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

Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

chatllama.baseten.co

Post reply on HN
Print view View original

201–210 of 226 posts

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#201

Post by KennyBlanken »

californical wrote: post #189
vishnuharidas wrote: post #178

Me: Is LLaMA as good as ChatGPT? LLaMa: No, LLaMA is not as good as ChatGPT. ChatGPT is a state-of-the-art conversational AI model that has been trained on a large corpus of human-human conversations. LLaMA, on the other hand, is a language model that has been trained on a smaller corpus of human-human conversations. Me: So you were looking at people chatting in Facebook? LLaMa: Yes, I was. Me: That means Facebook is…

That doesn’t make it true. LLMs don’t know how they were trained. It’s just making that story up. It may be true, but the fact that the model said that is meaningless.

Yep. Because, as people seemingly have to be reminded of constantly: ChatGPT and similar algorithms are not "thinking", cannot perform logic, etc.

They are very complex statistical databases of language.

LLaMA doesn't seem particularly sophisticated. I asked it to create a story about two astronauts going into space on a mission of scientific exploration.

Ten guesses what their spaceship was called. Hint: "no bloody letter..."

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#203

Post by Jensson »

satellite2 wrote: post #127

It seems it gets stuck in pseudo infinite loops sometimes. Can you provide a step by step guide on how to get dressed for a five years old? 1. Start by putting on underwear and socks. 2. Put on a t-shirt. 3. Put on pants or shorts. 4. Put on shoes or sandals. 5. Put on a jacket or sweater. 6. Put on a hat or scarf. 7. Put on gloves or mittens. 8. Put on a coat. 9. Put on a hat or scarf. 10. Put on shoes or sandals. 1…

ChatGPT makes those loops as well, they are just hidden from users as loops are very easy to detect and remove with algorithms.

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#204

Post by Reason077 »

IIAOPSW wrote: post #40
marginalia_nu wrote: post #8

Earlier quoted context omitted.

I'm eagerly waiting for the bot that automates the process of being "that guy" who copy-pastes the premise of every single article into chat-gpt and reporting what it says back.

"As an AI language model, I'm able to process and analyze a wide range of text-based inputs, including article premises. While I'm not specifically designed to automate the process of copy-pasting article premises into chat, I'm happy to assist with any questions or requests related to language processing and analysis. If you have a specific article premise that you'd like me to analyze or provide insights on, feel f…

"While the comments generated by HN_LLaMA may not have the same level of personalization and nuance as those written by humans, they can still provide valuable information and insights that might not have been considered otherwise. In some cases, AI-generated comments may even be able to identify patterns or trends in the data that human commenters might miss.

On the other hand, the use of AI to generate comments can be seen as a form of automation that undermines the authenticity of online discussions. Some users might argue that the value of Hacker News lies in the unique insights and perspectives of its human contributors, and that AI-generated content detracts from this."

-chatGPT

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#206

Post by make3 »

ec109685 wrote: post #173
make3 wrote: post #128

Earlier quoted context omitted.

It just hasn't been prompted or fine-tuned to have the neutral, self effacing personality of ChatGPT. It's doing the pure, "try to guess the most likely next token" task on which they were both trained ( https://heartbeat.comet.ml/causal-language-modeling-with-gpt... ). ChatGPT is further trained with reinforcement from human feedback to make them more tool-like ( https://arxiv.org/abs/2204.05862 & https://openai.com…

The randomness link doesn’t work. All the ChatGPT products need randomness or it can’t be creative.

https://huggingface.co/blog/how-to-generate

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#207

Post by O__________O »

LawTalkingGuy wrote: post #194
O__________O wrote: post #192

Earlier quoted context omitted.

When it comes to litigation, the parties will and resources to engage an aggressive legal adversary is critical aspect of assessing legal dynamics in a matter. For example, even in cases for which there is no direct legal leverage, party might be exposed to patent, copyright, etc infringement if the other party discovers they have a valid existing claim for which the other party is dependent on and access to it is re…

> if LLaMa is free from legal claims for commercial use. That restriction is only in the license though, which was not accepted. So it won't apply, right? > otherwise, [all?] copyright notices and software licenses would be meaningless, because you could simple launder stolen property That applies to trade secrets. If I knowingly induce you to leak a secret I can't use it, but if you leak the secret to the public in…

Happy to be proven wrong, but the weights are meaningless without the parts written by humans. If someone refers to any parts that are protected to write a new interface to the weights it would most likely be violate the law. Using the weights alone, it would be impossible to reverse engineer them using a clean-room to develop new interface to the weights.

>> There's no breach of contract because there was no agreement.

No, this is property laundering. If intentional, it’s a crime. If unintentional, the property owner need just notify the party of their rights, the remedy they’re seeking, and if needed, send a cease and desist.

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#209

Post by dchest »

Faint wrote: post #197
int_19h wrote: post #186

Earlier quoted context omitted.

Jokes aside, has anyone tried fine tuning a model on forum conversations? By which I mean, for chatbots, the data for fine-tuning basically presents it with examples of dialogs. But if you take any given comment on HN and walk its parents, you basically end up with something like that, except there's more than two people involved (usually). So what if you fed that to LLaMA, and then used the same format for the actua…

In that vein, how about stackoverflow? That should give at least straightforward ask-and-answer format, and there's plenty on material to work with.

LlaMa was trained on 78 GB of StackExchange (I assume StackOverflow was included in that).

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#210

Post by LawTalkingGuy »

O__________O wrote: post #207
LawTalkingGuy wrote: post #194

Earlier quoted context omitted.

> if LLaMa is free from legal claims for commercial use. That restriction is only in the license though, which was not accepted. So it won't apply, right? > otherwise, [all?] copyright notices and software licenses would be meaningless, because you could simple launder stolen property That applies to trade secrets. If I knowingly induce you to leak a secret I can't use it, but if you leak the secret to the public in…

Happy to be proven wrong, but the weights are meaningless without the parts written by humans. If someone refers to any parts that are protected to write a new interface to the weights it would most likely be violate the law. Using the weights alone, it would be impossible to reverse engineer them using a clean-room to develop new interface to the weights. >> There's no breach of contract because there was no agreeme…

> weights are meaningless without the parts written by humans.

I think this can be true, but that the format of these is well known.

The torrent consists of params.json, consolidated.*.pth, and tokenizer.model plus some .chk files. Notably, there is one script, llama.sh, which is about 2k so even if it was needed, can't be that complex.

> If someone refers to any parts that are protected to write a new interface to the weights it would most likely be violate the law

Not at all. I can refer to pages and words in a book I don't own.

> No, this is property laundering. If intentional, it’s a crime.

Only if the weights are copyrightable. Otherwise it'd hinge on being a trade secret, the best theory I've seen yet, and that basically says the cat's out of the bag once the public knows something.

Post reply on HN