Comment Re: AI is like a Ouija (Score 1) 52
You fail your nickname.
You fail your nickname.
Which, of course, is AWESOME.
A human who knows something will reject an obviously wrong answer, but since the LLM knows literally nothing and the AI companies won't pay for it to check even its own work (which won't solve the problem but will REDUCE the major fails) it will just happily shit out a catastrophe.
You forget your metamucil this morning, grandpa?
People aren't "printing guns", at least not with plastic printers.
Yes, they are. They aren't printing every single part of the gun, but yeah, they are printing guns. And I say that as someone who plans to print one eventually, though probably not while I live in California. You can make your own rifled barrels with EDM, too, so you actually can manufacture every part of the firearm yourself.
Many printers, including Bambu Labs', don't have endstop sensors. They run to the end and detect the stepper stall.
Yeah, that's also done with a sensor. It's done with current sensing. And it's not a hard stop, it's a soft stop. So, exactly what I said it was. Note I didn't mention a switch or hall sensor.
They're direct driven by the stepper motors and don't have the power to "strip belts or cogs."
Then they can kill the steppers. That's not better.
To answer your question if the output of an LLM is nondeterministic then of course it is by definition nondeterministic.
If you are quibbling about technical details such as logits only being influenced by randomness and not themselves being random then randomly perturb the weights of the model or introduce noise into the calculations until you are satisfied. If there is some technical detail to quibble about please explain why the quibbling is relevant to assertions related to consciousness.
OK... so now you are quibbling about the definition of deterministic and nondeterministic, and I happen to disagree with you.
THIS is why I provided the example of passing Ollama a static seed - it is entirely deterministic. You seem to refuse to accept that point, and that's the sort of thing that gets people yelling, "This CANNOT be overstated. LLMs are software, they execute on machines that are entirely deterministic and do not work unless they are. Non-determinism is literally simulated in AI. This must be said over and over.", as dfghjk had stated.
We cannot proceed to explain how that relates to consciousness if we can't even get past agreeing on what nondeterminism is.
Nondeterminism for the context of this discussion is when it is physically impossible to predict the output of a system from its inputs in advance.
If you execute an LLM using a PRNG with a known seed value the output of the LLM is deterministic.
If you execute an LLM using a hardware random source based on thermal noise the output of the LLM is nondeterministic.
This isn't rocket science. Still the same question remains WTF does determinism have to do with consciousness?
Then we disagree on whether or not LLM's are nondeterministic.
However, your definition contradicts the example. You noted nondeterminism is, "when it is physically impossible to predict the output of a system from its inputs in advance," but you then say an LLM with a random seed is nondeterministic. Is the random seed somehow not one of its inputs? If I know the inputs, I can predict the output 100% of the time, assuming the random seed (which is an input passed as part of the call) is considered one of the inputs (because it is).
Once again, you prove that, "This CANNOT be overstated. LLMs are software, they execute on machines that are entirely deterministic and do not work unless they are."
And if LLMs are deterministic, then they can not be considered conscious for those whose definition of consciousness includes a requirement for being nondeterministic. Note, that is NOT the same as saying consciousness is or is not determinism; It's saying that it depends on it as one of the attributes of consciousness.
I am asking for an explanation of assertions related to determinism and consciousness that someone else made. These claims were not made by me. I have no duty to provide any definition of anything. I'm asking for information not quibbling over definitions.
You're asking for a definition while claiming you are not quibbling over definitions. All you are doing is quibbling over definitions.
FWIW, I have no interest in defining consciousness either, but I do have an interest in the definition of deterministic behavior.
So why are you wasting my time by demanding that I provide YOU with a definition when you are the one making the claims?
Why are you wasting my time?
Did I make such a demand? Are you referring to this request, "Please provide your definition of consciousness"?
What claims (about consciousness) did I make? Perhaps you're confusing other replies in this thread?
As to why, it's right there - I have an interest in the definition of deterministic behavior. I believe that's something we, as people, should be able to agree upon, whereas consciousness tends to be more philosophical, akin to trying to define the "soul". Your last reply has confirmed that your understanding of deterministic behavior, while similar, differs from my own to the point that you consider an LLM with a random seed to be nondeterministic, yet one with a static seed to be fully deterministic.
I would offer that something that is fully deterministic can not exhibit free will. Apply that as you see fit to the above.
Once the machine is paid off I'm never touching the card again.
But current LLM chats are more aptly compared with a ouija board.
Absolutely not. Ouija has nothing in it which doesn't come from the players. LLM is based on its training data and random numbers. The two could not be more different.
Super Socket 7 lived about three years longer than it should have, but I saved a ton of money building computers on that platform.
Same here. I had a Cyrix 6x86 and a couple of different K6s including a K6/3+ before I went to Slot 1 and then Slot A, and it's been all AMD since...
I just bought a Mac Mini last month. The no questions asked educational discount took $100 off. Also signed up for the Apple card giving me 12 months of 0% interest payments.
I think you are too optimistic.
People who believe in conspiracy theories, will simply move on to another conspiracy theory, if their pet theory is disproven or proven. There is a kind of person who simply believes in conspiracy theories, they are primed and ready at all times to believe them.
Same thing here. My firewall/router is a core 2 duo box that set me back $35. I would like to replace it but it keeps chugging along without complaint.
How about a modern kernel on a 486? https://github.com/Sharktastic...
F-150 people don't strike me as the type to accept anything really innovative like the lightning.
Ford wasn't counting on people switching from gas F-150s to the Lightning. They were counting on the F-150 name helping to sell their new product. This has essentially been effective, in that they sold about as many of those as they reasonably could have hoped to have done. People who bought them generally seem very happy with them, though not very many people were ever going to be in the market for that vehicle.
With that said, Ford should have parlayed that success into a lighter, cheaper EV pickup. Call it the F-100 Lightning. I'd guess they haven't been able to execute on the supply parts of the picture, and that's the real reason it hasn't happened. It took Tesla a while to get into the swing of having battery supply at a good cost. Ford simply isn't there yet, so they are using protectionism to delay the progress of the entire market in an effort to to catch up.
The best solution is replacement of their electronics and thus freedom from their entire software stack. There are mods for a number of printers for doing this. I for one have a FlashForge Adventurer 5M and while it is hackable and they haven't done anything stupid to users yet, they do have horribly inadequate RAM (128MB!) so there's a project to replace all the electronics with a Pi and a common control board.
On occasion there have been commercial replacement boards for some printers. Biqu has made some for a couple of Bambu models.
The nice thing about standards is that there are so many of them to choose from. -- Andrew S. Tanenbaum