It is like tons of people cannot actually think. Almost if they are a different species.

Well, one of my two banks has an app as 2nd factor and the other I have a hard-token for. But for this story to make sense, it sounds like Bank of America allows username and password. Or maybe email as not-really-2nd-factor and the agent gets access to the email as well. Or something like that.

I get that banks do not want users to use LLMs. Some of their customers will get hit really badly before this becomes regular enough it will not be in the news anymore. Some of those hit will blame the bank. And the one thing banks really dislike is bad press.

Probably not. Would be nice though. Well, not that I use Google search anymore, but the others probably have the same situation.

Looks like basically nobody using AI can make it pay for itself. If they are honest.

Yes. That was what I tried at the beginning of the craze. Stuff you can easily look up (but should usually not need to look up), LLMs can do (mostly), but in the engineering I do I often run into things you cannot look up. Total failure and sometimes hallucinations on top. Since then, I occasionally check back, but nothing has changed really. Simple, well-known things LLMs can usually do, things that require actual insight LLMs cannot do. No change there. The illusion has gotten a bit better, but that is it.

Ah, no. Not even remotely. You are pushing a fallacy here, the "fallacy of experience" (https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fantsstyle.medium.com%2Fthe-logical-fallacy-of-experience-1367653709f0). For some simple examples that may be within your grasp: You do not have to do hard drugs to be able to know doing them is not a good experience. Your also do not have to participate in a pyramid scheme to understand it is a scam and there is nothing to be gained. You may want to look up "cognitive surrender", to find out that using LLMs is actually pretty dangerous to you. You may want to look up the negative experiences actually good coders make with LLMs.

That said, I have used LLMs "productively", namely in research showing how utterly limited it is. This research is initiated by me and done by students in thesis work. And just last week I got (again) a few nice examples of failures from current frontier (!) models. Will probably make a good paper eventually. For the time being, I will stay away from using LLMs, too dangerous and no actual benefit to my work. My work is mostly insight-based, something LLMs cannot do, even if they can by now fake it well enough to impress people that are not that smart.

You guys do know that there's a process in motion right now to make this debate meaningless and redefine "planet" again? As it stands, the current definition of planet you're arguing over doesn't include any "exoplants". It's actually weird, solar-system centric definition that clearly has no future. They're looking to define clearing its own orbit as 10^23 kg.

If you haven't read the interview, in it she relates a story where her young daughter's teacher asked her to write three things that made her child special. Apparently she is such an awful mother that she couldn't do that, and turned to AI for assistance. The AI shat out some generic Hallmark style platitudes, and she was delighted.

People like her want to dehumanize all of us to that level.

Well, ignorance, optimism and naiveness stem from not understanding the real world. Lack of risk management in doing things that can go horribly wrong fall in the same class. In the case at hand, anybody that actually looks at what gets reported in the mainstream press should be aware that LLMs are neither reliable not trustworthy. Coming to that conclusion does not require specialist knowledge or high intelligence. You just need to see what is.

If "stupid" is not scientific enough for you, replace with "intellectually limited with regards to insight into the functioning of reality". "Stupid" is a useful short form for that, as you seem to acknowledge in your statement about marketing.

I do fully agree to that statement about marketing. A prime current AI example is all those bad and mediocre coders claiming AI is great and makes them so much more productive, not realizing that AI will make coders on their level redundant and unemployed.

No actual disagreement with what you're saying there, but a few things about your example. A. You were dealing with at least one competent engineer and B. you're talking about mechanical engineering. We both agree that you need to hammer out the details with the client, but many clients looking for software to be developed are not engineers and frequently not even all that technical plus the ease of rolling out a mockup or prototype is generally a lot easier in a pure software environment. So sometimes hammering out the details can involve a mockup as a version of the specification for them to pore over. Especially for UI and workflow elements. Beyond that, sometimes dealing directly with the engineers whose requirement it is can be a bit of a luxury. Often you're dealing with one or more layers of middlemen. Worst cases can include your salespeople working out the requirement with their salespeople. If you lose the battle to have at least a constructive conversation over their heads, you're generally stuck with the sure and certain knowledge that there's some fundamental misunderstanding in the middle.

So, that's really the comparison I'm making here. In the traditional paradigm, you ideally have competent engineers/developers or even technically inclined end users who can work everything out. This new paradigm threatens to have the "developer" become the one who has no idea about what has been asked for, what it would really be used for, how to ask the right questions to improve the requirements documents, etc. They just become a middleman who feeds that into the AI.

You think (or rather your LLM thinks) banks are depending on the user login as a security barrier protecting their internal software? That is not the case. Banks assume a significant part of the devices users are using to log into the online banking are compromised and hence they assume attackers have access to everything they expose to customers. They have used that assumption for at least the last 20 years, probably much longer.

Hence, sounds plausible - is total bullshit. Typical LLM quality resulting from lack of insight in their training data and no native insight whatsoever.

Ah, sorry. I am forgetting you are probably thinking about an US bachelor's. Things are a bit different in Europe, even if named similar. My mistake.

What crappy ass bachelor's program is that? I do deep dives and I teach on bachelor level.

These days, I actually think FizzBuzz is not bad. I read about it long ago on the Coding Horror blog, wrote it in 4 different languages on the commandline in a few minutes, was unimpressed and forgot about it again. But later I got back to it and it actually has something going for it. Especially when a candidate starts to discuss different approaches and aspects, because the smart ones can see that it is utterly trivial, deliver a minimal (!) solution and will try to add value by discussing its aspects. Good to hear it still works and also works as detector for LLM users in cognitive surrender.

For color I would have to look up color escapes for an xterm. It probably understand some ANSI escapes, but I never tried. ;-)

Nope. You fail. This never meant AI and it does not mean AI now. You are a scam victim.

If we ever get AGI, this may (or may not) be different, but LLMs cannot get you that.