Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Meet “Claude”: Anthropic’s rival to ChatGPT

scale.com

Post reply on HN
Print view View original

131–140 of 165 posts

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#131

Post by flanked-evergl »

_flux wrote: post #128
flanked-evergl wrote: post #122

Earlier quoted context omitted.

> Remember Microsoft's Tay? Ah, right, it was shut within days because it quickly became an asshole. If we consider the context, which is not something that posts on twitter under Microsoft's brand name, but something you communicate with in private. Who exactly are you worried about here, the person who will prospectively coerce the language model into being an asshole? If they don't want to do that, they could just…

> If I make ChatGPT say something egregious, and post that on Twitter or Facebook, I'm posting it, and I'm liable, just as I would be liable if I used a word processor with spell checking to make text and post it on Twitter or Facebook. While arguably that would be the case, practically it might be a different look for the company whose algorithm generated the text in the first place—and from the audience, which migh…

> I mean surely you have noticed that social media can get stirred up for no fundamentally sound reasons.

Sure, but therein lies the problem, you can't really use reasoning, sound or otherwise, to insulate yourself from fundamentally unsound reasoning. Facebook was responsible for January 6 more than the upstarts that got kicked off AWS, but Facebook had no real negative consequences from their involvement.

I can literally post a screenshot from ChatGPT which shows anything, and given the complexity of how it works (take DAN for example), it is really difficult to dispute authenticity. If this can bring down ChatGPT then OpenAI should just close down now.

In an ideal world, ChatGPT would be value neutral, and fact positive, so if I ask it about fact, it should give me to the best of its rather poor ability a descriptive statement about fact, if I ask it about values, it should give me a descriptive claims about values (e.g. most value systems sees such act as wrong or right), not mask a prescriptive claim as a descriptive claim (e.g. it is wrong or right).

If I ask it to write fictional prose, then it should write what I ask it. If I ask it to write factual prose, it should similarly do what I ask it. I don't really see where this falls apart or turn into 4chan.

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#132

Post by flanked-evergl »

amelius wrote: post #129
flanked-evergl wrote: post #114

Earlier quoted context omitted.

So is my editor, my computer, my fingers, my eyes, my brain, the internet, google, stackoverflow, wikipedia, English, etc.

Yeah, but there is still a human between the code and its execution. I'm sure one of these days someone will hook GPT directly to a terminal.

> I'm sure one of these days someone will hook GPT directly to a terminal.

And then there would be someone between ChatGPT and execution in more or less the same way I am inbetween the code I write and execution of said code, which in most cases I'm not, as after I write the code, the computer executes it without checking with me.

If someone gives ChatGPT unfettered interaction with the internet, they are doing it, and they are liable.

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#133

Post by danielbln »

anhner wrote: post #60

I'm hoping of one day running GPT3/ChatGPT on my local computer, similarly to how one can run Stable Diffusion now. I would love to have a personal conversation with these AI systems, use them as a sort of assistant, without the worry of being spied on. At the moment I can't use it as more than a glorified search engine, because of the privacy implications of running it on the cloud.

There will always be an edge to these massively operated central models that you can't easily run at home, due to compute and cost. Things like 4-bit quantization will help making these foundational models significantly easier to operate on smaller hardware, and of course you might not need a foundational model for manby local usecases, but maybe a smaller, more specialized LLM is enough (those can even be trained by the bigger models).

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#134

Post by viraptor »

flanked-evergl wrote: post #122
viraptor wrote: post #98

Earlier quoted context omitted.

We already had a learning AI without hard limits. Remember Microsoft's Tay? Ah, right, it was shut within days because it quickly became an asshole. ( https://www.theverge.com/2016/3/24/11297050/tay-microsoft-ch... ) It's not even going to be objectively wrong a lot of the time. For example slaves are indeed a pretty efficient way to run a business. It's our limits and ethics that stops (most of) us from doing it. Wi…

> Remember Microsoft's Tay? Ah, right, it was shut within days because it quickly became an asshole. If we consider the context, which is not something that posts on twitter under Microsoft's brand name, but something you communicate with in private. Who exactly are you worried about here, the person who will prospectively coerce the language model into being an asshole? If they don't want to do that, they could just…

> the person who will prospectively coerce the language model into being an asshole?

The language models are trained on internet content already. If you ignore ethics, it means you're feeding it bias and racism along with everything else.

Chatgpt is not blatant about it normally but I'm sure you've seen examples like "write a function that takes a race argument and returns length of prison sentence" where it obviously leaks through. I'm sure there are some non-obvious ways it leaks in other answers too. Sometimes subtle ways you will not realise are biased and will not want to take liability for.

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#136

Post by belter »

These models are using industry ML Algos and known techniques. It's not like some unknown startup suddenly discovered, gradient descent or Deep Learning RNN's and is keeping these confidential. Why would Microsoft consider it worthwhile to even contemplate the possibility of paying $10B for these or similar?

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#137

Post by armchairhacker »

zone411 wrote: post #32

Here is a fun example of what it can do: https://twitter.com/jayelmnop/status/1612243602633068549 .

This example is way better than ChatGPT and actually pretty creative.

However for some of these really good responses I always wonder if you’re example is close to one which has been given “preloaded” responses or explicit reinforcement…because if you ask ChatGPT a common question like “why did the chicken cross the road?” the model’s response seems especially unique and better than usual. Even if the specific question isn’t common, maybe it’s been trained on a more general but still reinforced category, like asking “why did the fox cross the road” would get you almost the same “preloaded” response but with chicken adjectives/verbs replaced with fox ones.

I doubt Claude has been trained on Fast and Furious or movie titles specifically, but perhaps it has been explicitly trained to know what “exaggerated” responses means. Even if not, reinforcement focusing on specific areas may be a good technique for future language models.

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#138

Post by jacooper »

jillesvangurp wrote: post #69
anhner wrote: post #60

I'm hoping of one day running GPT3/ChatGPT on my local computer, similarly to how one can run Stable Diffusion now. I would love to have a personal conversation with these AI systems, use them as a sort of assistant, without the worry of being spied on. At the moment I can't use it as more than a glorified search engine, because of the privacy implications of running it on the cloud.

There will always be this uncomfortable balance between need to know information to be useful being essentially the same as everything an adversary would need to really act against you. With something like chat gpt being plugged into some voice assistant thing and having access to all your documents, emails, and other content, you could imagine having conversations about work content, content creation, calendar manag…

> That's where AI becomes really useful. But the issue of trust is a big one

Well google already has access to all of this data, the difference is unlike google assistant, ChatGPT actually can do something useful.

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#139

Post by phonebucket »

flanked-evergl wrote: post #87

> Claude feels not only safer but more fun than ChatGPT. I may be the minority here, but I really don't concern myself with ChatGPT safety and I am not entirely sure what the reason is why people are very worried about its safetly. It is safer than most things I have in my house, including a kettle, a saw, a hammer, a screwdriver, my actual PC, every kitchen appliance I have. Of course it can be misused, like any too…

> Of course it can be misused, like any tool, but no amount of safety features in ChatGPT will make users of it more or less careful in their use of it. If someone using ChatGPT cares nothing for using it safely then it will likely end poorly, just like it will end poorly if I use a hammer without any care for using it safely.

I'm not worried about hammers when I have them. I'm worried about hammers when someone who wants to hurt me has them.

Same with ChatGPT.

Re: Meet “Claude”: Anthropic’s rival to ChatGPT

#140

Post by otabdeveloper4 »

flanked-evergl wrote: post #87

> Claude feels not only safer but more fun than ChatGPT. I may be the minority here, but I really don't concern myself with ChatGPT safety and I am not entirely sure what the reason is why people are very worried about its safetly. It is safer than most things I have in my house, including a kettle, a saw, a hammer, a screwdriver, my actual PC, every kitchen appliance I have. Of course it can be misused, like any too…

But ChatGPT might exhibit wrongthink tendencies, that would be plus-ungood.
Post reply on HN