Exploring AI and GPT-3 Applications
ChatGPT Chrome Extensions
- Discussion on cool ChatGPT related Chrome extensions
RL without Human Feedback
- Shared interesting read on potential to use RL without human feedback
- Discussion on interesting OSS projects in the space
- Mention of ShareGPT
AI Dating Expert
- Discussion on a bot trained on Bumble chats to reply like the user
Prompt Injection Attacks
- Warning about prompt injection attacks
- Discussion on using vector search on past swipes and bios
- Discussion on OpenAI branding and trademark on GPT
- Discussion on using Qdrant for vector search
Fine Tuning OpenAI Models
- Discussion on fine tuning OpenAI models on code
- Tips and do’s/don’ts for fine tuning
- Use of embeddings and GPT-3.5 for creating diagrams
GPU Hosting Providers
- Discussion on affordable spot GPU/cloud providers
- Discussion on cloning voices using data collection and cleaning
- Use of open source models and hiring freelancers
AutoGPT and Fixie
- Discussion on using AutoGPT and Fixie for useful projects
- Use of Fixie’s SDK
Context Window and Fine Tuning
- Discussion on using context window and fine tuning for product design
- Use of multi-tenant setup and few-shot prompting
- Discussion on Indian clients’ concerns about data usage for model training
- Fine print of OpenAI’s data usage policy
- Boris Power’s advice on using prompting and fine tuning
Other Tools and Resources
- Use of Weaviate for vector search
- Use of Modal Labs for Python-friendly web endpoints
- Discussion on using serverless and stateful use cases
- Discussion on Anthropic as a competitor to OpenAI
The description and link can be mismatched because of extraction errors.
- https://gist.github.com/yoavg/6bff0fecd65950898eba1bb321cfbd81 - V. interesting read on potential to use RL without human feedback.
- Feels like a dialogue straight outta ‘Indian Matchmaking’: https://openai.com/brand +1 on this
- https://techcrunch.com/2023/04/24/gpt-may-be-trademarked-soon-if-openai-has-its-way/ - OpenAI is trying to trademark GPT and prevent any use of [insertword]GPT that may suggest an association/affiliation/partnership with OpenAI when there may be none.
- https://replit.com/bounties/@YoheiNakajima/scrape-an-api-and-se - off-topic - anyone wants to complete a Bounty for Yohei Nakajima? Too little money for trivial work?
- https://github.com/openai/chatgpt-retrieval-plugin/pull/59#issuecomment-1514694410: The message discusses difficulty in setting up Qdrant and Chroma with the chatgpt-retrieval plugin and also asks for recommendations on affordable spot GPU/cloud providers.
- https://arize.com/observe-2023/ - The URL is provided as context and does not require training or inference. The message in the same link as the URL is related to the link and may require zero shot or one shot inference.
- https://github.com/xtekky/gpt4free: A possible solution for accessing GPT4 without a paid subscription is shared in response to a request for someone to share their GPT4 access for a day and a half.
- https://github.com/voicepaw/so-vits-svc-fork (Nope. mentioned in response to a question about using https://beta.elevenlabs.io/)
- https://github.com/voicepaw/so-vits-svc-fork - Someone leaked pretrained models of many popular singers a week back. Was that for this project?
- https://docs.google.com/spreadsheets/d/1qzeFdpUPr7E0jOFwWSXd8LF30ZLjz1CSVEBiG8gPHTU/edit#gid=1792554832 - Probably. Wait I’ll share. They even got Freddy. (Context: someone is sharing a Google Sheets link and mentioning that Freddy is included in it)
- The message “They even got Freddy” is related to the URL https://twitter.com/tivadardanka/status/1649721970886594561?s=21&t=r3oag1xERfq9yMvHrl3kqA, which is followed by the URL https://aiagent.app/.
- https://twitter.com/tivadardanka/status/1649721970886594561?s=21&t=r3oag1xERfq9yMvHrl3kqA - A tweet mentioning a cool website and asking about the creator. The website mentioned is https://aiagent.app/.
- https://weaviate.io/developers/weaviate/installation/weaviate-cloud-services - The message mentions using Weaviate’s managed SaaS service WCS and expresses curiosity about why it hasn’t been found useful.
- The website https://whiterabbit.ai/ is given as an example of AI-assisted healthcare, which can minimize effort and improve patient outcomes. The message also mentions the difficulty of finding a good and affordable approach for hosting a stable diffusion-based web app.
- https://gooey.ai/compare-large-language-models/?example_id=7ihhyv3l - A link related to serverless and modal and banana type services. The message in the same link as the URL is related to the link.
- https://arize.com/observe-2023/ - The message mentions that the author owns the website and invites users to DM for bugs and feedback. The message also states that the author has spent a lot of time trying different GPU hosting platforms.
- https://modal.com/docs/guide/webhooks: A recommendation for Modal Labs as a beginner-friendly GPU hosting platform with Python-friendly web endpoints.
- https://manifold.markets/IsaacKing/will-any-llm-have-a-context-window - A discussion about whether any language model will have a context window, with a related discussion on Hacker News.
- https://manifold.markets/IsaacKing/will-any-llm-have-a-context-window - Discussion about the deployment of FMs in production using few-shot in-context learning. Related discussion can be found at https://news.ycombinator.com/item?id=35682424.
- https://www.researchgate.net/publication/370213628_Scaling_Transformer_to_1M_tokens_and_beyond_with_RMT: Personal experience from a limited sample set - more about the use-case than regulations. The message in the same link as the URL is about Indian clients’ concern regarding their data being used for model training in different industries.
- https://learn.microsoft.com/en-us/legal/cognitive-services/openai/data-privacy: OpenAI engineer Boris Power’s insights on the use of prompting and fine tuning for AI models.
- https://openai.com/blog/new-ways-to-manage-your-data-in-chatgpt - Instructions to turn off chat history on ChatGPT and assurance that all data is deleted within 30 days and not used for training.