TechTakes

My problem is that it's not that fucking useful. I got the Pixel 9 specifically because of its advertised AI chip for the assistant and I swear it's just gotten worse since the Pixel 7. I used to be able to ask Google anything through the assistant, and now 90% of my questions are answered with "can't find the information."

They also advertised (or at least heavily alluded to) the use of the AI chip when you are in low network areas but it works just as good outside of 4g+ coverage as it ever did without the stupid chip.

Whats the point of adding AI branded nonsense if there's no practical use for it. And that doesn't even start to cover the issues with AI's reliability as a source of information. Garbage in = garbage out.

That's not fair! I care! A lot!

Just had to buy a new laptop for new place of employment. It took real time, effort, and care, but I've finally found a recent laptop matching my hardware requirements and sense of aesthetics at a reasonable price, without that hideous copilot button :)

The average technical person realises ai is shit.
The average non-technical person doesn't need an ai computer, because chatgpt is free.

they are useless... Copilot is not worth even $50 of an update

I will not go into ethics.

AI as in the current LLMs is useful. But it has to be used for the right things. It is not a search engine. It is not perfect at extracting information from text.

I am not a fan of AI "art", but I do actually find some AI music tolerable.

AI "art" has gotten worse in my opinion. It's more "correct", but has this nasty glow to it. I liked it better with the jank and seven fingers. It did not have the same AI slop glowy filter.

Development It helps speed up development, not by giving me code I can use necessarily. But because it shows me a path forward. I don't get stuck on tasks that should be easy anymore. I don't have to think long for it to suggest me a way to map x object to y datatype.

It's also useful for writing unit tests.

Summarization I use it to summarize articles with a custom prompt. While not perfect, it helps me decide if this is clickbait/interesting or not. I use the summaries to decide if I want to read the article or not.

Do you see any obvious issues with this? Give it a plan or something and ask if it sees any issues. It will come up with tons of bullshit and a few useful issues sometimes. It is like a sparring partner.

AI Dungeon AI Dungeon is fun, it was even fun when GPT2 was a thing, but now it's a lot better.

Low effort image recognition Set up a camera, send a picture regularly and ask "Is X in photo? Yes/No"

(I know there is better ways of doing this, and even great pretrained machine learning APIs for it)

HomeAssistant I have these sensors (give list of sensors). I want an automation that does x, y, z. It writes a working one in seconds.

Math Yes, it is fairly good at math. You can't trust it, but if you know a fair bit of math and know how to verify, it can help you use the right formulas. If you are stuck it can help you move forward.

Fuck AI

I care. I care enough to crater copilot.

People expect AI to be default feature. Image search was once what was "AI". Photo recognition was once what was "AI". Voice recognition was once what was "AI". These all fall under the field of AI/ML. It's until the next state of the art comes along. Then it's no longer "AI" but a standard feature.

I have no idea why this phenomenon is but that's the way it's been. When the field of AI/ML makes its leap to the next frontier. The current "AI" which is LLMs will longer be "AI" but a standard feature.

Maybe because fictional media as set the goalpost at AGI. So nobody is expecting to be buying "AI" hardware until they are buying an AGI machine that is a conscious cybernetic lifeform. Otherwise practical AI as we know it is assumed to be just another software package that runs on any computer.

Most features are relabelled years old shit...google on tap is now gemini screen search.

Things like chatbots have gotten better but bleh, I dont want to give up my privacy for this shit

Local AI kind of sucks right now, and everything is massively over branded as AI ready these days.

There aren’t a lot of compelling local use cases and the memory constraints of local mean you end up with fairly weak models.

You need a high end high memory local setup to get decent token rates, and I’m finding right now 30-70b models are the minimum viable size.

That doesn’t compare with speed of online models running on GPUs that cost more than luxury cars.

Oh hey, I got one of those buttons on my new laptop that literally never booted into Windows. Pressing it Linux says it's "Meta + CTRL" (I think), which is pretty useful. Got it for the good price/performance/build-quality ratio.

Didn't yet find a good use for that fancy NPU, the XDNA driver just arrived a month ago or so. Perhaps for use with Upscayl or something actually useful.

The only real purpose of AI is to get sweet VC money. Beyond that...

WTF is an AI computer? Is that some marketing bullshit?

Oh we care alright. We care about keeping it OUT of our FUCKING LIVES.