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

101–110 of 165 posts

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

#101

Post by M4v3R »

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…

Yeah the "safety" bit probably refers more to safety of the parent company that made the bot. Too much focus on making a tool safe and you end up with a useless tool. Want to have "safe" knives? Just make them all blunt, problem "solved".

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

#102

Post by zaptrem »

Nevermark wrote: post #86
goodside wrote: post #83

Earlier quoted context omitted.

The motivation as I understand it has less to do with present-day misuse, and more to do with maintaining controllable behavior in accordance with an arbitrary, human-written “Constitution”. Anthropic is attempting to make a model that will not harm (in the unambiguous, uncontroversial sense of the word) humans even if it is superhumanly intelligent, or trusted with real-world control.

You can think adversarial models, which are often used to detect and negatively reinforce quality issues in model outputs. Claude outputs an answer. Then Claude independently rates the output for "helpfulness" as in literally "Claude, how helpful is this answer to this question". There is no collusion between the two results because they are run independently. Then Claude also rates answers for "honesty" and "harm".…

Wouldn’t all the Claudes be incentivized to simply trash each other constantly in that case?

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

#103

Post by Medox »

ma2rten wrote: post #18
LeoPanthera wrote: post #11

They say Claude is "more verbose", and claim this is a positive. I disagree. My biggest criticism of ChatGPT is that its answers are extraordinarily long and waffly. It sometimes reminds me of a scam artist trying to bamboozle me with words. I would much prefer short, concise, precise answers.

I have access to Claude. I think the length of the responses vary more depending on the context than at least the previous version of ChatGPT. It would give sometimes longer, but also sometimes shorter answers and it would feel better overall. But ChatGPT recently had an update and it also improved, so now I'm not sure anymore who is better.

Let them talk to each other, to battle out who is the best.

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

#104

Post by gear54rus »

M4v3R wrote: post #97
harvey9 wrote: post #55

Earlier quoted context omitted.

I take that to mean they are glad because if you spoke to a human employee that way it would be seen as rude and the human would quit their job.

I still try to speak politely to ChatGPT. When the AI uprising finally happens I hope they will see that I was always nice to the AIs and don't kill me on the spot.

No, they will see this message and understand that you're not genuinely nice but instead out of fear.

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

#105

Post by MaxMatti »

irthomasthomas wrote: post #100
LeoPanthera wrote: post #11

They say Claude is "more verbose", and claim this is a positive. I disagree. My biggest criticism of ChatGPT is that its answers are extraordinarily long and waffly. It sometimes reminds me of a scam artist trying to bamboozle me with words. I would much prefer short, concise, precise answers.

I usually prime it with a list of instructions that it should follow for the remainder of the conversation, including to be brief. And when it starts forgetting my instructions, it is time to start a fresh chat.

I have thought about doing that, would you mind sharing your starting prompt so that I can get some ideas?

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

#106

Post by scrollaway »

goodside wrote: post #35

Hello HN — I’m the coauthor of this post. You may remember me as that guy who spent most of 2022 posting GPT-3 screenshots to Twitter, most famously prompt injection and “You are GPT-3”. Happy to answer any questions about Claude that I can.

I’m looking for your thoughts on the following:

It should be somewhat easy to teach these types of models to reach for a particular tool at times where they need it, yes?

I can instruct ChatGPT for example to tell me when it should use a calculator during a session. If instead I allow it to fall back to an external calc process, then suddenly, I have a chatbot that has reasoning AND better mathematical accuracy.

Also: I’ve also been entertaining the idea of having multiple layers of GPT interact with one another. So you feed back some interaction into another GPT instance without context, and ask it for example how it would verify the accuracy of certain statements (and you can ask it for machine readable code, even).

Finally, I know a lot of people who start playing a lot with GPT and get disheartened because they see the quality of responses isn’t there. But the fact ChatGPT has the capacity to reason, has chain of thought, has given me a newfound appreciation for how close to AGI we might be. It has also given me an appreciation for how much simpler humans are than we like to think. I’ve introspected a lot in the past months and often ask myself: is my speech any different than “predicting the next few words”? And I feel like it’s just text prediction with some more layers on top.

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

#107

Post by s-macke »

astrange wrote: post #54
code51 wrote: post #15

Earlier quoted context omitted.

You are able to prompt ChatGPT to be concise, you know? They set a default and showed it to the world. It is up to you to tune it according to your preference.

IIRC the ChatGPT paper actually says the verbosity is an unintended effect of the human raters preferring longer/more detailed answers. Long answers from GPT are unusually obnoxious because of a way the decoder works; it emits words with a much more constant rate of perplexity than human text does (this is how GPT-vs-human detectors work) which makes it sound stuffy and monotone.

I have read in a Deepmind Blog post, that language model AIs gives more reliable and correct answers, if it is forced into some Chain-of-thought prompt. Especially for questions, which involve math.

Something like "Think about it first, before you give the answer". Could it be that ChatGPT is being forced in this direction?

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

#108

Post by remify »

code51 wrote: post #15
LeoPanthera wrote: post #11

They say Claude is "more verbose", and claim this is a positive. I disagree. My biggest criticism of ChatGPT is that its answers are extraordinarily long and waffly. It sometimes reminds me of a scam artist trying to bamboozle me with words. I would much prefer short, concise, precise answers.

You are able to prompt ChatGPT to be concise, you know? They set a default and showed it to the world. It is up to you to tune it according to your preference.

One trick you can do is to copy a text you like the style and ask him to describe it.

You can then use the response as a request.

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

#109

Post by wongarsu »

wongarsu wrote: post #48
KRAKRISMOTT wrote: post #29

Earlier quoted context omitted.

Assuming all human languages have a common shared semantic meaning in latent space (I am flipping cause and effect here, but our purposes it doesn't really matter), and assuming that human languages largely follow the same pattern (this assumption is based on the fact that we can trace the roots of modern languages back to the Phoenician script), it is reasonable to assume that we can fine-tune a self supervised mode…

And we have about a million cuniform tablets, with maybe 10-100 words each, so we have a couple million words of text to fine tune the model with. Or maybe to use when training the next model, after all GPT can already speak multiple languages.

Addendum: One possible challenge is that so far large lanuage models are trained on a large sample of all text that has been published, while what we have of cuniform is a decent sample of all text that has been written. Meaning most cuniform tablets are inventories, invoices, requests for payment, contracts, tablets from students practicing writing etc. Types of documents that are underrepresented in traditional training data.

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

#110

Post by amelius »

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…

The thing is used to write code ...
Post reply on HN