# Large Language Models [[2023-06-16]] I've by now worked so much with [[GPT-4]] and others that the older links below seem quite tame. I'm collecting the stuff I've come up with / I'm playing around with in [[SK - 2n0 - Large Language Models]]. The hot shit about [[Natural Language Processing]]. See also: [[LLM Prompt Engineering]] ## Links - [Introduction - Hugging Face Course](https://huggingface.co/course/chapter1/1) - [Scale Spellbook: The platform for large language model apps.](https://scale.com/spellbook) This one is a pretty advanced [[Reverse Engineering]] of [[Github Copilot]]: - [copilot-explorer | Hacky repo to see what the Copilot extension sends to the server](https://thakkarparth007.github.io/copilot-explorer/posts/copilot-internals) - [Model index for researchers - OpenAI API](https://beta.openai.com/docs/model-index-for-researchers) [[Reinforcement Learning]] for [[Large Language Models]] - [Illustrating Reinforcement Learning from Human Feedback (RLHF)](https://huggingface.co/blog/rlhf) ## ChatGPT alternatives Competitors/alternatives to ChatGPT (from [https://twitter.com/goodside/status/1606611869661384706](https://twitter.com/goodside/status/1606611869661384706)) - [Poe](https://poe.quora.com) - [Jasper Chat | AI Chat for Content Creators](https://www.jasper.ai/chat) - [You.com | The AI Search Engine You Control](https://you.com) - Ghostwriter chat: [https://twitter.com/amasad/status/1606139822837338112](https://twitter.com/amasad/status/1606139822837338112) ### Opensource ChatGPT - [GitHub - lucidrains/PaLM-rlhf-pytorch: Implementation of RLHF (Reinforcement Learning with Human Feedback) on top of the PaLM architecture. Basically ChatGPT but with PaLM](https://github.com/lucidrains/PaLM-rlhf-pytorch) - [GitHub - LAION-AI/Open-Assistant](https://github.com/LAION-AI/Open-Assistant) Distributed training: - [GitHub - CarperAI/trlx: A repo for distributed training of language models with Reinforcement Learning via Human Feedback (RLHF)](https://github.com/CarperAI/trlx) ## Papers - [[2212.10001] Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters](https://arxiv.org/abs/2212.10001) - [https://twitter.com/giffmana/status/1608568387583737856](https://twitter.com/giffmana/status/1608568387583737856) ![[Large Language Models-1672359073845.jpeg]] [How does GPT Obtain its Ability? Tracing Emergent Abilities of Language Models to their Sources](https://yaofu.notion.site/How-does-GPT-Obtain-its-Ability-Tracing-Emergent-Abilities-of-Language-Models-to-their-Sources-b9a57ac0fcf74f30a1ab9e3e36fa1dc1) ## Tools From [https://twitter.com/goodside/status/1588247865503010816](https://twitter.com/goodside/status/1588247865503010816): > First up, [@humanloop](https://twitter.com/humanloop) has a Playground that brings variable interpolation to prompts and lets you turn them into API endpoints. Once you're deployed, it also lets you collect past generations along with user behavioral feedback for fine-tunes. [https://humanloop.com](https://t.co/qco5m5JPm8) > [@everyprompt](https://twitter.com/everyprompt) extends Playground in a similar way: Putting variables in prompts and giving you a single button to go from prompt to API. Has nice developer-oriented touches in the UI too — e.g. displaying invisible chars as ghosts. [https://everyprompt.com](https://t.co/YFxDhz7X8G) > [@dust4ai](https://twitter.com/dust4ai) takes some work to wrap your head around, but it's very powerful — gives a collapsible tree UI for representing k-shot example datasets, prompt templates, and prompt chaining with intermediate JS code. Replaces a lot of code around prompt APIs. [http://dust.tt](https://t.co/l4Gi2MEFsO) > Just dropped: Spellbook from [@scale_AI](https://twitter.com/scale_AI) . Great ideas in here — couples prompts with uploaded training data to automatically write k-shots, give evaluation metrics for prompt variants; also lets you turn prompts to spreadsheet functions ### Langchain > [@hwchase17](https://twitter.com/hwchase17)'s Python package Langchain implements and lets you easily compose many published LLM prompting techniques. Implements self-asking, web search, REPL math, and several of my own prompts. - [GitHub - hwchase17/langchain: ⚡ Building applications with LLMs through composability ⚡](https://github.com/hwchase17/langchain) Can help with: 1. LLM and Prompts 2. Chains 3. Data Augmented Generation 4. Agents 5. Memory 6. [BETA] Evaluation ## Cool applications - GPT3() function in excel: [https://twitter.com/shubroski/status/1587136794797244417?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1587136794797244417%7Ctwgr%5E1cef16bc839dcf6f1c8d8a9f33c54f4cad1d0e22%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fwww.geoffreylitt.com%2F2022%2F11%2F23%2Fdynamic-documents.html](https://twitter.com/shubroski/status/1587136794797244417?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1587136794797244417%7Ctwgr%5E1cef16bc839dcf6f1c8d8a9f33c54f4cad1d0e22%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fwww.geoffreylitt.com%2F2022%2F11%2F23%2Fdynamic-documents.html) ![[Large Language Models-1672360027273.jpeg]] ![[Large Language Models-1672360057768.jpeg]] ![[Large Language Models-1672360077512.jpeg]] ![[Large Language Models-1672360098091.jpeg]] ![[Large Language Models-1672360167771.jpeg]] ### Teleprompter On-device teleprompter that helps you become more charismatic by providing you with various quotes to say. - [GitHub - danielgross/teleprompter](https://github.com/danielgross/teleprompter) ![[Large Language Models-1672357227745.jpeg]]