Community to discuss about Llama, the family of large language models created by Meta AI.
There are 4 ways to achieve the mission as far I understood:
π
AFAIK there is none, and there is a good reason why. I built a small guide for my company 2/3 months ago. Most of it is now in some version of deprecated, the field is moving extremely quickly and every general guide needs to be updated constantly. For now the only way I see is to use all the inevitably somewhat deprecated guides out there and keep your mind open for updates.
I consider myself somewhat obsessed. The breakthrough seems to have slowed, we aren't experiencing daily breakthroughs. Seems to be more like weekly now, but still far too fast for any aggregation document to stay useful beyond 2 months.
In two months from now half of what's written would no longer be relevant.
The best noob-accessible explanation of LLMs I've found so far: https://blog.rfox.eu/en/Programming/How_to_run_your_own_LLM_GPT.html
The most entertaining (IMHO) explanation, which is (at best) 60% accurate: https://www.reddit.com/r/LocalLLaMA/comments/12ld62s/the_state_of_llm_ais_as_explained_by_somebody_who/
Trial and error
Sure, anything else? :D Even regulars here don't know how to keep up. You can try asking Bing.
Pretraining = Unsupervised Learning
Fine Tuning = Supervised Learning
Human Feedback = Reinforcement Learning
These three steps produce what are currently calling AI.
A modern LLM like Mistal7B is made of 32 layers of 4096x4096 transformer nodes
In pretraining,
Coherent data is fed thru the network one word at a time (in this case the entire internets text) and the models node-connection-weights are automatically adjusted towards the values such that given a list of words it correctly predicts the next one.
In finetuning,
This time Data Pairs are fed thru, (example prompt AND example correct answer) this bangs the model over the head and forces it to respond to our prompt formatting, it's also where we make it helpful and do what it's told.
In Human Feedback,
(Abbreviated to RLHF) We let the model mutate slightly, having it generate multiple responses with slightly differing internal weights and having actual humans select their favorites, over time this draws the model towards not just generalizing from text examples, but also towards actually pleasing humans with words (what ever that process might entail)
All intelligence emerges during the pure prediction/pretraining stage, Finetuning and RLHF actually damage the model SIGNIFICANTLY! but working with pure text prediction engines requires much more though than simple prompt engineering.
There's a strong mathematical proof that states Modeling == Prediction == Compression == Intelligence
Saying It's essentially impossible to get any one of these without also getting the other three,
Accurate modeling provides Prediction (by simply running the model forward in time), Accurate Prediction provides Compression (by only storing the difference from the prediction)
And intelligence (I.E. Getting what you want) is simply a mater of using your Compressed Model of the world to Predict what might happen if you performed various actions and selecting the one where you get what you want.
Modern Open Source LLMs (like OpenChat, DarkSeekCoder etc) are actually superior to GPT4 now...
The reason they might seem behind to casual users is the RLHF step which GPT4 has received A TON OF!
This is an expensive step requiring many peoples time, and all open models simply skip this.
The thing is using techniques (like simply asking a million times) we find the knowledge and skill in these OS models are CLEARLY far and beyond the latest available OpenAI GPT model (its 11/9/2023 now)
Hope that helps! model LLM based AI is actually extremely simple, we create an intelligence beast using prediction and then we bang it over the head (giving it serious literal brain damage) to behave for us, then we listen closely to it and slap it in the face for the tiniest mistake until we're happy with it.
It's still just an insanely high dimensional word-by-word predictor, it's just been traumatized by humans to please us.
This pretty much sums it up
Enjoy!
I really don't like this at all because it really is a gross misrepresentation of what's really going on under the hood.
It's really a tool used to distort the reality of the mechanisms occurring under the hood, which are actually interpretable.
The news on LLM's are just like new models that are 99.9999% of ChatGPT-4 this hour, but in two days nobody even remembers them.
