Probably another case of "I don't want people training AI on my posts/images so I'm nuking my entire online existence".
micheal65536
Without knowing anything about this model or what it was trained on or how it was trained, it's impossible to say exactly why it displays this behavior. But there is no "hidden layer" in llama.cpp that allows for "hardcoded"/"built-in" content.
It is absolutely possible for the model to "override pretty much anything in the system context". Consider any regular "censored" model, and how any attempt at adding system instructions to change/disable this behavior is mostly ignored. This model is probably doing much the same thing except with a "built-in story" rather than a message that says "As an AI assistant, I am not able to ...".
As I say, without knowing anything more about what model this is or what the training data looked like, it's impossible to say exactly why/how it has learned this behavior or even if it's intentional (this could just be a side-effect of the model being trained on a small selection of specific stories, or perhaps those stories were over-represented in the training data).
IMO, local LLMs lack the capabilities or depth of understanding to be useful for most practical tasks (e.g. writing code, automation, language analysis). This will heavily skew any local LLM "usage statistics" further towards RP/storytelling (a significant proportion of which will always be NSFW in nature).
Stable Diffusion 2 base model is trained using what we would today refer to as a "censored" dataset. Stable Diffusion 1 dataset included NSFW images, the base model doesn't seem particularly biased towards or away from them and can be further trained in either direction as it has the foundational understanding of what those things are.
There doesn't appear to be a model anywhere, unless that has been published completely separately and not mentioned anywhere in the code documentation.
So... If this doesn't actually increase the context window or otherwise increase the amount of text that the LLM is actually able to see/process, then how is it fundamentally different to just "manually" truncating the input to fit in the context size like everyone's already been doing?
I tried getting it to write out a simple melody using MIDI note numbers once. I didn't think of asking it for LilyPond format, I couldn't think of a text-based format for music notation at the time.
It was able to produce a mostly accurate output for a few popular children's songs. It was also able to "improvise" a short blues riff (mostly keeping to the correct scale, and showing some awareness of/reference to common blues themes), and write an "answer" phrase (which was suitable and made musical sense) to a prompt phrase that I provided.
Someone explain to me why there are so many frameworks focused on LLM-based "agents" (LangChain, {{guidance}}, and now whatever this is) and how these are practically useful, when I have yet to find a model that can even successfully perform a simple database query to answer an easy question (searching for one or two items by keyword, retrieving their quantity, and adding the quantities together if applicable) regardless of the model, prompt template, and function API used.
To be honest, the same could be said of LLaMa/Facebook (which doesn't particularly claim to be "open", but I don't see many people criticising Facebook for doing a potential future marketing "bait and switch" with their LLMs).
They're only giving these away for free because they aren't commercially viable. If anyone actually develops a leading-edge LLM, I doubt they will be giving it away for free regardless of their prior "ethics".
And the chance of a leading-edge LLM being developed by someone other than a company with prior plans to market it commercially is quite small, as they wouldn't attract the same funding to cover the development costs.
IMO the availability of the dataset is less important than the model, especially if the model is under a license that allows fairly unrestricted use.
Datasets aren't useful to most people and carry more risk of a lawsuit or being ripped off by a competitor than the model. Publishing a dataset with copyrighted content is legally grey at best, while the verdict is still out regarding a model trained on that dataset and the model also carries with it some short-term plausible deniability.
There are only a few popular LLM models. A few more if you count variations such as "uncensored" etc. Most of the others tend to not perform well or don't have much difference from the more popular ones.
I would think that the difference is likely for two reasons:
-
LLMs require more effort in curating the dataset for training. Whereas a Stable Diffusion model can be trained by grabbing a bunch of pictures of a particular subject or style and throwing them in a directory, an LLM requires careful gathering and reformatting of text. If you want an LLM to write dialog for a particular character, for example, you would need to try to find or write a lot of existing dialog for that character, which is generally harder than just searching for images on the internet.
-
LLMs are already more versatile. For example, most of the popular LLMs will already write dialog for a particular character (or at least attempt to) just by being given a description of the character and possibly a short snippet of sample dialog. Fine-tuning doesn't give any significant performance improvement in that regard. If you want the LLM to write in a specific style, such as Old English, it is usually sufficient to just instruct it to do so and perhaps prime the conversation with a sentence or two written in that style.
In that case ChatGPT is correct, it cannot work with links. You will need to download the video transcript (subtitles) yourself and ask it to summarise that. This definitely works, people have been doing it for months.