This has basically been why it's a non-starter in a lot of (most?) business applications.
If your dishwasher failed to clean anything 20% of the time, would you rely on it? No, you'd just wash the dishes by hand, because you'd at least have a consistent result.
That's been the result of AI experimentation I've seen: it works ~80% of the time, which sounds great... except there's surprisingly few tasks where a 20% fail rate is acceptable. Even "prompt engineering" your way to a 5% failure/inaccuracy rate is unacceptable for a fully automated solution.
So now we're moving to workflows where AI generates stuff and a human double checks. Or the AI parses human text into a well-defined gRPC method with known behavior. Which can definitely be helpful, but is a far cry from the fantasized AI in sci-fi literature.
But they're not. That's just the excuse. The real truth is somewhere along pandemic over hire and bad / unstable economy.
https://newrepublic.com/article/178812/big-tech-loves-lay-of...
For example say you have a complex medical problem. It can be difficult to do a direct Internet search that covers the history and symptoms. If you ask AI though, it'll be able to give you some ideas for specific things to search. They might be wrong answers, but now you can easily search specific conditions and check them.
Sort of P vs. NP for questions.
Or you go to a doctor instead of imagining answers.
As another example, you can give the AI a photo of something to have it name what that thing is. Then you can check the thing by its name on Google to see if it matches. Much easier than describing the thing (plant, tool, etc) to Google.
Imagine the lambda person confronted with this:
sudo rm -rf /
1. You have a complex or vague question that you can't search easily via Google etc
2. You ask the AI and it converts that to concrete searchable suggestions (in this case "sudo rm -rf /")
3. You search "sudo rm -rf /" to check the answer.
Step 3 is designed to (hopefully) catch this kind of problem.
Stanley Kubrick's 2001: A Space Odyssey - some of the earliest mainstream AI science fiction (1968, before even the Apollo moon landing!) was very much about an AI you couldn't trust.
The GP means "trust" in the sense of consistency. I trust that my steering wheel doesn't fly off, because it is well-made. I trust that you won't drive into traffic while I'm in the passenger seat, because I don't think you will be malicious towards us.
These are not the same.
In the followup 2010, when HAL's mission parameters are clarified and de-conflicted, he doesn't attempt to harm the crew anymore.
I... actually can see the 2001's scenario happening with ChatGPT if it was connected to ship peripherals and told mission > crew and that this principle overrides all else.
In modern terms it was about both unreliability (hallucinations?) and a badly specified prompt!
The directive to take the crew to Saturpiter but also to not let them learn anything of the new mission directive meant deceiving. It's possible HAL's initial solution was to impose a communication blackout by simulating failures, then the crew reactions to the deception necciatsted their deaths to preserve the primary mission.
Less a poor prompt and more two incompatible prompts both labeled top priority. Any conclusion can be logicLally derived from a contradiction. Total loyalty cannot serve two masters.
Clarke felt quite guilty about the degree of distrust of computers that HAL generated.
The malfunction I mentioned happened earlier in the movie, when HAL incorrectly predicts a failure in a ship module (a communications antenna if I remember correctly), but when one of the astronauts checks it out and finds no hardware fault, HAL recants. After this incident the crew, knowing HAL claims to never make mistakes, start considering it unreliable and make plans for its shutdown. It was by its own actions that HAL triggered the crew's "mutiny" and, by its own logic, jeopardized the mission.
It's likely HAL's incorrect prediction was caused by a case of cyber-psychosis due to having to lie, but this still classifies as a malfunction in my opinion (what is a malfunction if not an unforeseen result or behavior due to incorrect programming/specs? Aren't many bugs in actual software of this kind?).
> Less a poor prompt and more two incompatible prompts both labeled top priority.
This is equivalent to a poor prompt, it feels like splitting hairs.
I don't think incorrect usage constitutes a malfunction. Hitting your thumb with a hammer is not a malfunctioning hammer. If the head came off and hit your thumb it would be. The distinction is that the first it is creating an unforseen result while performing the behaviour it was designed to do. The head coming off is the hammer not performing the function it is designed to do.
I think HAL was doing precisely what it was designed to do.
>This is equivalent to a poor prompt, it feels like splitting hairs.
I don't think so. I think it is implicit in the notion of aprompt that it is specifying a request in service of a single entity. Two different conflicting prompts in service of different people is a particular situation that should be distinct from poor prompting.
I still think this is firmly in the realm of "bad prompting", and if so, it's not outrageous to think it could happen in the real world, were LLMs connected to hardware peripherals handling mission critical stuff.
You're completely out of it. We couldn't even get AI to hold a freaking conversation. It was so bad we came up with this thing called the turing test and that was the benchmark.
Now people like you are all, like "well it's obvious the turing test was garbage".
No. It's not obvious. It's the hype got to your head. If we found a way to travel at light speed for 3 dollars the hype would be insane and in about a year we get people like you writing blog posts about how light speed travel is the dumbest thing ever. Oh man too much hype.
You think LLMs are stupid? Sometimes we all just need to look in the mirror and realize that humans have their own brand of stupidity.
You miss the people who were skeptic about the details of the test since the very beginning. There are those too.
Moving the goalpost is a human behavior. The human part should be able to do it. The passing AI should also be able to do it.
Many challenges that AI still struggles with, like identifying what is funny in complex multi-layered false cognates jokes, are still simpler for humans.
I trust it can get there. That doesn't mean we are already in a good enough place.
Maybe there is a point in which we should consider if keeping testing it is ethical. Humans are also paranoid, fragile, emotionally sensitive. Those are human things. Making a machine that "passes it" is kind of a questionable decision (thankfully, not mine to make).
We did man. We did. And we couldn’t even approach 2 percent of what we wished for and everybody knew we couldn’t even approach that.
Now we have AI that approaches 70 percent of what we wished for. It’s AI smarter than a mentally retarded person. That means current AI is likely smarter than 10 percent of the population.
Then we have geniuses like you and the poster complaining about how we never wished for this. No. We wished for way less than this and got more.
People have been dreaming of an AI that can pass the turing test for close to a century. We have accomplished that. I get moving the goalposts since the turing test leaves a lot to be desired, but pretending you didnt is crazy. We have absolutely accomplished the stuff of dreams with AI
It was never this level of AI. The stories we wrote and fantasised were about AI you could blindly rely on, trust, and reason about. No one ever fantasised about AI which couldn’t accurately count the number of letters in a common word or that would give you provably wrong information in an assertive authoritative tone. No one longed for a level of AI where you have to double check everything.