One of the most insightful posts I’ve read recently. I especially enjoy the rationale behind the options you chose to reduce costs and going into detail on where you find the most savings.
I know the post primarily focuses on neural search, but I’m wondering you tried integrating hybrid BM-25 + embeddings search and if this led to any improvements. Also, what reranking models did you find most useful and cost efficient?
Recently I’ve pretty much replaced searching Yelp or TripAdvisor with a combination of Instagram Reels/TikToks and Google maps for travel and food recommendations. I usually just save a lot of places that I have to dig through and watch again to dig through and add to my saved Google maps lists. Because this takes me a fair bit of time when preparing for a trip, I decided to automate part of it by building a pipeline to “watch” the video for me by getting the audio and video frame data and taking into account key POIs that I then export out to a note. Later I plan to also integrate this with other travel or food tracking software like Wanderlog, Google maps list, or Beli eats. So far it’s been working well, but def expensive (so thank god for free GCP credits haha)
I’m currently building a service that allows someone to text a link or content they want summarized either via SMS or Discord and the service will return a summary based on certain parameters specified. I realized that there are numerous summarization tools based on Chrome extensions, but I do most my reading from my phone and thought that linking a URL and having a service figure out how to parse and summarize (using GPT) would be a lot easier flow. Lmk if anyone wants to try it
Definitely understand the pre-job jitters and anxiety. I felt that a lot prior to my previous job and the impostor syndrome was heavy. Here are a couple of things that helped me:
1. During onboarding, keep a log of important things you learn and people you can contact about certain topics. Be sure to understand how you can seek help, where you can seek help, and who you can talk to.
2. Understand that they picked you after the interview process which means they see the potential in you. Instead of thinking: “what if I don’t live up” why not reframe it as “what can I do to live up to what they want and what I want for myself”
3. Set boundaries clearly and utilize tools to help you do this: for example, schedule off slack notifications after 5 pm, let your coworkers know hours you’ll be online and how quickly they can expect you to respond during work hours and that when off work you won’t respond, plan stuff deliberately after work so you’re busy and can’t answer anyways
You got this! Feel free to message if you have any other questions
What are Pinecone’s advantages compared to other vector databases? From what I understand one of the founders has experience building similar features at AWS previously.
As someone who has tried a variety of productivity tools: Trello, Jira, Notepad, religious google calendar-ing, Notion, I've only really stuck with two things: notepad and Notion.
Notepad just allows me to get simple thoughts down and easily reorganize information. I then adapted that workflow with Notion and have been pretty happy with it so far: tracking daily todos, trip plans, workout routines.
What I've enjoyed most about notion is that it lets me do something as simple as my daily routines but also helps me plan/track things for technical side projects as well. I really think the flexibility, which many other tools don't have, is what I'm drawn to here
Recently a friend and I were discussing our pain points around the flood of information surrounding AI. Every minute a new article comes out touting the latest and greatest. This re-introduces the age old problem of too many articles, too little time.
Then, we got to thinking: what if we could use AI to help us summarize this? Would this just introduce noise, or would it be helpful? Would people be entertained? Would we like it at least?....and a million other questions we're still pondering.
This manifested into an experiment, Brief Bear's Week in AI Newsletter, aimed at answering the question: could an AI generated newsletter become the most interesting 5-minute read in AI?
We'd love to get feedback on the content and also hear about your gripes with and delight for AI generated content. We intend to start a discussion on these topics.
I know the post primarily focuses on neural search, but I’m wondering you tried integrating hybrid BM-25 + embeddings search and if this led to any improvements. Also, what reranking models did you find most useful and cost efficient?