Live data from Hacker News

There is no A.I.

newyorker.com

431–440 of 523 posts

Re: There is no A.I.

#431

Earlier quoted context omitted.

At each new token processed (edit: either if the token comes from the user or has been generated), this dynamic internal state is updated via a formula based on the model parameters and latest processed token. To make this more clear, if you process 1000 tokens the neural network will have gone through a sequence of 1000 states. Each of these states can be probed and analysed, by training a classifier on top of this…

> To make this more clear, if you process 1000 tokens the neural network will have gone through a sequence of 1000 states. That isn't how transformer architectures work

^^ See above comment and "Transformers are RNNs" paper to convince yourself.

There are various ways of seeing how transformer architectures work.

For the past data, using a temporal causal mask, you can compute all past features as they would have been seen by applying the causal mask, which allows you to seemingly do all the past computations in parallel, which hide the sequential aspect of it.

What I've described correspond to the the simpler subvariant figure 1 of "Attention is all you need" https://arxiv.org/pdf/2006.16236.pdf , of the transformer architecture in the case where you don't give the past as input to the input branch, put rather prepend it to the output, which is the what people usually do in practice (see llama) ). If you want to use both branches of the transformer architecture, so that you can do filtering and smoothing (aka not using a causal mask) for the past data, it creates a "bottleneck" in your architecture, as you synthesize all the past into a finite sized context vector that you then use in the right branch of the transformer.

But alternatively you can loop along the time dimension first, and make the sequence more apparent. Due to how the (causally mask) transformer architecture is defined this gives exactly the same computation. It's just some loop reordering.

For the generation part, the sequential aspect is more evident as token are produce one by one sequentially and feedback to the transformer for the next token prediction.

Re: There is no A.I.

#432

Earlier quoted context omitted.

This is what I got on ChatGPT today. I assume it used GPT4: Prompt ChatGPT Actual Match 397,356 * 930,547 369,685,207,932 369,758,433,732 FALSE 36,330 * 26,951 979,458,630 979,129,830 FALSE 8,681 * 9,330 80,911,430 80,993,730 FALSE 278 * 903 250,734 251,034 FALSE 82 * 77 6,314 6,314 TRUE Edit: # of correct digits (counting from leftmost) only exceeds 3 on the smallest pair. It drops to two, as well, on the 3x3 set.

You probably used GPT-3.5. That said, I didn't manage to get GPT-4 to calculate 36,330 * 26,951 correctly. I suggested casting out 9s, casting out 11s, doing long multiplication, reversing the digits - nothing. I have a theory that it does arithmetic badly because the logic goes right-to-left, when LLMs write left-to-right. If the digits were to be reversed, it might not make as many mistakes. I ran out of attempts b…

I just had the following conversation on chatGPT

> Human: what is 36,330 26,951 * (input is expressed in reverse post fix notation)

> chatGPT: To evaluate this expression using reverse Polish notation, we need to use a stack to keep track of the intermediate results. Here's how we can evaluate the expression:

1. Read the first number "36,330" and push it onto the stack. 2. Read the second number "26,951" and push it onto the stack. 3. Read the operator "", pop the top two numbers (26,951 and 36,330) from the stack, multiply them, and push the result (971,914,830) onto the stack. 4. The stack now contains only one number, which is the final result.

Therefore, 36,330 26,951 = 971,914,830 in reverse Polish notation.

I dont think this is about notation.

Re: There is no A.I.

#434
post #318

> If the new tech isn’t true artificial intelligence, then what is it? In my view, the most accurate way to understand what we are building today is as an innovative form of social collaboration. What an utterly vapid and ignorant view. In the rest of the article Lanier goes on to argue that it's all just "simple mathematics". Yeah, yeah, and computers are just simple NAND gates, amirite? I've waffled over the years…

It’s fair to say that LLMs are surprisingly good at predicting the next term in a series, in a way that we did not expect.

Calling this reasoning, implies a greater set of abilities than what LLMs have, and this skews conversations.

We end up spending time discussing implications and untangling wires due to this anthropomorphization.

Re: There is no A.I.

#435

Earlier quoted context omitted.

Thank you, the philosophers and religious people are more than welcome to discuss what is "true" intelligence or humanity I'm more focused on the fact that GDP as it was measured in 2020 is going to 100x in the next 5 years.

Currently LLMs cannot construct factories, grow crops, or transport goods. Even if it leads to brilliant inventions that were otherwise impossible before, it takes a while for humans to physically make or do things. How confident are you in your prediction? I’d be willing to make a wager with you that would pay off big if you’re right.

I will not need more money if I'm right. Were you offering to wager something other that I would value if I'm right?

Re: There is no A.I.

#436
post #318

> If the new tech isn’t true artificial intelligence, then what is it? In my view, the most accurate way to understand what we are building today is as an innovative form of social collaboration. What an utterly vapid and ignorant view. In the rest of the article Lanier goes on to argue that it's all just "simple mathematics". Yeah, yeah, and computers are just simple NAND gates, amirite? I've waffled over the years…

But it’s true, it is a form of social collaboration. It’s based on a vast training dataset consisting entirely of things humans have written. It’s tuned via reinforcement learning via hiring thousands of people, many of whom are paid poverty wages. And it is further refined by the daily training it gets from its hundred million human users. The core of the article is in the title. The tech is artificial but it’s not…

A new artificial (or otherwise) intelligence would also be a new form of social collaboration. And we shouldn't really insist that "its not intelligent" is true, when the semantics of intelligence is so loaded/vauge/unclear.

Maybe it isn't intelligent, but just because it is "statistics" doesn't imply that. Everything that can be thought can be thought with just NAND gates, Our brain is just atoms bouncing around.

Re: There is no A.I.

#437

Earlier quoted context omitted.

> Is this supposed to be reassuring? I have no idea why people think AIs will be better at general intelligence than they are at driving cars. At least why they think GI will improve any faster than driving.

You have no idea why a software system with very few physical constraints and which can iterate incredibly fast might improve faster than an algorithm that has to drive a physical vehicle through a world full of edge cases?

The fact that we have zero awareness about the edge cases in the world of general intelligence doesn't give us right to think there aren't any. The world of the minds isn't necessarily simpler than physical world and we know about it even less.

For example think of all cognitive biases and myriad of illusions we have to struggle very hard with when we apply out minds to any problem and want to have any hopes of achieving even remotely useful results.

AI alarmists speak of human minds as if they are some perfect tool of progress completely forgetting that those minds failed to achieve any significant progress for tens of thousands of years and what pushed us forward wasn't more intelligent mind. Just a practice of how to deal with shortcomings of our minds.

Another thing is resource cost. Look at how much it costs to run a single instance of semi-retarded LLM. I'd say we are lucky if we are only one computational revolution away from AGI like the one we got from discovering semiconductors.

Humanity dreams of golems for millenia. First we made them out of sticks and clay. Now we are making them out of iron and silicon which is a significant progress but I'm not sure how many such technological jumps we'll need to enjoy or suffer the creation of actual golems.

Re: There is no A.I.

#438

Earlier quoted context omitted.

The attention matrix/matrices are being updated as you talk to it, within the context window.

False. They are being updated during the response only. Edit: false-ish I suppose. You’re not wrong. But it doesn’t persist in a way people would assume and consider meaningful.

Yeah, I was referring to dialogue style interaction, like ChatGPT. It DOES have an internal state depending on version of from 4000 to 32000 tokens. It’s wiped clean between sessions, like a Meeseeks ceasing to exist once its task is finished.

Re: There is no A.I.

#439

Earlier quoted context omitted.

Only in a very anthropocentric sense. How do we know an ant colony doesn’t reason abstractly (or a human town for that matter)? What about slime mold or ameba? Both can solve a maze as well as humans. What makes you think a forest ecosystem isn’t capable of abstract though? It is only if we narrow thought to mean precisely human-like thought when humans and human creations are uniquely capable of something. To that e…

> Only in a very anthropocentric sense. yes, that is the sense in which we are discussing intelligence in order to debate whether the human brain and LLMs operate on similar phenomena

The fact that both humans and LLMs can both reason abstractly is an uninteresting fact if we define “abstract reasoning” to be exactly what humans do, and then create models with the goal of recreating exactly that. This is than simply a statement of an accurate model, and the word intelligence is there only to confuse.

This would be like finding a flower which produces a unique fragrance, then create a perfume which approaches the same fragrance and then conclude that since these are the only two things in the universe which can create this fragrance there must be something special about that perfume.

Re: There is no A.I.

#440

Earlier quoted context omitted.

Well, and let's not treat it like dancing bear either. I've been following AI/ML for years so I'm not blown away like so many other people. I know enough I'm also not mystified by the current popular form. Which is to say there have been previous. LLMs are not AGI but still "a form of AI". Unfortunately AI has become a buzzword, actually it has been for years. While many are just coming to the party I feel a sense of…

This is the problem right here AI as a definition is now fairly meaningless. Are they self aware ... no and they're not designed to be. Can they do intelligent things?... well yes when they're trained to with huge amounts of probably flawed data. I use these models everyday to look up answers to coding problems which prior to this required minutes of wading through ad riddled noise.

I didn't think stackoverflow was that bad. Ba dum tiss. :)
Post reply on HN