Fuck AI

It's a question of using the right tool for the right job, but CEOs that don't even know the job, nor the tools it needs, fall for charlatans selling them snake oil that'll magically increase revenue, cut expenses or otherwise make them look good when the time for allotting their bonus comes around. And why wouldn't they? They're not the ones to pay for their fuckup. They'll step down, get a chonky farewell gift and the employees taking pay cuts, losing their jobs or just suffering under the increased workload are left holding the bag.

For text processing with negligible precision, LLMs may be a great fit. For repetitive, primitive coding tasks in the hands of a senior coder, they can save time. For replacing humans, they lack critical thinking and semantic understanding.

Add to that the issue that, even if AI did bring the promised savings, that wouldn't benefit the general public, particularly the people whose prospects for a job are now fucked along with their prospects for a living. This is a social factor that no discussion of tool aptitude or AI applications can solve.

We're on the same page about the problem, I think. And as I said in my original comment, there are absolutely reasonable applications for LLMs, they just won't pay back the investments chucked into the research quick enough. So we get instead our current "AI" moment.

Now, as a counterpoint to your argument about black and white positions — I see a lot of people filling the gray areas with what is essentially whataboutism:

• "The next model will be much better"
• "People said the same thing about the internet"
• "Sure it sucks and make stuff up, but it's good enough for me"
• "You can't stop progress"...

In that context, I think it's important to take a black and white stance, staying off that slippery slop(e) of bad faith arguments. "AI" as it is being marketed to us is bullshit, and needs to stop.

But once that bubble is burst (and it will one way or another), research needs to continue in a more focused manner into the fields where LLMs can actually make a positive difference (hint: it's neither search engines or your operating system).

It’ll be sad and unnecessary if we throw the cancer diagnostics out with the work stealing, resource wasting, slop machine.

It would be a tragedy if we accepted the latter so that we can obtain the former.

Limited models trained on good datasets have limited applications in analysis of similar data.

LLMs are a gigantic soul-sucking scam with no practical usage.

You are creating a distinction without a difference.