Technology

Regex

I don't use AI for anything. I consider the LLMs pretty useless since they are prone to spewing BS.
I would probably play around with stable diffusion if I had a GPU that would run it at a reasonable speed though.

Guitar amp and pedal modeling.

Sometimes it's helpful if I'm having trouble making a specific excel formula

I use it to ask questions that I can't find search results for or don't have the words to ask. Also for d&d character art I share with my playgroup lol.

if you know programming then you can have it do basic stuff (or even mid complexity stuff if you do it step by step). Just the other day I directed it to produce a code in js using 3js which does scatter plots. The code did run into a couple issues which it was able to solve itself when pointed by me. There was only one problem it could not solve despite several attempts (having a grid does not move with camera controls) so I had to figure that out myself. It was pretty impressive. Overall an expert in 3js would do that maybe in 10 minutes, it took me a couple hours. If I had to do it via searching online it would probably take me a couple of days since I know nought about js.

I also had it write bash scripts a couple of times. It is generally pretty good with writing basic stuff and piecing them together especially if you know programming so you can check it and write intelligible prompts about problems in the code.

1 Get random error or have other tech issue

2 Certainly private search engines will be able to find a solution (they cannot)

3 Certainly non private search engines can find the solution (they can not)

4 "Chat GPT, the heck is this [error code or something]" Then usually I get a correct and well explained answer.

I've installed immich on my DIY NAS, it has ml face/object recognition and it works nice

I sometimes have Señor GPT rewrite my nonsensical ramblings into coherent and decipherable text. I recently did that for a paper in my last class. lol
I wrote a bunch of shit, had GPT rewrite it, added a couple quotes from my sources and called it a day.

I'm also currently on a single player, open world adventure with GPT. Myself and the townspeople iust confronted the suspicious characters on the edge of town. They claim to not be baddies but they're being super sus. I might just attack anyway.

I would generally say they're great with anything you happy being 100% right 90% of the time.

Ansible.

I fucking hate YAML, and I hate Ansible 'programming' (see "HTML 'programming' language" for rage context).

Chatgpt - I'll use the one in bing or the one in regular-skype - feeds me stuff I can copy/paste/review, and I can get on with my day having lost fewer brain cells to the rage of existing in a world with Ansible fanboys who seem to have forgotten there is NOTHING Ansible does now that we weren't doing in 2003 ... and that the state of the art is 2 generations PAST that glorified mess.

Having used puppet and chef and seen mgmtconfig, I can only applaud RedHat for going with the worst-of-two options and promoting it so hard it appeared viable.

I don't mean to dunk on Michael. Just, James' idea was way better and RH still went with Michael's, and I one day need to know whether the person who had the final say got help.

I make porn

But I also used it In tech debug, but less and less.

One use-case for me has been converting code from a language I know to a language I don't. Usually, just small snippets. The code is usually full of holes, but I'm good enough with the logic to duct tape those puppies!

Entertainment.

Customer support tier .5

It can be hella great for finding what you need on a big website that is poorly organized, laid out, or just enormous in content. I could see it being incredible for things like irs.gov, your healthcare providers website, etc. in getting the requested content in user hands without them having to familiarize themselves with constantly changing layouts, pages, branding, etc.

To go back to the IRS example, there are websites in the last 5 years that started to have better content library search functionality, but I guess for me having AI able to contextualize the request and then get you what you want specifically would be incredible. "Tax rule for x kind of business in y situation for 2024"---that shit takes hours if you're pretty competent sometimes, and current websites might just say "here is the 2024 tax code PLOP" or "here is an answer that doesn't apply to your situation" etc. "tomato growing tips for zone 3a during drought" on a gardening site, etc.

I'm in HR so benefits are a big one...the absolute mountain of content, even if you understand it, even experts can't have perfect recall and quick, easy answers through a mountain of text seems like an area AI could deliver real value.

That said, companies using AI as an excuse to them eliminate support jobs because customers "have AI" are greedy dipshits as AI and LLMs are a risk at best and outside of a narrow library and intense testing are going to always be more work for the company as you not only have to fix the wrong answer situations but also get the right answer the old fashioned way. You still need humans and hopefully AI can make their work more interesting, nuanced and fulfilling.

I think it's mischaracterising the argument against AI to boil it down to "AI is useless" (and I say that as much as a criticism of those who are critical of genAI as I do of those who want to defend it; far too many people express the argument reductively as "AI is useless" when that's not exactly what's really being meant).

The problem is not that genAI is never useful for anything. It is sometimes useful for some things. The problem is that being sometimes useful for some things does not remotely justify what the technology costs. I mean that both on the macro scale - untold climate damage, vast amounts of wasted resources - and on the micro scale; OpenAI alone loses $2.35 for every $1.00 they make.

That is fundamentally unsustainable. If you like genAI for whatever use cases you've found for it, and you really don't care about the climate toll and other externalities, then you can look forward to paying upwards of $50-$100 a month to actually use it, once we're out of the "Give it to 'em cheap/free to get' em hooked" phase, because that's what it'll take to make these models profitable. In fact that's kind of a lowball estimate.

I know plenty of people who find this tech occasionally useful as a way of searching for the answer to a question or producing a small snippet of code, but I can't imagine anyone who finds those uses so compelling that they'd throw "Canadian cell phone contract" levels of money at it.

So far, I've only found it really useful for two things. One is generating text, where I've found using an LLM to generate a title based on a given piece of text is more effective than using other summarisation models, especially for a short piece of text.

I've also found it okay for basic, generic scripts, like trying to figure out what the equivalent Powershell commands for a bash script would be to do something quick, rather than try and learn it from scratch.

The useless/useful dichotomy is kinda misguided because that judgement will almost always depend on cost and we don’t have a good understanding of actual costs of running these models. I have copilot enabled in my IDE, and it saves me from a few searches here and there and autocompletes stuff that would have taken me some time to type. So not exactly useless, but right now it’s being payed for by VC’s who expect a return on their investment, so what does that look like? Before we know, it’s hard to say whether these things are relevant

making my tone proper in emails

I use a lot of AI/DL-based tools in my personal life and hobbies. As a photographer, DL-based denoising means I can get better photos, especially in low light. DL-based deconvolution tools help to sharpen my astrophotos as well. The deep learning based subject tracking on my camera also helps me get more in focus shots of wildlife. As a birder, tools like Merlin BirdID's audio recognition and image classification methods are helpful when I encounter a bird I don't yet know how to identify.

I don't typically use GenAI (LLMs, diffusion models) in my personal life, but Microsoft Copilot does help me write visualization scripts for my research. I can never remember the right methods for visualization libraries in Python, and Copilot/ChatGPT do a pretty good job at that.

When I need to make a joke about how inept AI is, I'll use AI to capture an example of it saying the most efficient way to get to the moon is to put a 2 liter bottle of coke in your asshole, wide end first, remove the cap and immediately sit on an opened sleeve of mentos.

I mainly use it if I want to rephrase text passages and to correct the grammar and spelling of texts. However, I only use it when writing important assignemts for choop/university.

Currently, mainly just cooking.

In the future, I'm hoping to leverage it to create video content. I've actually been disappointed in its usefulness for writing sci-fi, it tends to want to argue. But based on the surreal images that it can created I am hoping that can be translated into creating 3D scenes that can be used to extract video.