Dataleap retweeted
Tim Draper said something that hit me right in the chest: "A startup is a mission. A great entrepreneur is a missionary. Not a mercenary."
I will be honest, I've lived both sides of this.
After 5 pivots over 4 years, there were moments where I tried to play 4D chess. Figure out the perfect positioning. How we win this market. How we get acquired. How we make everyone rich. I was surrounded by mercenary energy, and it seeped in.
Most VCs are mercenaries. I'll say it. Their first question is always "How does your business make money?" And when you're a missionary who needs money from mercenaries, it puts you under insane pressure to think like one.
That nearly broke me.
What brought me back was one question: ๐ช๐ต๐ ๐ฎ๐บ ๐ ๐ฑ๐ผ๐ถ๐ป๐ด ๐๐ต๐ถ๐ ๐๐ผ ๐บ๐๐๐ฒ๐น๐ณ?
Not "how do I make this work." Not "what's the TAM." Just... why do I keep getting back up after every pivot rips my heart out?
And the honest answer is: I think of myself like a creator. My medium is software. I love dreaming up the future and then building it. Dataleap is literally the product I want to use every day. My personal vision is that it's the last work software I ever need. Just me, great ideas, and agents that handle the rest. That feeling of building something beautiful is the fuel. Not the exit. Not the valuation.
๐ง๐ต๐ฎ๐'๐ ๐๐ต๐ ๐ ๐น๐ฒ๐ณ๐ ๐๐ผ๐ผ๐ด๐น๐ฒ.
I was in the APM program. The most prestigious product role you can get. Great pay, great brand. But it was a big company with small scope, super bureaucratic, and after four months I knew: this is not what I was born to do.
Here's the thing founders don't talk about enough: ๐ถ๐ป ๐๐ต๐ฒ ๐ฎ๐ด๐ฒ ๐ผ๐ณ ๐๐, ๐๐ผ๐๐ฟ ๐ก๐ผ๐ฟ๐๐ต ๐ฆ๐๐ฎ๐ฟ ๐ถ๐ ๐๐ต๐ฒ ๐ผ๐ป๐น๐ ๐๐ต๐ถ๐ป๐ด ๐๐ต๐ฎ๐ ๐ธ๐ฒ๐ฒ๐ฝ๐ ๐๐ผ๐ ๐๐ฎ๐ป๐ฒ.
Everything changes every single day. I joked to a friend recently that PMF doesn't really exist anymore. We have product-market fit this month. Product Month Fit. Next month? Who knows. The world moves too fast. Companies had total market domination in the middle of last year and are outdated today. New paradigms emerge weekly. The speed is completely insane.
The only thing that keeps you grounded through all of that chaos is knowing why you're doing this. Not the market opportunity. Not the competitor landscape. Your actual, personal why.
Ours is simple: follow our curiosity. Build what we want to exist in the world. Design the future. Drown out the noise of "get rich quick."
Because we're not in this for a quick flip. We're in this for the long run. To invent the future of work for ourselves and for the companies that want to live at the forefront.
Tim Draper's test for founders: Would you still work on this if capital markets closed tomorrow?
My answer after 5 pivots, near-burnout, and walking away from Google: ๐ฎ๐ฏ๐๐ผ๐น๐๐๐ฒ๐น๐. Because I'm not building for the exit. I'm building because there's something inside me that wants to be expressed.
That's a very different kind of fuel.
Finding the right information from your long-term memory vector store is key for managing the context of an agent well. Trying different algorithms here can lead to significantly better results! Try it!
Random note on k-Nearest Neighbor lookups on embeddings: in my experience much better results can be obtained by training SVMs instead. Not too widely known.
Short example:
github.com/karpathy/randomfuโฆ
Works because SVM ranking considers the unique aspects of your query w.r.t. data.
Dataleap retweeted
๐งต1/ First things first: What's vector search? It's a technique to find similar data points within a data set. You might be familiar with LIKE queries in SQL - this is a similar concept, but with more complexity and possibilities. ๐ต๏ธโโ๏ธ
๐งต1/ First things first: What's vector search? It's a technique to find similar data points within a data set. You might be familiar with LIKE queries in SQL - this is a similar concept, but with more complexity and possibilities. ๐ต๏ธโโ๏ธ
7/ Creating a vector index is simple with the right tools. Just import the necessary libraries, initialize your API key, and create an index. You can then insert your data as tuples containing the ID and vector representation of each object. ๐
Dataleap retweeted
1/ ๐จ Let's talk about AI Chatbot Security Risks: As companies race to deploy chatbots powered by large language models (LLMs), new security risks emerge that we need to address. Let's dive into these risks & discuss how we can approach them. Thread๐
Dataleap retweeted
๐งต 1/ Excited to share some tips on how to work with long documents leveraging @langchain chains! Let's dive into the four common methods and their pros and cons. Remember, there's no one-size-fits-all solution โ context is key! ๐
1/ ๐จ Let's talk about AI Chatbot Security Risks: As companies race to deploy chatbots powered by large language models (LLMs), new security risks emerge that we need to address. Let's dive into these risks & discuss how we can approach them. Thread๐
5/ ๐ It may take years before we establish best security practices for AI chatbots, and in the meantime, expect a surge of chatbot exploits & companies scrambling to fix them.



