Not looking for an easy way out, ha! More curious about ways to actually be more diligent with the practice so that I am actually learning and retaining the knowledge. Curious about what other people have done to take learning into their own hands. Making the problems into Anki cards seems like one approach. I appreciate the note in the other comment though that writing the problems out by hand is possibly more reflective of how you may encounter the problem in an interview setting. Getting too comfortable with my IDE could be a detriment....
Here is something actionable you can start doing immediately: recognize when you are jumping into habitual reactions in conversation. (e.g. problem-solving, giving advice, storytelling, devil's advocate, etc). I have a list of those here under "concepts": https://josephrmartinez.github.io/reactionsjournal/
Just chill out and try not doing that for a few seconds. Not saying this is always the wrong thing to do, but if your intention is to have your friends and colleagues appreciate your listening skills I think this is the starting point. Be more present to the conversation; responding to what they are actually saying and not just habitually reacting.
Eep! Thanks for pointing that out. I'll just put the list here:
- Giving advice
- Changing the subject
- Dismissing feelings
- Leading questions
- Devil's advocate
- Sympathy (e.g. "I know exactly what you're going through...")
- Commiserating
- Defensiveness
- Static labeling
- Sarcasm
- Shutting down
- Intellectualizing
- One-upping
- Storytelling
Not saying these are all bad all the time. Just things to maybe start noticing. See what happens when you just hold off on these types of reactions for a moment.
You don't even have to do anything. Just... chill. Be silent. Think about what the person said.
I've found this is a more actionable place to start than something like "read this book" or "take this course"
You can select between local model (ollama), claude 3.5 sonnet, or gpt-4. I've been surprised to find sonnet much better in performance and price for this task.
Quick landing page feedback:
Could you update it so that the first two screen captures are not the same content? And then pick a different second example that isn't also about leaving a note? More variation in the landing page examples would give me a better intro.
Is it possible to just call them "researchers at xyz institute" and link to their lab rather than "Scandinavian researchers"?
It might be inappropriate or disallowed to use the name of the institution in separate commercial promotion.
But mentioning Scandinavian seems savvy. If they hadn't said they were Scandinavian, I would've assumed the startup was totally full of poo, because many of the signs thus far look like stereotypical US techbro sociopathy. If I had that reaction as a US techbro myself, then presumably some Europeans they might be speaking to (as prospective investors, hires, or users) would also have that reaction.
Thank you for answering for me. And yes this is the case, the contracts with Scandinavian Universities are very slow and heavy. But the process has started.
Just a starting point. But if you are interested in this space, fork away and build it into something useful!
My personal take is that the current tools on the market are too expensive. The cost should go way, way down. This should stay open source. Patients should have easy access to full audio recordings and transcriptions of their medical appointments. One can dream!
Having worked in computing around medical facilities and providers, unfortunately, I believe this is just a pipe dream.
Companies that charge a lot for the software will buy up the smaller companies providing services until only a few are left. They'll bundle it up with an expensive price tag. Even more so, providers tend to have protection when they buy a piece of software used by a large number of other providers. They get to bandy about saying "industry standard". I promise you, you do not want to be in front of a medical malpractice jury saying "well we slapped these parts together that we downloaded online", it won't go well for you, even if that's what the big software service did just the same.
(initial introductions, "accidentally" stumbling on the owner's favorite bar/beach/...) Yeah, I'm doing my damndest to make the world a better place. There's all of these great tools for doctors, but they're struggling, getting hit by the crossfire of malpractice suits and interoperability difficulties.
(later meetings) We just landed <xyz llc>! They're world-class at helping patients with <abc>, but they didn't have the resources to survive in the current political climate. The founders were floundering pouring their own money in to try to keep their patients happy, but with our help the patients are better off and the founders can finally retire comfortably.
(much later) I see how much you care about your business and how it's really about the people you serve; that's why I got into it too. When it's finally time for you to sell, you definitely don't have to go with us, but please try to hand it off the right way to somebody who cares. Private equity vultures don't make anybody happy.
There are a lot of these youtube-openai api projects, but I really don't want to open another page to interact with youtube videos. This is a Chrome extension to perform tasks on YouTube videos without leaving your current tab.
This extension simply pulls down a transcript for the video you are currently viewing and sends this to the OpenAI API with your task request.
"Generate a vegetarian version of this recipe."
"Extract the materials list for this project."
"What was the book they recommended?"
This extension does not use a speech to text model to transcribe the video or an image recognition model to actually pass along information about what took place in the video. Automatically generated YouTube transcripts are not great quality, but they tend to be totally fine for the lightweight use cases when you just have a simple question on a video with dialogue.
gpt-3.5-turbo-1106 with the 16k token context window is used by default for videos under about ten minutes. For longer videos, the gpt-4-1106-preview model with a 128k context window is automatically selected. You should be able to use this for videos of to about four hours in length, but this is highly dependent on the volume of dialogue.
Best thought of as a very simple starting point for configuring a diy extension of your own. I actually use this on a regular basis now! Its nice for just getting around ads and not suffering through long youtube videos sent by well-meaning friends. Would love to see how other people configure this to be even better. Fork away!
This can be a very useful tool especially for cherry picking long videos for a specific recommendation or book review. A second order use case would be running it in a loop to get all book recommendations from a podcast series or say patio11 posts.
It seems hopelessly confused, because there are no open source models, only some that allow free redistribution of the model weights. Perhaps they should be called "open" and the word "source" should be dropped.
I made a simple Chrome extension that similarly pulls down the video transcript and sends this to the openai chat completions endpoint:
https://github.com/josephrmartinez/AskYouTube
This extension allows me to "ask" the model to perform a task on the video content:
- "Give me the materials list" (for a diy video)
- "What was the recommended book?" (for a 2+ hour podcast where they made a reference I can't find again easily)
- "Extract the recommended protocol" (for 3+ hour health videos)
- "Provide a counter argument" (for when I'm getting bored...)
Big plus is that you DO NOT need to wait for the ad to play through. I can just navigate to the video and send in a query without having to watch any ads.
Youtube transcripts are pretty rough. At first, I used Whisper to create a better transcript. But my primary use is to ask something of the youtube video - I found that slinging the so-so transcript along with my task was totally fine. Really simple project: Chrome extension in just html, css, and js. FastAPI server for the openai endpoint. Server function does a quick tokenization on the transcript to determine if I need to use the gpt4 model for the 128k context window or if the gptt3.5 16k context window is okay.
I've found that to be the case. I typically don't want a full transcript -- I want the materials list, or a summary, or a counterargument. I've found it is totally sufficient to just plop the transcript into an LLM and ask for my desired output. No need to clean of the transcript ahead of time.
