Ollama that gives an interface like the usual LLM products. It seems to offer some additional features and just uses Ollama as infrastructure.
1: Using LLMs as a thesaurus is by far the best use-case I’ve found. It plays to the fundamental strengths of the software because they are models of language based on large samples of writing. I can ask for a word based on feeling, tone, connotation, etc. I can say I want a word that is like (or not like) another word. As a model of words and their relationship to other words, it makes a very good thesaurus. ↩