Saharsh Agrawal

Early-stage fundraising is getting complicated due to SAFEs. Should they be limited to early rounds? • Angel Rounds (SAFEs): $150K-$200K raised, small valuation caps, light dilution. • Small Pre-Seed (SAFEs): $680K raised, $9M cap, 6%-8% dilution. • Big Pre-Seed (SAFEs): $1.5M raised, $12M cap, 11%-13% dilution. • Seed (Priced vs SAFEs): ~$3.3M raised, $13.7M valuation (priced) vs $18M cap (SAFEs), slightly lower dilution with SAFEs. • Post-Seed Extensions: More common, but costly in equity. #Fundraising #Startups #VC #Investment

7

Keep Learning New Things... Indeed, the path to efficiency requires discipline, consistency, and sacrifice. I can't believe that it is already the 3rd week of the Headstarter Fellowship! To be honest, I felt quite lazy throughout this week, with low dopamine levels. However, I'm getting back on track and am ready for the next week of the fellowship. During the previous week, I had the opportunity to work on some exciting projects: - The Pantry App Project: I focused on the Firebase backend, learning how to integrate features such as authentication, analytics, and database CRUD operations. Live: https://lnkd.in/dCGFCiH5 Github: https://lnkd.in/dWpHgc4y Pathway to Tech Event 2: Here's My Key Takeaways: Dennis Crowley: Founder of Foursquare ($150M Revenue) When asked about the ideal tech stack, He emphasized the importance of being "willing to fill in gaps." It's not just about knowing specific technologies but being able to and ready to learn. His experience on failures "People quit on themselves or quit on their startup.", reminding me to stay resilient and never quit Tariq Afaq: First Engineer at AppLovin ($30Bn IPO) Tariq's talk was packed with valuable information. when he mentioned that this field is more than just coding, like being an engineer we can solve problems under pressure and get things back on track He also mentioned that you don't need to be super intelligent to succeed. Instead, focus on creating in-depth products, being quick and responsive, and working hard. Julian Alvarez: Ex-Meta, CEO of Wisdolia ($20K+ MRR) The most energetic speaker at this event! , His thoughts on how startup founders stick to their ideas rather than believing in their team's capabilities become the reason for their failure something exceptional and new for me to learn. His startup Wisdolia— is truly something. It generates flashcards for you from any YouTube video, blog, article, or PDF. I can already see it becoming a vital part of my learning routine! Hamza Ali - CEO of Olostep I was deeply inspired by Hamza Ali's vision for Olostep, which is enabling APIs to interact with the web as humans do. It's truly a brilliant idea that showcases his commitment to shaping the future of tech. 🔗 Here's the link to the recording: https://lnkd.in/dzX6Saaz

7

3 Comments

India’s entrepreneurial ecosystem is thriving, offering vast, untapped potential. Unlike China, where state control has tightened, India provides a more supportive environment for startups. Here are some key points: ✅ 1. Growing Startup Ecosystem India’s startup ecosystem is the third-largest globally, with over 100 unicorns valued at around $340 billion. The surge in new businesses is driven by a young population, increasing internet use, and supportive government policies. ✅ 2. Untapped Markets The “Bharat” segment, referring to the population outside major cities, presents significant opportunities. Startups are targeting this segment with affordable products and services, bridging the digital divide between urban and rural areas. ✅ 3. Deep Science and Technology India’s deep science tech ecosystem is ripe for investment. Innovations in biotechnology, renewable energy, and advanced manufacturing are set to make global impacts. 📶 Recent Examples ✅ Zepto: Offers 10-minute grocery delivery, quickly becoming a unicorn. Jio Financial Services: Partnered with BlackRock to revolutionize wealth management in India. ✅ Semiconductor Startups: AGNIT Semiconductors and Aura Semiconductor are leading the industry. ✅ Dharaksha Ecosolutions: Converts crop residue into biodegradable packaging. ✅ Digital Payments and Agri-tech: Paytm and Ninjacart are transforming digital transactions and agricultural supply chains. India’s startup ecosystem is not just about financial growth but also societal progress. With the right investments, India is poised to become a global leader in innovation and entrepreneurship. if you are looking for corporate venture strategies or you are a startup or innovation looking for corporate ventures funding in India. Feel free to connect. [email protected]

2

NVIDIA's Jensen said "ChatGPT democratized computing, Llama2 democratized generative AI" ...and now Llama3 is the next level 🔥 It's fascinating to watch Meta's strategic moves. With Llama3, they prioritized training on a massive 15T token dataset to pack all into a lean 70B param model, instead of building a massive model. This allows Llama 3 to match trillion+ parameter models like GPT-4, but at 1/10th the compute, storage and inference costs! 💰 This technique was published by Google DeepMind a few years ago https://lnkd.in/gx7VU8aA Meta is not a dark horse in AI anymore. They might be the top dog. https://lnkd.in/gRRJYuxt #llama2 #llama3 #llm #chatgpt #gpt4 #ai #ml

16

5 Comments

AI memory is fundamental for conversational AI and AI agents. It is critical for building personalized experiences and improving reasoning. Check out Letta’s course to understand how to use AI memory to enhance the customization and quality of your AI app.

15

📜 A new blog post explaining the fundamentals of RL for LLMs—the perfect place to intuitively understand PPO and RLHF from first principles. 👇👇 Pre-trained LLMs are often compared to Lovecraft's Shoggoth, a monstrous, shape-shifting beast. PPO is the algorithm that brought it under control. Take a look! 🔗 https://lnkd.in/eXqDvrUV

29

1 Comment

I get excited about the little things in the world of LLMs, because they're generally the things that will make the most impact. Anthropic's Prompt Cacheing is one of those things. Prompt Caching enables builders to cache frequently used context between API calls. Meaning, if you have an agent, that has frequent use, you can cache parts of your prompt, to save on cost and latency. According to Anthropic, cache write tokens are 25% more expensive than base input tokens, but Cache read tokens are 90% cheaper than base input tokens. Interestingly, this is going to mean that you'll need to start structuring your prompt a little differently. Generally, you would structure your prompt like this: 1. Input Data (user question, chunks, guidelines) 2. Task (implicitly stated within instructions) 3. Instructions 4. Output specification Whereas, now with prompt caching you would do something more like this: 1. Task (clearly stated at the beginning) 2. Instructions ** This part is cached** 3. Input Data (placed at the end) 4. Output instruction Super excited to have a play with this.

57

7 Comments

Meta researchers develop ‘thinking’ LLMs  Meta researchers just introduced a new method called Thought Preference Optimization (TPO) to train large language models to ‘think’ before responding to general instructions — not just reasoning tasks. The details: →TPO prompts models to generate internal thoughts before responding to user instructions, similar to how humans think before speaking. →The AI's thoughts are kept private, with only the final answer shown to users with the AI using trial-and-error without direct supervision to optimize outputs. →TPO outperforms standard models on key benchmarks for non-reasoning tasks like marketing and creative writing but declines in math-related tasks. →The approach builds on the recent OpenAI ‘Strawberry’ research and o1 model release, which takes time to reason. Why it matters: Meta’s own Yann LeCun may scoff at the notion, but this method shows the potential of AI to ‘think’ for broader tasks than just math and reasoning. By allowing models to learn useful thought processes independently, TPO could also enable more capable and flexible AI assistants across a wider range of domains.

6

1 Comment

Optimize for 🚀 Bio Cores First, Silicon 🤖 Cores Second David Heinemeier Hansson makes a compelling case: prioritize human programmers (bio cores) over computer processing power (silicon cores). Here’s why it matters: 👉 Investing in programmer efficiency is far more valuable than solely optimizing for computer speed. 👉 Human labor costs far exceed computing costs. Boosting human productivity should be the main focus. 👉 In the age of AI and high-level programming languages, the bio cores drive your real gains. Bottom line: Make your programmers more productive, and watch your team (and tech) thrive. Read the full article here: https://lnkd.in/gK8HEGtk P.S. : The image below is brilliantly summarized using Napkin.ai for a quick visual breakdown.

6

Can you really build a company the same way you build an engineering team? I recently touched upon this on the ML Street podcast with Tim Scarfe but let me break it down further. If you’re Jensen Huang running NVIDIA, the answer is yes.  With one of the flattest structures in comparison to company size in the world, Jensen manages 50+ direct reports.  Jensen’s centralized approach to management works as employees aren’t deemed more powerful because of the information they possess. This creates a culture of transparency and freedom of information that… 1. Shows people how and why their work impacts overall business objectives 2. Motivates employees when they know what kind of impact they’re making 3. Aligns everyone to company objectives, reducing friction 4. Builds trust as leadership is pressured to be candid and accountable 5. Fosters innovation through information gathering and collaboration across the company At CentML, we’re a blend of all of the above, where we believe in honesty, craftsmanship, and collaboration, not only within our engineering teams, but our company as a whole. Although, we were in fact founded by a group of engineers, so perhaps a bit of the engineering team rubs off on how we’ve grown the company after all. #ai #engineering #hiring #company #nvidia #centml #management #collaboration #honesty #craftsmanship

44

1 Comment

Join us for a conversation with Peggy Rayzis on building a "Developer Intent led" GTM motion! We explore the key drivers that led to the wild success of Apollo Graph QL... followed by an AMA Link to register in comments. Feel free to share any questions

7

2 Comments

Headed out soon! Come geek out with me and see why I ❤️ Streamlit! (And why it’s a perfect pairing with LaunchDarkly when testing out models from Hugging Face) #AI #MachineLearning #FeatureManagement

3

A $23bn offer from Google just got rejected. This would have been Google’s biggest deal in history. Wiz is a new cyber company with insane growth. - Founded in 2020 - $350m in ARR - 40% of Fortune 100 companies are its consumers. During COVID, tech giants made millions off cloud computing, but now employees are back in the office – Google was looking for places to grow aka Cybersecurity (this would also bolster Google Cloud). Google is outclassed by competitors like Microsoft in cyber and saw buying Wiz as a shortcut to catch up. Wiz was a relatively new player in the space, but that growth speaks for itself. Google was willing to pay almost double the company’s previous valuation of $12bn just three months ago. So why did Wiz reject the offer? Likely antitrust concerns (particularly coming off the CrowdStrike debacle). Lina Khan is watching the Tech Giants hawkishly. But to quote Wiz’s CEO “Our next milestones are $1 billion in ARR and an IPO” so it might just be a difference in visions. And good luck to them. #Wiz #Google #acquisitions

33

1 Comment

Elon Musk’s xAI and the Surprise Launch of Grok-2 Elon Musk's company xAI has launched Grok-2 and Grok-2 mini in beta, positioning xAI as a formidable competitor against AI giants like OpenAI and Google. Previously seen as a minor player, Grok-2 now reportedly surpasses GPT-4 and Claude 3.5 Opus in key benchmarks. Grok-2 boasts significant improvements in reasoning, reading, math, and science, particularly excelling in visual tasks. A strategic partnership with Black Forest Labs, known for their cutting-edge image generation technology, further enhances Grok-2’s capabilities. Independent validation is needed, but this move heightens the pressure on OpenAI and Google, making the industry eager to see their next steps. Link: https://x.ai/

5

1 Comment

Breaking: Utah-based Recursion acquires UK-based AI-drug discovery comapany Exscientia in arguably the largest AI in bio deal so far. Recursion Pharmaceuticals (RXRX.O) has agreed to buy smaller rival Exscientia (EXAI.O) for $688 million in an all-stock deal, according to a statement seen by Reuters and further reported by BioPharmaTrend.com (link in the comments). The deal is reflective of a larger pattern in the AI for drug discovery space, where a consolidation is ongoing with a small number of leading companies taking over smaller or financially troubled players with otherwise strong and innovative technological resources and know-how. Read further 👇 #artificialintelligence #drugdiscovery #mergerandacquisition #biotech Image credit: BiopharmaTrend

514

34 Comments

A few weeks ago Meenal Nalwaya who heads up multimodal LLMs at Meta and I were chatting about what is going on in the world of agentic infra globally. There have been a bunch of opinions on what agents are, what they can do, but there was no clear canonical architecture that was emerging. Even doing something as basic as a personal agent to book "a flight ticket to SFO tomorrow" needs the agent to know what "today" is, where "here" is to book something from here to SFO, my flight + seat + meal preferences, my FF number, my ID & credit card info, then once booked, schedule an uber ride, get my boarding passes & place it on my calendar, and ensure a pickup & a hotel once I land, and so much more. A simple task like this has so, so many complexities. We felt we needed to dig into what is the architectural framework needed to solve some of this. We also felt that this problem needs a 2-sided approach - you can build a bicycle but then you also need the roads to build bicycle lanes or it doesn't work. While we believe this utopian world where agents can book us a flight, a car, a hotel, is still not here yet, we wanted to write a series of articles to help us think about the opportunities and challenges ahead. This is the first part of a series of articles, feedback & comments welcome - link in the first comment - and hope you enjoy the Homer Simpson references :) part 1: "Endless Homers, Infinite Agents":

463

32 Comments