Mist
REEEEeyore
- 31,192
- 23,342
Yes, exactly. But it's important to talk about the tech as it works right now not some imagined future tech.
This is still just "a neat trick that happens when you upscale autocomplete by a fucking lot."
Chat GPT AI
- Thread starter Sanrith Descartes
- Start date
-
Guest, it's time once again for the massively important and exciting FoH Asshat Tournament!
Go here and give us your nominations!Who's been the biggest Asshat in the last year? Give us your worst ones!
This is still just "a neat trick that happens when you upscale autocomplete by a fucking lot."
Aldarion
Egg Nazi
- 9,717
- 26,638
Reading this debate about "when machines have agency" makes me think a lot of people are still missing the importance of LLMs.
Its not that they're a real boy now. The important point to grasp is not about them at all. Its about us.
Its that a large fraction of human thought and speech is nothing more than a biological upscaled autocomplete.
Anyone still thinking about the Turing test or anything like that is, I fear, missing the point. We asked for decades when a machine would join us over here on this side of the line. If you're paying attention right now, you should be asking yourself if there even is a line, and if so, are we on this side of it in the first place?
Its not that they're a real boy now. The important point to grasp is not about them at all. Its about us.
Its that a large fraction of human thought and speech is nothing more than a biological upscaled autocomplete.
Anyone still thinking about the Turing test or anything like that is, I fear, missing the point. We asked for decades when a machine would join us over here on this side of the line. If you're paying attention right now, you should be asking yourself if there even is a line, and if so, are we on this side of it in the first place?
- 2
- 1
- 16,738
- 7,764
The domain itself looks suspicious, let alone directly linking a .mp4 file.
I have NoScript and Ublock Origins running in addition to Windows Defender. The best piece of security is using your brain.
- 1
- 1
Mist
REEEEeyore
- 31,192
- 23,342
NPCs believing that they're NPCs.
Just because a machine can algorithmically emulate human language patterns when trained on hundreds of billions of parameters of prior human knowledge and exabytes of text does not mean this is what humans are doing. Humans need nowhere near this size of a training set in order to start generating novel speech. Humans can also generate new ideas, something LLMs have not been demonstrated to do. Humans can even invent new tokens, new language, and do it all the fucking time.
But you're right, some swath of the population probably is just a small language model.
Humans don't start from a blank slate. How much information have our ancestors ingested over three and a half billion years of evolution? How much energy did that process take?
Mist
REEEEeyore
- 31,192
- 23,342
Not quite relevant to what I'm talking about. You're confusing knowledge vs language acquisition. Each new human is a blank slate when it comes to language tokens, phoneme acquisition, etc. Only the capacity for language is passed down genetically.
You could say that each new human benefits from the optimization that has been performed on our languages over time to make them easier to acquire, but then you actually look at our languages and realize they're anything but optimized for acquisition.
This is where having an understanding of developmental psychology/developmental linguistics comes in handy for understanding vast differences between humans and LLMs.
They're not unrelated. We're a whole lot better primed for acquiring language knowledge than a math equation is. A more efficient acquisition structure with some built-in predicates absolutely effects the amount of training required. This is the basis for your scoffing at the amount of training data LLMs require.
My point was that I don't think it's honest to deny credit to all the information embedded in our biology.
Last edited:
GuardianX
Perpetually Pessimistic
- 7,164
- 18,153
I am desperately trying to understand your point..
You state "Only the CAPACITY for language is passed down genetically", how is this any different than a LLM? An LLM is the capacity for language. His concept was that, if I'm understanding correctly, humans evolved though billions of years the capacity for language. A point that you also make on the opposite edge of the same side of a coin. Humans and LLM's are both learning over billions of generations (iterations).
Then you make a statement about acquisition of language, but languages were never created to be easily acquired, they were formed for social integration and the conveyance of ideas THROUGH a common point of social integration.
Then you are making a point about Psychology and Linguistics (Social Integration) as it pertains to Humans VS a Library? Or are you using LLM's as a stand in for the concept of "AI"?
I dunno I look at AI like this, Artificial (Humanistic in nature) Intelligence (LLM's). They have the knowledge but the social integration is lacking for now..
Denamian
Night Janitor
- 7,459
- 19,675
- 1
Mist
REEEEeyore
- 31,192
- 23,342
No, an LLM is a pre-trained model of a language or usually multiple languages. The LLM is the output of the training function, producing a very long static string of digits representing a map of the data it was trained on. That static map is then used to turn prompts into responses.
Proceeding from false premises, your entire post is wrong and displays no understanding of what you're talking about.
- 1
- 16,738
- 7,764
My point is that I'm not clicking on a link to "goody2.ai", ever. Nor "brain.wtf", for that matter. The original post reads like a textbook phishing email.
I'm glad the links are safe. I did not intend for you to take that risk.
- 1
Denamian
Night Janitor
- 7,459
- 19,675
I have a perfectly sandboxed environment for opening these type of links.
I also can't be bothered to use it and just open it another browser.
It's time to come to terms that there are shitloads of TLDs now. People and companies will use them for shits and giggles and completely legitimate sites nowadays. The days of .com, .net, .org, .gov, etc being the only legit TLDs in use is coming to a close.
I also can't be bothered to use it and just open it another browser.
It's time to come to terms that there are shitloads of TLDs now. People and companies will use them for shits and giggles and completely legitimate sites nowadays. The days of .com, .net, .org, .gov, etc being the only legit TLDs in use is coming to a close.
GuardianX
Perpetually Pessimistic
- 7,164
- 18,153
Party 2 makes several points, and their correctness depends on the context and interpretation:
1. **Language Acquisition in Humans**: It's generally accepted in linguistics and developmental psychology that humans are born with the capacity for language acquisition. This capacity includes the ability to recognize and produce phonemes, words, and sentences. However, the specifics of language (e.g., vocabulary, grammar) are learned through exposure to language in the environment.
2. **Optimization of Languages**: Party 2 suggests that languages are not optimized for easy acquisition. This viewpoint is debatable. While languages naturally evolve based on cultural, social, and historical factors rather than with the specific goal of being easy to acquire, linguists do recognize patterns in language structures that facilitate learning, such as regular grammar rules and frequent vocabulary usage.
3. **Comparison with LLMs**: Party 2 draws a comparison between human language acquisition and the functioning of Large Language Models (LLMs). While LLMs are trained on vast amounts of text data to learn patterns in language, they don't acquire language in the same way humans do. LLMs are trained algorithms that process and generate text based on statistical patterns, whereas humans learn language through exposure, interaction, and cognitive development.
Overall, Party 2's statements capture some nuances of language acquisition and its intersection with LLMs, but there may be room for further clarification or refinement in their arguments.
---
Congratulations Mist...your point even requires clarity from GPT3.5
Mist
REEEEeyore
- 31,192
- 23,342
You have no idea what you're talking about, so you then asked a bullshit machine, which then agreed with me, but you lack the comprehension to realize it.
Good job, and an excellent example of why humanity is doomed and probably deserves it.
- 1
GuardianX
Perpetually Pessimistic
- 7,164
- 18,153
If you can't explain what you mean, then so be it, but even chat GPT agrees that you need to clarify some things.
As usual, it's been a pleasure talking to you about AI mist.
- 1
LLM's as I understand it are just statistical models calculating the probability of what word is likely to come after it in a sequence given its training data. They have trouble with complex instructions specifically because of this. There's no real brain or logic behind it other than the parameters they feed it. True AI still has a long way to go.
