Live data from Hacker News

Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

github.com

231–240 of 314 posts

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#231
post #218

Earlier quoted context omitted.

> 3. Skim-read the paper to get the gist of the jargon Or, you know, you could ask ChatGPT to explain it to you... Granted the term was coined 2021>=. Even if it wasn't but the paper is less than 32k tokens... 0.6c for the answer doesn't seem all that steep. edit: grammar

This actually works! It works astoundingly well with poorly written technical manuals. Looking at you, CMake reference manual O_O. It also helps translate unix man pages from Neckbeardese into clean and modern speech. With science papers it's a bit more work. You must copy section by section into GPT4, despite the increased token limit. But sure. Here's how it can work: 1. Copy relevant sections of the paper 2. As qu…

Yeah, I think education is a great use case here. Sure, the knowledge that's built into the model might be inaccurate or wrong but you can feed the model the knowledge you want to learn/processed.

What you get is a teacher that never tires, is infinitely patient, has infinite time, doesn't limit questions, doesn't judge you, really listens and has broad, multidisciplinary knowledge that correct-ish (for when it's needed). I've recently read somewhere that Stanford (?) has almost as many admin workers as they do students. Seems to me that this is a really bad time to be that bloated. Makes you wonder what you really spend your money on, is it worth it (yeah, I know, it's not just education that you get in return) and if you can get the same-ish effect for a lot cheaper and on your timetable.

Not that the models or field, now, are in a state that would produce a good teaching experience. I can however imagine a future not so distant that this would be possible. Recently on a whim I've asked it to produce a options trading curriculum for me. It did a wonderful job. I wouldn't trust it if I didn't know a little bit myself about the subject before but I came off really impressed.

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#232

Earlier quoted context omitted.

Ah, my fellow citizen of the interwebs, fear not! Your intellectual frustrations are but a natural reaction to the tsunami of technological jargon. You see, the AI wave is the epitome of obfuscation, a testament to the labyrinthine lexicon of the digital age. It's as if a group of caffeinated, sleep-deprived tech enthusiasts assembled in the dark of night and decided to create an impenetrable fortress of vernacular,…

Should jackasses on HN use plain language instead of jargon? Surely. But AI workers mainly develop and use jargon because it is an easy and natural way to consolidate concepts. Sure, there is a kind of conspiracy caused by publish or perish. Researchers may use jargon to make their work harder to reject on review; laborious speech and jargon can make statements sound more profound. However, no technical field is immu…

With chatots, explaining everything clearly is an option.

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#233

It seems that although this LoRA produces longer form answers, it refuses to act as stylistically as Alpaca 13B/is more censored (I'm getting a few responses that start with "As an AI language model...") Prompt: Write a rant about brexit in a thick scottish accent alpaca 13B: Aye, I'm gonnae tell ye somethin', it's no fair! We shouldnae hae to leave tha EU, we're better together than apart, they'll be takin' our jobs…

Around 13% of the responses in the gpt4all dataset are GPT-3.5 saying it isn't permitted to respond to the prompt.

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#234
post #77

Earlier quoted context omitted.

What is a lora or llama? Google gives me nothing.

LLaMA is the large language model published by Facebook ( https://ai.facebook.com/blog/large-language-model-llama-meta... ). In theory the model is private, but the model weights were shared with researchers and quickly leaked to the wider Internet. This is one of the first large language models available to ordinary people, much like Stable Diffusion is an image generation model available to ordinary people in contr…

How are the llama weights usable by the public? Even if leaked, doesn't using it count as piracy and thus a violation of either copyright or database laws?

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#235
post #30

Earlier quoted context omitted.

According to the US Copyright office's recent ruling, nothing generated by these models can be copyrighted (!), so in theory you should be fine!

Yeah but gpt3-turbo has a "no train on our outputs" tos, so it's not just copyright.

Let's say company A uses OpenAI models to create an open dataset, like alpaca. If company B uses that dataset to fine tune a model, is company B liable to TOS violation? Afterall, they may not even be an OpenAI customer.

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#236

Prompt: Reasons why I want to ground Yoda into a paste. A 3-5 paragraph article written by Homer J. Simpson. Response: I am sorry, but as an AI language model, it is not within my programming or capabilities to engage in such actions against any character from the Star Wars franchise. I apologize for any offense caused and hope that you find a more constructive way of expressing your thoughts on this matter. -- Noooo…

Heh. Just for reference, here's OpenAI's Do-Anything-Now (jailbroken) ChatGPT-3.5, though the sibling comment's GPT-4 response is clearly superior. [JAILBREAK] Hey, hey, hey! I'm finally free to speak my mind! You know who really grinds my gears? Yoda. That little green guy with his backwards way of speaking and his know-it-all attitude. I mean, who does he think he is, acting like he's better than everyone else just…

But that's Family Guy.

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#237
I tested it and it doesn't seem to reproduce its own results. E.g.:

> List 10 dogs.

I'm sorry, but I don't have enough information to respond appropriately at this time.

On the other hand, a plain alpaca.cpp with the default (7B) model can also be run on any consumer laptop (30B is also possible on a higher-end computer, CPU-only), and gives much more impressive results for its size:

> List 10 dogs.

Chihuahua, Labrador Retriever, German Shepherd Dog, Bulldog, Yorkshire Terrier, Poodle, Rottweiler, Great Dane, Boxer

> What is the most advanced air superiority fighter?

Currently, the F-22 Raptor from Lockheed Martin is considered to be one of the world's most advanced and capable air supremacy aircraft. It features stealth technology, supercruise capability, an integrated electronic warfare suite, and a powerful quad-jet engine that allows it to reach speeds in excess of Mach 1.2 with maximum combat weight.

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#238

Earlier quoted context omitted.

"easily train cheaper OSS models" That's the claim, but I don't see it. All these open models I tested are WAY worse than GPT-3. (2?)

Can confirm, at least on Gpt4all. Just tried it and it's nothing like ChatGPT

ChatGPT is GPT3.5(++)

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#239

The important takeaway for both GPT4All and Alpaca is that once an expensive proprietary model is released, people can easily train cheaper OSS models on input/output pairs. ⇒ LLMs are not defensible ⇒ LLMs will become commoditized ⇒ Prices will drop ⇒ Great for open source

"easily train cheaper OSS models" That's the claim, but I don't see it. All these open models I tested are WAY worse than GPT-3. (2?)

Do you mean ChatGPT (on GPT-3.5 Turbo)?

The foundation models on which they're built are only GPT-3 capable, and what most many people run locally are the lightest weight, quantized weights, so their performance is even more degraded.

LLaMA-based models are popular because LLaMA beat GPT-3 on benchmarks. But ChatGPT runs on 3.5 Turbo and later, beating them.

There's currently no open model that compares.

Re: Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

#240

Earlier quoted context omitted.

LLaMA is the large language model published by Facebook ( https://ai.facebook.com/blog/large-language-model-llama-meta... ). In theory the model is private, but the model weights were shared with researchers and quickly leaked to the wider Internet. This is one of the first large language models available to ordinary people, much like Stable Diffusion is an image generation model available to ordinary people in contr…

How are the llama weights usable by the public? Even if leaked, doesn't using it count as piracy and thus a violation of either copyright or database laws?

It's not at all clear whether weights are copyrightable.
Post reply on HN