Agree. Checks are the way.

They are all some binary file that completes a task. That LLMs have a statistically derived “personality” (if one would be dangerous enough to call it that) to make people like them was a purposeful mistake in development to trick people into embracing the technology. LLM is non-deterministic by nature which means results can never be trusted.

Honestly, the Photoshop filter analogy may be wrong because the filter is deterministic. The same settings produce the same results every time as it is math on an image file. LLMs literally can’t do that.

LLM tech is only seeing more “reliable” results now as they often throw the tool against a sandbox environment where it has to execute deterministic code to compensate for the random unreliability, and it still screws up.

Linus knows code, the tool sped up his process. In the absence of his knowledge, it would be useless.

Neat toy, but way too inefficient to bank the future on, and very dangerous to have such a religious reaction to.

LLMs are akin to fancy Photoshop filters for code. They can be used well by professionals and terrible by n00bs.

They are but a (very inefficient) tool.

It’s so messed up that some states legally allow merchants to charge a “credit card fee”. Providers of services much like the above can charge if it’s a debit, credit, etc. even if they both say “Visa” on them. The concept of money in the US is so obfuscated now.

To directly quote a painted piece of wood in a pile of rubble on the side of the road after a big flood in Colorado in 2013: Fuck Allstate.

100%. Everything they say should always be inverted.

Insane how weighted response filtering and graph search have existed for eons, to say nothing about the fact the search query could be pre-parsed by an LLM/“AI”, and only now do these moronic companies use hard keyword filters.

Makes you wonder if it’s really “for the children” or just attempts at hard-censoring to train their future AI models.

“I now have 2 homes, a wife and kids, a dog, a boat, 3 cars, all on my 1920s programming union salary! Wife doesn’t even need to have a job! Just remember to show confidence, it’s easy!”

Interviews right now in software:

• Apply to job
• 1.5 months later receive email to take an AI-“proctored” code interview where you have to show your ID, share screen, cam, mic like some Orwellian njghtmare
• Maybe get a soft-interview with employees where they are just asking you about local weather while seemingly personable, but actually just trying to find out if you live where you say you live
• Possible technical interview with humans where it’s “pair programming” but actually just a human-proctored coding puzzle irrelevant to the job description and the “interviewer” is looking at another monitor surfing reddit
• Contemplate if homelessness isn’t that bad versus all this bullshit
• Maybe retail? Construction? Farming?
• Work at CostCo?

All to get a job to be a condom between Claude and source control as that other recent lemmy post shared.

Is there a browser plugin to constantly query their AI in the background? Wasting their power might send a message.

“Conversing” with an LLM is a different dialect than speaking to humans if one wants consistent output. Keyboard still wins. LLM is exhausting. Humans are dope.

OpenAI employee builds a rapport with an Apple employee on-site. Apple employee leaves laptop unlocked with OpenAI employee in their office and goes to the bathroom. No joke, it’s just that simple.

You’d think they would have learned that from the cloud or any SaaS provider multiple times over already.

My favorite edition is yt-dlp.

Snap Inc. is its own company. Not that it is a good company.

Many of these schools forbid stuff like this

Coming soon to a USA near you: Constitutional amendment 28: Freedom of oppression. Indoctrination starting at the college/university level.