Headstarter

Stay Connected , Share as much you can PS: Headstarter | Yasin Ehsan 🚀 #QuickStartAI #AIInnovation #ChatSupport #ComingSoon #NextLevelAI

Skeleton of Thought RAG by hand ✍️ + Langflow. I am designing a series of exercises to teach advanced RAG techniques. This is No. 4, just in time for the Halloween! 🎃 = Series  = Advanced RAG by hand  ✍️ + Langflow 1. Query Rewrite RAG 2. Multi-Query RAG 3. Hyde RAG 4. Skeleton of Thought RAG more to come ... = Background = Over the past few months, I've done quite a few sketches to teach advanced RAG techniques in various webinars. I was often asked: "Your drawings help me understand the idea. But how can I implement it in practice?" I am experimenting with Langflow to see if it is the right tool to bridge the gap between my high-level drawings and practical implementations. If successful, I plan to use it in my course next semester. = Design = 1. Match my Langflow exercises to my high-level hand sketches as closely as possible 2. Keep the layout as similar across the exercises as possible, making it easy to compare them and understand their differences and similarities. 3. Make data visible in their own separate blocks, rather than embedded into other blocks as inputs and outputs. For example, query, answer, retrieved chunks, prompt template, all have their own blocks. 4. Use distinct colors to denote Retrieval 🟧, Augmented ⬛️, and Generation 🟩. #halloween  #RAG #aibyhand [REPOST♻️] to share this exercise with your network!

Booking The Win Early vs. Under Promising and Over Deliver Every now and then someone will say to me, with pride, that they're under promising and intend to over deliver. When we interview people at Groq we look for a trait, we call it booking the win early. Suppose I tell you that, "if we do X the chip can be twice as fast." If you're someone who books the win early you don't hear that, what you hear is, "If we don't do X, the chip will lose half of its potential speed!" You book the possibility the moment you hear it, and failure to recognize the win is a loss. The vast majority of people are afraid to get their hopes up, it hurts when your expectations aren't met. That's the point. Loss hurts more than success feels good. That's called loss bias. People who book the win early do it because loss hurts more than winning, they're psychologically using that bias as motivation. The pain is the point. Do you know anyone who under promises and over delivers? I don't. I've met people who under promise and under deliver, people who over promise and over deliver, and people who over promise and under deliver. I have never met anyone who under promises and over delivers. Overcommitment is a necessary condition of over achievement. This isn't just about individuals, it's about teams. The best people want to work with other people who are ambitious - people who are irrationally compelled. In engineering in particular where it's about finding the 10,000 ways that don't work, and then once you find something that sticks you keep doing it, it's especially hard. The great teams over promise, and then use their collective aversion to failure to over deliver. So here's my advice. Over promise and over deliver. Book the win early. Sign your ego up to the possibility of failure and commit to the possibility of greatness. Surround yourself with others willing to commit to the possibility of greatness. Stop being a coward, and stay away from cowards because they'll turn you into a coward. And if you and I ever meet, don't even think about under promising.

Wow! Meta dropped an open NotebookLM recipe: NotebookLlama 🔥 It uses L3.2 1B/ 3B for pre-processing the PDF, L3.1 70B for Transcript creation, L3.1 8B for re-writes and Parler TTS for Text to Speech ⚡ Step 1: Pre-process PDF: Use Llama-3.2-1B-Instruct to pre-process the PDF and save it in a .txt file. Step 2: Transcript Writer: Use Llama-3.1-70B-Instruct to write a podcast transcript from the text Step 3: Dramatic Re-Writer: Use Llama-3.1-8B-Instruct model to make the transcript more dramatic Step 4: Text-To-Speech Workflow: Use parler-tts/parler-tts-mini-v1 and bark/suno to generate a conversational podcast There's still some rough edges, but it already sounds pretty fire - Link to notebook in the comments ✨ Wow! Meta dropped an open NotebookLM recipe: NotebookLlama 🔥 It uses L3.2 1B/ 3B for pre-processing the PDF, L3.1 70B for Transcript creation, L3.1 8B for re-writes and Parler TTS for Text to Speech ⚡ Step 1: Pre-process PDF: Use Llama-3.2-1B-Instruct to pre-process the PDF and save it in a .txt file. Step 2: Transcript Writer: Use Llama-3.1-70B-Instruct to write a podcast transcript from the text Step 3: Dramatic Re-Writer: Use Llama-3.1-8B-Instruct model to make the transcript more dramatic Step 4: Text-To-Speech Workflow: Use parler-tts/parler-tts-mini-v1 and bark/suno to generate a conversational podcast There's still some rough edges, but it already sounds pretty fire - Link to notebook in the comments ✨

🔥Communication is all you need Wrote a blog on how communication is the hardest part of building agents. From this axiom, a bunch of fun hot takes can be derived: ✒️Why prompt engineering isn’t going away 🤖Why you need an agent framework 🪶Why we made LangSmith the most user friendly “LLM Ops” tool 🛂Why it matters that LangGraph is the most controllable agent framework out there 🧱Why agent frameworks like LangGraph are here to stay 💻Why code will make up a large part of the "cognitive architecture" of an agent 🖼️Why UI/UX is the most important place to be innovating in AI 👪Why building agents is a multidisciplinary endeavor 😱Why people have asked us to expose LangSmith traces to their end users https://lnkd.in/g8KgFKZi

Steve Jobs on the limits we impose on ourselves: "Once you discover one simple fact" "Everything around you was made up by people who were no smarter than you" "You can change it" "You can influence it" "You can build your own things" -- Found this valuable? Follow Intro to learn how the world's most successful entrepreneurs built $100M+ businesses.

2 Ls make a W. If you feeling low bc of no offer or intv Keep it pushing. My first hackathon was photoshop. My second hackathon had empty ReadME. My third was win (bc I was only one in that category). Keep it pushing. 2 Ls make a dub. Wyt?

Copilot sings! Pretty cool, eh?

Meet Christen Xie, a Software Engineer Intern at Apple and Tesla Christen just joined the Headstarter Accelerator program to level up his skills in AI and we are super excited to see what he builds over the next 4 months! With all the new AI tools coming out, the best time to learn is now. We've devised the Headstarter Accelerator curriculum to combine classical / traditional ML with the latest in Generative AI, along with full stack Software Engineering to get you ready for AI Engineering roles at startups or big tech companies If you would like to join the Headstarter Accelerator, please fill out this application here: https://lnkd.in/eVzHdgRt

It’s no longer news to anyone that Meta’s stock has been on fire, but how much would your TC have changed if you happened to join Meta at a 5 year low back in early 2023? Had a few folks reach out regarding my previous Uber stock growth post asking to cover Meta's stock growth so let’s take a look at a Meta E5 research scientist offer (senior level) submitted in January 2023 and see how it’s grown since then. Meta still utilizes a standard 4-year vesting schedule evenly vesting 25% each year. The total stock grant issued at time of hire was $700k over 4 years. With $META trading at $130 back in January ‘23, that comes to about 1,346 shares vested per year, or 5,384 total shares granted at time-of-hire. Using this month's lowest stock price to approximate the stock price for the 3rd year, this is what the equity growth has looked like: Year 1 (Jan. 2023): 1346 shares * $130 per share = $175k Year 2 (Jan. 2024): 1346 shares * $351 per share = $472k Year 3 (Jan. 2025): 1346 shares * $563 per share (using estimate) = $757k Year 4: (Jan. 2026): 1346 shares * $??? = $??? That's a ~4x increase in stock value at a relative mature company within a pretty short duration. This *is* a unique case in that the new hire happened to catch Meta stock at a 5-year low. If they didn’t sell any stock, their currently vested stock would now be worth ~$2.15M (representing 3,814 shares: 1,346 Y1 + 1,346 Y2 + 1122 vested so far in Y3). This year this individual is set to make just shy of $1M in total compensation ($215k base + $757k stock this year). Since there was some confusion around gross vs net on my previous post, this grant value is the gross value of your stock, before any taxes or withholding as with base salary. View this specific offer: https://lnkd.in/gJT44pZd View Meta research scientist salaries here: https://lnkd.in/gPsxk_jp #salarytransparency #stockgrowth #rsus