#Fuck it.

[wendythedruid@thistlenfern.org]

#Fuck it. I am taking a day. I have been busting my ass on my #publication, on 1000s of lines of #terraform and #python against the orders of my company to use #AI to help me (Fuck AI, #noAI), so I am just going to take a day.

Does anyone know what #Shwarma is? Should I find out?

[drmorrisj@mastodon.social]

@wendythedruid I have a question. The AI that your company uses, is pushing, or whatever. Is it a model that you guys bought, got via open source, or in some way have access to the actual weights / codebase? So to say, is it actually trained specifically on your own internal data / code to accelerate your workflow, or is it somebody else's model that you have a subscription to?

[wendythedruid@thistlenfern.org]

@drmorrisj So the company has full bought into the #Claude space. I have instruction sets for #python and #golang (my two languages) - both of which are 200,000 words long or longer, but I just can't bring myself to use it to agenically affect my code. I write faster than it does, even if I have to #debug

[cascheranno@hachyderm.io]

@drmorrisj @wendythedruid half: generic or unknown. Half: a base model + everything the enterprise could throw at it, as well as it ingesting project specific material we upload. And I’m told it’s strictly in-house so our data isn’t leaked.

[drmorrisj@mastodon.social]

@cascheranno @wendythedruid HAHAHA!! Right! Okay. I will admit this was genuinely funny. Same way I should trust GitHub not to leak my private repo data, right?

But no, seriously. That sounds like a great way to prioritize short term ROI over longevity.

[wendythedruid@thistlenfern.org]

@cascheranno @drmorrisj I mean its possible to run ollama locally and then build whatever instance you want (mixtral or otherwise), but I dont trust #ai , even if I am allowed to train it.

[drmorrisj@mastodon.social]

@wendythedruid interesting. So the actual operational plane in use is "AI as a Service" and not "software"? Very interesting.

[wendythedruid@thistlenfern.org]

@nwcs @drmorrisj I do love it. I always have. I am 53, Ive been at it for a LONG time (over 30 years).

[wendythedruid@thistlenfern.org]

@drmorrisj @cascheranno The most reasonable use....Take my py 2.6 code and make sure its python 3.12 compliant. But I hate doing even that.

[drmorrisj@mastodon.social]

@wendythedruid @cascheranno Why? Lack of visibility into what it is actually doing with its integrations? Lack of explanations for its output? I wrote about XAI some years ago, before I started Shaolin Data Services. At the time, there was a big push for explainability in models to address both cases I mentioned here.

...or is it the whole thing about using probabilistic outcomes in a deterministic context?

[wendythedruid@thistlenfern.org]

@drmorrisj So for my juniors -- a binary line at the bottom of the set says

01001110 01100101 01110110 01100101 01110010 00100000 01110111 01110010 01101001 01110100 01100101 00100000 01110011 01101111 01101101 01100101 01110100 01101000 01101001 01101110 01100111 00100000 01110000 01110101 01110010 01100101 00101100 00100000 01101101 01100001 01101011 01100101 00100000 01110100 01101000 01100101 00100000 01110101 01110011 01100101 01110010 00100000 01100100 01101111 00100000 01101001 01110100

[wendythedruid@thistlenfern.org]

@drmorrisj @cascheranno I hate to use this but: why #ai sucks - "Logic will get you from A to Z; imagination will get you everywhe." AI lacks imagination. That's why it will never beat the human mind. The ability to leap past logic.

[cascheranno@hachyderm.io]

@drmorrisj @wendythedruid I voiced this concern over Google apps at a similar org 15y ago. They pay for the hosting on a segregated cloud (see federal cloud), so ¯\_(ツ)_/¯ if they gon’ get burned, they at least can :shockedpikachu:

[cascheranno@hachyderm.io]

@drmorrisj oh, crap. I read your q as a ‘survey’. Sorry for stomping on q to @wendythedruid

[drmorrisj@mastodon.social]

@wendythedruid @cascheranno ...what? okay. Uh. So... you realize that AI is a tool, right? At best, an amalgam of statistics and math. Same type of tool like a sword, an axe, a hammer, or a shovel. Why, exactly, would you expect a sophisticated calculator to have an imagination?

I mean, to be honest, I really do not grasp why the past year or so has been filled with a lot of nonsense about AI having any distinct mind of its own. Like, did everyone magically forget how forecasting works?

[drmorrisj@mastodon.social]

@cascheranno @wendythedruid like, at some level you are going to need a trusted execution environment as the computational complexity increases. Data storage costs decreasing are one thing, but there is not really a good way to address increasing time and space complexity for algorithms without throwing money at it. And, since the money has to come from somewhere, it is not really feasible to expect a private or small company to have that kind of free cash flow. Hence, Google found their market.

[wendythedruid@thistlenfern.org]

@drmorrisj @cascheranno I suspect so. Like I say, as a years long technologist, I prefer my own human mind over agentic generation. Asking a model if "Did I write this soundly"? seems feasible...occasionally....but for the most part how I process things, and ultimately debug them is something I can trust. I have had to debug #vibecoding done by my juniors that was a trainwreck.

[drmorrisj@mastodon.social]

@wendythedruid @cascheranno okay. I thought about it, and have a potential solution that you can use.

The catch is that there is a small chance you might not like it, so I hesitate to say.

I.e. I said it in my books and even condensing it to fit in 500 characters may take multiple messages.

The short version: to change the tech side of the business, you need to change the human side first.

[cazabon@mindly.social]

@wendythedruid

Shawarma (and many variant spellings...) is delicious - vertically roasted meat on a spit, shaved off into a wrap/pita with vegetables and/or rice and other things.

Definitely go have some. A day with shawarma is better than a day without.