5 ms·

>But it's not a sentient AI. It's just a language model. Just a beefed up auto-correct. There is a large space between "sentient" and "beefed up autocorrect".

>But it's not a sentient AI. It's just a language model. Just a beefed up auto-correct. There is a large space between "sentient" and "beefed up autocorrect". Why do people insist on going for the most reductive description they can muster?

pixl97 4y ago
Because the average person you speak to would consider beefed up autocorrect to be near magic as it is. Once you get near to the limits of an individuals comprehension adding more incomprehensible statements/ideas doesn't really change much, their answer is still 'magic'.
mistermann 4y ago
Heuristics, aka HumanGPT.
phire 4y ago
Don't mistake my "reductive description" as disapproval. I'm actually really impressed and I see a bright future. But I think it's really important that we don't give GPT3 more credit that it deserves. All the discourse around GPT3 and it's derivatives like Github copilot or ChatGPT have shown that people (even tech literate people) have a strong bias towards anthropomorphising it as some kind of "proto-sentient" AI. In my opinion, this bias is actually very damaging towards the reputation of language models. People start expecting way too much from them, and then feeling confused or even betrayed when the language models start confidently sprouting bullshit. Also, I don't think "beefed up autopredict" is reductive at all. (Though I might have said "beefed up autocorrect" in my previous comment, whoops). The 10,000 ft view of it's runtime architecture is identical to autopredict. You take some input context, encode it into tokens and feed them into a neural network, and to get a prediction of the next few words. The innovation in GPT3 has nothing to do with moving away from that basic architecture. The innovation is all about improving the encoding of tokens, changing the internal architecture of the neural network support larger models and make better predictions, improving the training process and finally the training data itself. They also massively increased the size of both the training set and the model... but I'm not sure that counts as innovation. IMO, GPT3 is actually way more impressive once you start thinking of it as nothing more than a beefed up autopredict.
- naijaboiler 4y ago
  Correct. It is indeed beefed up auto predict
  - hackinthebochs 4y ago
    It is beefed up auto predict in the same way that brains are beefed up electrochemical integrators. While technically true, it is maximally uninsightful. We should be aiming for maximum insight, not the opposite.
    - phire 4y ago
      I get the impression that you are massively underselling auto-predict. If you took GPT3's architecture and scaled it down to the size and training set of a typical auto-predict, it would produce near identical results. You wouldn't be able to tell the two apart. Likewise, if we took an auto-predict architecture from 8 years ago, scaled it up to the size of GPT3 and could train it on GPT3's training set, it would produce similar output to GPT3 and we would see the exact same emergent intelligence capabilities. (Though, it's probably not possible to compete training in a practical time-frame, the real innovation of GPT3 was optimising the architecture to make training such a large model practical) I think very insightful to point out just how similar the two are. Because it shows the capabilities of language models are not because of any architectural element, but are emergent from the model and it's training data. It also makes me excited for what will happen when they move beyond the "beefed up auto-predict" architecture. (Arguably, Bing has taken a small step in this direction by bolting a search engine onto it)
      - hackinthebochs 4y ago
        >If you took GPT3's architecture and scaled it down to the size and training set of a typical auto-predict, it would produce near identical results. This is almost certainly not true. The number of parameters is an important feature related to the quality of output. If you scaled the architecture down significantly, it would be significantly less capable[1]. But perhaps I misunderstand your point. >Likewise, if we took an auto-predict architecture from 8 years ago, scaled it up to the size of GPT3 and could train it on GPT3's training set, it would produce similar output to GPT3 This is also not true. The transformer is a key piece in the emergent abilities of language models. The difficulties in scaling RNNs are well known. Self-supervised learning is powerful, but it needs to be paired with a flexible architecture to see the kinds of gains we see with LLMs. Stacked Transformers with self-attention are extremely flexible in finding novel circuits in service to modelling the training data. The question is how to characterize this model in a way that doesn't short-sell what it is doing. Reductively describing it in terms of its training regime is just to treat the resulting model as explanatorily irrelevant. But the complexity and the capabilities are in the information dynamics encoded in the model parameters. The goal is to understand that. [1] https://arxiv.org/abs/2001.08361 https://arxiv.org/abs/2001.08361

pixl97 4y ago

Because the average person you speak to would consider beefed up autocorrect to be near magic as it is. Once you get near to the limits of an individuals comprehension adding more incomprehensible statements/ideas doesn't really change much, their answer is still 'magic'.

mistermann 4y ago

Heuristics, aka HumanGPT.

phire 4y ago

Don't mistake my "reductive description" as disapproval. I'm actually really impressed and I see a bright future. But I think it's really important that we don't give GPT3 more credit that it deserves. All the discourse around GPT3 and it's derivatives like Github copilot or ChatGPT have shown that people (even tech literate people) have a strong bias towards anthropomorphising it as some kind of "proto-sentient" AI. In my opinion, this bias is actually very damaging towards the reputation of language models. People start expecting way too much from them, and then feeling confused or even betrayed when the language models start confidently sprouting bullshit. Also, I don't think "beefed up autopredict" is reductive at all. (Though I might have said "beefed up autocorrect" in my previous comment, whoops). The 10,000 ft view of it's runtime architecture is identical to autopredict. You take some input context, encode it into tokens and feed them into a neural network, and to get a prediction of the next few words. The innovation in GPT3 has nothing to do with moving away from that basic architecture. The innovation is all about improving the encoding of tokens, changing the internal architecture of the neural network support larger models and make better predictions, improving the training process and finally the training data itself. They also massively increased the size of both the training set and the model... but I'm not sure that counts as innovation. IMO, GPT3 is actually way more impressive once you start thinking of it as nothing more than a beefed up autopredict.

naijaboiler 4y ago
Correct. It is indeed beefed up auto predict
- hackinthebochs 4y ago
  It is beefed up auto predict in the same way that brains are beefed up electrochemical integrators. While technically true, it is maximally uninsightful. We should be aiming for maximum insight, not the opposite.
  - phire 4y ago
    I get the impression that you are massively underselling auto-predict. If you took GPT3's architecture and scaled it down to the size and training set of a typical auto-predict, it would produce near identical results. You wouldn't be able to tell the two apart. Likewise, if we took an auto-predict architecture from 8 years ago, scaled it up to the size of GPT3 and could train it on GPT3's training set, it would produce similar output to GPT3 and we would see the exact same emergent intelligence capabilities. (Though, it's probably not possible to compete training in a practical time-frame, the real innovation of GPT3 was optimising the architecture to make training such a large model practical) I think very insightful to point out just how similar the two are. Because it shows the capabilities of language models are not because of any architectural element, but are emergent from the model and it's training data. It also makes me excited for what will happen when they move beyond the "beefed up auto-predict" architecture. (Arguably, Bing has taken a small step in this direction by bolting a search engine onto it)
    - hackinthebochs 4y ago
      >If you took GPT3's architecture and scaled it down to the size and training set of a typical auto-predict, it would produce near identical results. This is almost certainly not true. The number of parameters is an important feature related to the quality of output. If you scaled the architecture down significantly, it would be significantly less capable[1]. But perhaps I misunderstand your point. >Likewise, if we took an auto-predict architecture from 8 years ago, scaled it up to the size of GPT3 and could train it on GPT3's training set, it would produce similar output to GPT3 This is also not true. The transformer is a key piece in the emergent abilities of language models. The difficulties in scaling RNNs are well known. Self-supervised learning is powerful, but it needs to be paired with a flexible architecture to see the kinds of gains we see with LLMs. Stacked Transformers with self-attention are extremely flexible in finding novel circuits in service to modelling the training data. The question is how to characterize this model in a way that doesn't short-sell what it is doing. Reductively describing it in terms of its training regime is just to treat the resulting model as explanatorily irrelevant. But the complexity and the capabilities are in the information dynamics encoded in the model parameters. The goal is to understand that. [1] https://arxiv.org/abs/2001.08361 https://arxiv.org/abs/2001.08361