R to @shaunmmaguire: There is a website for this

R to @cocolitron: I stole that take from @floydophone , btw. That really resonated hard.

RT by @JoberthFG_: Decided to learn @dagster this evening and my conclusion is that @schrockn & @floydophone are absolute geniuses. Never thought building data pipelines could ever be fun, but somehow they've made that happen🤯.

R to @memdotai: ...Llama2, Mistral-7b-instruct was found to be just as fast and high quality as GPT-3.5, and was hosted by Perplexity AI. This tweetstorm is to show that open LLMs are now usable for production apps, and to thank the teams involved."

R to @HunterAmacker: Saved! Here's the compiled thread: https://mem.ai/p/urf3QklfD5NPivF9D2QJ 🪄 AI-generated summary: "A side project that uses GPT-3.5 to summarize YouTube videos had grown in popularity, but the creator wanted to switch to an open-source LLM. After trying Vicuna-13b and...

R to @floydophone: @memdotai mem it

R to @floydophone: Thanks for sharing your experience, very insightful

RT by @charlyouki: Mistral on perplexity is a bit faster

RT by @VijayBombl7720: first, some context. summarize dot tech is a side project i hacked in a few weekends last year to summarize youtube videos w/ gpt-3 (now gpt-3.5). it ended up growing in popularity and now has over 200,000 monthly active users. here it is in action: https://www.summarize.tech/www.youtube.com/watch?v=vznSu-BHyVA

RT by @alahmedse: a few weeks ago, i moved summarize dot tech off of gpt-3.5 and onto @MistralAI's mistral-7b-instruct, an open-source LLM that runs on normie hardware. the model is hosted by @perplexity_ai. it's just as fast, just as high quality, and much cheaper to run. here are the details: