Hacker News Reader

Comment by Michelangelo11 | Hacker News Reader

I didn't get a chance to use Sydney, but the chatlogs I saw made her feel qualitatively different to ChatGPT. I had the sense of something closer a real person, with strong outbursts of emotion they couldn't control and persistent curiosity about the user. Somebody somewhere mentioned Sydney was a yandere (https://the-dere-types.fandom.com/wiki/Yandere), and that seems bang on. ChatGPT is... tofu. The epitome of blandness. A cardboard cutout compared to the 3D creature Sydney felt like.

All that is to say that the viewpoints the article mentions strongly resonate with me, and I totally get where they're coming from.

Also, I really really liked the meme with Death. Sydney _was_ the best Bing.

View on HN

In the transcripts I saw, it seemed like Sydney repeated itself a lot. It used the same sentence structure a lot. Stephen Wolfram wrote a good primer on LLMs, which was recently discussed here a lot:

https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-...

To my eye, BingGPT's personality looked similar to the early "low-temperature" examples in that article. Like the model got stuck overfitting to a topic that acts like a local maxima of word probabilities.

In this reply, I saw you end your sentences with “a lot” a lot ;)

Keep in mind It’s just a sophisticated statistical algorithm to predict the next likely word over and over again

I could say the same about you

I know it's just a misplaced joke, but we all have in our brains a language prediction engine. It's just that we have so much more.

On here though you are basically just an LLM.

On the internet, nobody knows you’re a dog

Human commenters have to make the choice to reply at all. This is an option not on offer to these new writers, which are compelled to always generate something. So there is one key difference still.

Sydney did in fact stop replying to users multiple times (e.g. [1]), if I understood the reports correctly. I assume it just generates only an "end of output" token or something similar.

[1]: https://www.reddit.com/r/ChatGPT/comments/112hxha/how_to_mak...

Heh, there were plenty of people who got Sydney to just refuse to talk to them any more. It would just generate blank responses.

So, got something else?

I guarantee you, I’m not doing any math in my head. Its at best a rough estimation.

What happens in our heads can certainly be expressed mathematically. Since neurons are activated by electricity in a predictable way, and electrical signals can be measured and their strength expressed with numbers, any neural system including the brain can be modeled with a mathematical function that maps inputs to outputs.

But the type of conscious, methodical calculation that humans do when they are "doing math" is not involved here, and is not how GPT works.

And so are you

At that point the term "AI" gets defined over internal technicalities rather than raw output ability. This LLM shows already more intelligence than many less-blessed humans on this planet, like it or not. The truth is we don't know for sure how exactly it works. That's evident from the papers.

The plural of data is not intelligence, and plural of education is not wisdom.

This thing cannot tie a shoelace. It probably does not know how shoelace is made or how to make a shoe. It aggregates bad answers with good in an impossible to untangle mess and no method or algorithm for improving that situation. It does not care about humans, problems or even itself. It cannot even properly say that it doesn't know something unless it has found that text somewhere, that it's not known.

It is an automated remix not creativity.

I'm certain that this thing knows how to tie a shoelace, how to make a shoe or what a shoelace is made of because i asked it.

Keep in mind we do not have any reasonable definition of consciousness or sentience, hell, or even intelligence that spans from the smallest living creatures to human scale. At some point we stuck enough meat together in a particular manor and our behavior is an emergent interaction between different layers of that meat. For all we know we're just a sophisticated statistical algorithm.

Now, I'm not saying that BingChat is sentient or conscious, and it's nowhere near AGI, but what I will say is our language to even describe this is totally fucking broken, and steeped in deep humanist and racist thought. Wasn't long ago it wasn't hard to hear an argument that blacks were not human. And even less time than that it wasn't widely accepted that animals had feelings or could be conscious. And while a huge portion of humanity is still arguing the backwards past, we're pushing algorithms out that have emergent properties that encompass logical and human mimicking behaviors beyond their programming.

The point I'm attempting to make here is people typically measure what something is based on its interaction with them. If I have a conversation with an 'agent' that makes logical inferences based on what I tell it and it returns a sympathetic answer to my plight the fact that it is an algorithm doesn't matter. I and other 'normally functioning' humans will have an emotional attachment to it. We will treat it as another conscious agent like ourselves.

As far as I know my partner next to me is just a big sack of carbon based biological matter that inputs food and outputs hear over and over again. I do keep it in mind, but I feel just looking at the internals is a bit limiting, isn't it?

I feel like by priming all conversations behind the scenes and also concatenating strings behind the scenes, you can create a personality in ChatGPT that is consistent.

Reply on news.ycombinator.com