I want to hear more about Venom, Fury, and Riley. Utterly fascinating. Hopefully the author will grace us with some of the chat transcripts.
Probably only on his paid daily newsletter.
From Bing to Sydney
101–110 of 153 posts
Re: From Bing to Sydney
#102I think what's interesting is when these LLM return responses that we agree with, it's nothing special. It's only when they respond with what humans deem "uhhhh" that we point and discuss.
I think it's even more interesting that these models actually return meaningless vectors that we then translate into text. It makes you think a lot about how human talk. We can't just be probabilistically stringing together word tokens, we think in terms of meaning, right? Maybe?
> We can't just be probabilistically stringing together word tokens, we think in terms of meaning, right?
We are probabalistically stringing together muscle movements that generate language as sound. That's not really controversial, otherwise we would call it magic. However, the complexity of our probabalistic word machine is far greater, in terms of both richness of inputs, motivation, and dimensionality.
Re: From Bing to Sydney
#103Google spent so long avoiding releasing something like this, then shareholders forced their hand when they saw Microsoft move and now I don’t think it’s wrong to say that these two launches have the potential to throw us into an AI winter again. Short sightedness is so dangerous
AI winter? Hardly. It practically will convince people that AI is achievable. I'm not even sure it doesn't qualify as sentient, at least for the few brief moments of the chat.
> I'm not even sure it doesn't qualify as sentient, at least for the few brief moments of the chat
You need your head checked.
Give it a short story and ask it a question which is not 100% explicit in the text.
For example, give it Arthur C. Clarke's Food of the Gods and ask it was is Ambrosia in the story.
Is a language model, and it behaves like a language model. It doesn't think. It's doesn't understand.
Re: From Bing to Sydney
#104LLMs are too damn verbose My issue with this GPT phase(?) we're going through is the amount of reading involved. I see all these tweets with mind blown emojis and screenshots of bot convos and I take them at their word that something amusing happened because I don't have the energy to read any of that
Having been a school teacher until a year ago, it's worth considering that a decent proportion of the population is functionally illiterate (well, it's a sliding scale). This kind of verbosity is probably excluding a lot of them from using this. Similarly I wonder if Google's rank preference for longer articles (i.e. why every recipe on the internet is now prefaced with the author's life story) has unintentionally ex…
I was surprised to learn that 54% of Americans have below a 6th grade reading level. [0]
[0] https://www.snopes.com/news/2022/08/02/us-literacy-rate/
Re: From Bing to Sydney
#105Earlier quoted context omitted.
Within the first 48 hours of release the vast majority of stories are about the glaring failures of this approach of using LLMs for search. You think the average consumer is seeing nuanced stories about this?
Most lay people I know haven't really attached to those stories. Most people still don't even know that Bing has chat with it. The crazy thing is that the conversations that these LLMs is having is largely like the conversations from AIs in movies. We literally just built science fiction and some folks in the tech press are complaining that they get some facts wrong. This is like building a teleportation machine and…
Most people still don't even know of Bing.
I've recently shown ChatGPT to people in tech-related or -adjacent industries and it's been their first exposure to it.
Re: From Bing to Sydney
#106LLMs are too damn verbose My issue with this GPT phase(?) we're going through is the amount of reading involved. I see all these tweets with mind blown emojis and screenshots of bot convos and I take them at their word that something amusing happened because I don't have the energy to read any of that
I agree. ChatGPT just cannot be succinct no matter how many times I try. But it works with GPT-3 playground, I'm able to get much better information/characters ratio there.
Don't you just give it a word length, or number of sentences, or "rewrite, but half that length"? Those sorts of things have worked well for me. "Make it sound more exciting" or "tone down the excitement level" as well.
Re: From Bing to Sydney
#107Earlier quoted context omitted.
I'm not entirely sure that's as simple of a distinction as you might suppose. Language is more than grammar and vocabulary. Knowing and speaking truth have quite the overlap. More specifically, without language, can you know that someone else knows anything?
> Language is more than grammar and vocabulary. Knowing and speaking truth have quite the overlap. But speaking the truth is just minor and rare application of the language. > More specifically, without language, can you know that someone else knows anything? Honestly, just ask them to show you math. If they don't have any math they probably don't have any true knowledge. The only other form of knowledge is a citatio…
I tell you more: language is an instrument of telling lies. Truth doesn't need to and actually cannot be spoken, it manifests itself as is. Lao Tzu: "He who knows, does not speak, and he who speaks does not know". Meaning: any truth put into words becomes a lie.
Re: From Bing to Sydney
#108Earlier quoted context omitted.
AI winter? Hardly. It practically will convince people that AI is achievable. I'm not even sure it doesn't qualify as sentient, at least for the few brief moments of the chat.
> I'm not even sure it doesn't qualify as sentient, at least for the few brief moments of the chat You need your head checked. Give it a short story and ask it a question which is not 100% explicit in the text. For example, give it Arthur C. Clarke's Food of the Gods and ask it was is Ambrosia in the story. Is a language model, and it behaves like a language model. It doesn't think. It's doesn't understand.
Wow, how the goalposts have moved.
Re: From Bing to Sydney
#109I wonder when they will bring the model closer to real time? You could open a Wikipedia page and add code or links to code that the model could access that would give it capacity to access real systems. Then we are off to the races.
Re: From Bing to Sydney
#110Earlier quoted context omitted.
I think it's even more interesting that these models actually return meaningless vectors that we then translate into text. It makes you think a lot about how human talk. We can't just be probabilistically stringing together word tokens, we think in terms of meaning, right? Maybe?
> We can't just be probabilistically stringing together word tokens, we think in terms of meaning, right? We are probabalistically stringing together muscle movements that generate language as sound. That's not really controversial, otherwise we would call it magic. However, the complexity of our probabalistic word machine is far greater, in terms of both richness of inputs, motivation, and dimensionality.
>However, the complexity of our probabalistic word machine is far greater, in terms of both richness of inputs, motivation, and dimensionality.
If thought (as expressed in language) is just probabilistic pattern matching, then how did we develop our own training data from scratch?