Javier Luraschi

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

Please media types writing about the cost of AI: recognize there is a difference between huge GPU/TPU clusters needed to train foundation models vs. inference costs to run most apps on top of foundation models. The two are not synonymous, so you can dial back some of the doom and gloom about AI startups and your commingling of the two different AI business types.

11

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

Large Language Models (LLMs) as super lossy compression. I have been thinking a lot about demystifying what LLMs do because lots of y'all's are very mystified. In the linked article, I try to reframe them for you, hoping it will help you understand what they good for and what they are not good for. Side note: I am still looking for my next adventure. If communicating complicated concepts so that regular people can grasp them is something you need, you should message me. I write code and stuff too and am usually fun to be around. #WrittenByMe

6

3 Comments

Anthropic AI Releases Claude 3.5: A New AI Model that Surpasses GPT-4o on Multiple Benchmarks While Being 2x Faster than Claude 3 Opus Anthropic AI has launched Claude 3.5 Sonnet, marking the first release in its new Claude 3.5 model family. This latest iteration of Claude brings significant advancements in AI capabilities, setting a new benchmark in the industry for intelligence and performance. Claude 3.5 Sonnet is available for free on Claude.ai and the Claude iOS app. The model is accessible via the Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI. Enhanced rate limits are provided for Claude Pro and Team plan subscribers. The pricing structure is set at $3 per million input tokens and $15 per million output tokens, with a 200K token context window, making it cost-effective and highly efficient. Quick read: https://lnkd.in/g6DZAvHD Try it: https://lnkd.in/gfVBhz8S Anthropic Blog: https://lnkd.in/gSuW2WcX Anthropic

44

🧩 One thing we've noticed with #VentureIQ technology is how well it can distill and communicate complex technical material 🧩 Whether it's a #deeptech venture, a novel #AI application, or even a patent submission, we can quickly distill the business and strategic implications for investors...and sometimes even the founders/PIs themselves! 🤯 Based on our work so far, the tech can handle a wide array of research areas and technological sectors -- anything from blockchain algorithms to vaccine development. It handles the nuance of the field, the specific terminology, and extracts the key concepts which matter for venture-building. 🤓 To demonstrate, I took a patent I helped write with Eric Laber on reinforcement learning. I ran it through #Ventureiq to build a commercial assessment (key differentiators, target markets). Then for fun I had #NotebookLM make a podcast out of it. I've uploaded the result it here (DM me if you'd like the actual report): https://lnkd.in/eiTejsny This is more than just automation or streamlining operations - it's about unlocking clarity and insight when it was previously unattainable, and the ability to do it over and over again no matter the sector or vertical. Interested in more info? DM me so we can chat.

11

2 Comments

All the hype around GenAI means that every software maker seems to be stuffing chat interfaces into their products whenever they can. For the most part, the jury is still out on whether this is a good idea or not. However, design goes deeper than just the user interface, so it’s also useful to know about how the designs interact with the rest of the software. Once you move beyond chatbots into things like agents, there are also thorny questions around which bits of your workflow should still be done by a human, and which bits can be completely automated. True insight in this context lies in a gray area, across software, UX and AI. I just came across a wonderful podcast on designing modern AI applications https://lnkd.in/d_CuHM-C Here are my pickings from this podcast 𝘊𝘰𝘨𝘯𝘪𝘵𝘪𝘷𝘦 𝘢𝘯𝘥 𝘊𝘰𝘮𝘱𝘰𝘴𝘢𝘣𝘭𝘦 𝘈𝘳𝘤𝘩𝘪𝘵𝘦𝘤𝘵𝘶𝘳𝘦: Adopting a cognitive and composable architecture allows for modular, flexible AI systems. This design ensures that different components can be updated independently, providing a seamless and adaptive user experience. 𝘛𝘩𝘪𝘯𝘬 𝘰𝘧 𝘎𝘦𝘯𝘈𝘐 𝘚𝘰𝘭𝘶𝘵𝘪𝘰𝘯𝘴 𝘢𝘴 𝘚𝘬𝘪𝘭𝘭𝘴, 𝘕𝘰𝘵 𝘑𝘶𝘴𝘵 𝘊𝘩𝘢𝘵 𝘐𝘯𝘵𝘦𝘳𝘧𝘢𝘤𝘦𝘴: Instead of viewing GenAI solutions as isolated chat interfaces, consider them as a collection of specialized skills, similar to Alexa. Each skill can perform specific tasks, contributing to a cohesive and versatile system. 𝘏𝘶𝘮𝘢𝘯 𝘪𝘯 𝘵𝘩𝘦 𝘓𝘰𝘰𝘱 𝘍𝘪𝘳𝘴𝘵, 𝘈𝘶𝘵𝘰𝘮𝘢𝘵𝘪𝘰𝘯 𝘚𝘦𝘤𝘰𝘯𝘥: Integrating human oversight initially ensures that AI systems learn accurately from feedback and reduce errors. This approach builds trust and reliability before full automation is implemented. 𝘓𝘓𝘔 𝘧𝘰𝘳 𝘕𝘦𝘹𝘵 𝘉𝘦𝘴𝘵 𝘞𝘰𝘳𝘥, 𝘈𝘐 𝘈𝘨𝘦𝘯𝘵 𝘧𝘰𝘳 𝘕𝘦𝘹𝘵 𝘉𝘦𝘴𝘵 𝘈𝘤𝘵𝘪𝘰𝘯: Use LLMs for tasks that involve predicting the next best word, such as text generation. For decision-making and process automation, employ AI agents to determine the next best action. 𝘏𝘺𝘣𝘳𝘪𝘥 𝘈𝘱𝘱𝘳𝘰𝘢𝘤𝘩: 𝘋𝘦𝘵𝘦𝘳𝘮𝘪𝘯𝘪𝘴𝘵𝘪𝘤 & 𝘗𝘳𝘰𝘣𝘢𝘣𝘪𝘭𝘪𝘴𝘵𝘪𝘤 𝘈𝘐: Combining deterministic AI (rule-based systems) with probabilistic AI (ML/LLMs) enhances effectiveness and reliability, depending on the use case. This hybrid approach leverages the strengths of both methods. 𝘗𝘳𝘰𝘮𝘱𝘵 𝘰𝘳 𝘘𝘶𝘦𝘳𝘺 𝘊𝘭𝘢𝘴𝘴𝘪𝘧𝘪𝘤𝘢𝘵𝘪𝘰𝘯: Accurately classifying prompts or queries ensures the right model or API is used to serve the request. This improves efficiency and the relevance of responses, enhancing user satisfaction.

10

We have a packed house and waiting list for this Tuesday’s in person GTM roundtable in San Francisco. We are also thinking about having a seed, series A & B startup GTM session later this month on Zoom for those who live in other areas of the country or world. Please let me know your thoughts. #sales #gtm #ai #genai #startups #founders Rey Advising

17

This is just incredible. Neal Ghosh's demonstration of how #VentureIQ takes deeply technical and complicated content across a range of industrial use cases and ages/stages through an AI-generated conversational podcast (a la #notebookLM) is really mind-blowing. Check out his post, below, and if you haven't seen VentureIQ, yet, let's get you set up to learn more. You can visit VentureIQ.ai or DM any of us over here to book a demo!

2

LLMs are not vast, infallible databases. As models grow larger so does their tendency to confidently state falsehoods (hallucinations). Matei Zaharia, co-founder & CTO of Databricks, warns against treating LLMs as internet-scale encyclopedias and proposes we focus on what they excel at:  → Reasoning → Interpreting language → Understanding context For founders building AI applications, treat LLMs as powerful reasoning engines and not fact repositories. Build your knowledge base separately, feeding the LLM with trusted up-to-date information through retrieval mechanisms and tool-calling. This approach lets you harness the LLM's strengths in language analysis and generation while sidestepping the risks of its unreliable memory. Building with AI? Share your approach to balancing LLM reasoning power with factual accuracy in the comments. Let's hear it!

137

22 Comments

Introducing Ocular AI (YC W24) 2.0 (rebranded and refreshed) and our first-ever Launch Week: https://lnkd.in/gAzPpBJa We stand at the chiasm of a technological renaissance, driven by the symbiotic partnership between humans and machines. This dynamic relationship has ushered in an era where Artificial Intelligence augments human intelligence, creating technologies that are more sophisticated, accurate, and practical—all while preserving human oversight and expertise. AI enhances human capabilities, empowering us to solve complex problems while maintaining agency and control. At the heart of this advancement is real-world data, enriched by human intelligence. Together, they form the foundation for cutting-edge AI that is both impactful and ethical. When Louis Murerwa and I started Ocular AI (YC W24) as an Open-Source Enterprise Search project during Y Combinator, our goal was to make information retrieval more intelligent. Early on, we realized that data quality was the decisive factor in achieving meaningful AI results. Over the past few months, we’ve engaged with hundreds of founders, executives, and engineers from startups, enterprises, research labs, and companies in robotics and self-driving cars. The message is clear: data is the lifeblood of AI. That’s why Ocular AI (YC W24) is building the Data Foundry for the Multi-Modal AI Era. By combining AI scalability with human insight, we’re creating golden datasets that go beyond training models, paving the way for AI breakthroughs. We’re excited to announce Ocular AI (YC W24)’s first-ever Launch Week, from December 2-6! Our team (Vivek Lahole, Louis Murerwa) will unveil two flagship products, Ocular Foundry and Ocular Bolt, which lay the foundation of our vision. Foundry and Bolt will strengthen the AI-human intelligence bond, enabling teams to create, refine, and deploy data with precision, efficiency, and speed. Companies spend huge amounts on training data, but Foundry and Bolt are AI-native tools that will lower costs, reduce manual effort, and accelerate high-quality data collection. We’re replacing outdated, clunky, and expensive data software! 💡 What to Expect: - Daily announcements and product demos - Insights into tackling AI’s biggest challenges - Opportunities to connect with our team 🎯 Be part of the journey: - Explore the Launch Week page: https://lnkd.in/gAzPpBJa If you’re working on AI for: 1. Computer Vision 2. Robotics 3. Self-Driving Cars 4. Foundational Models - Join our Slack community for live updates: https://lnkd.in/gVSeGTC2 - Sign up for our Beta Waitlist: https://lnkd.in/gsgkwsBB We’re already powering data for robotics companies and would love to hear and learn from you!

172

37 Comments

In a recent Office Hours, Baris Gultekin, Head of AI at Snowflake & a friend from Google, shared his insights into Snowflake’s approach to AI. It was a great conversation to understand how one of the leading data companies is pushing AI forward. The video for the session is here : https://lnkd.in/gfBesq35 & the audio is here : https://lnkd.in/gUK8hR66. There are four major pushes for Snowflake within AI. 1. Bringing AI to Data with Cortex. Cortex is a suite of AI building blocks that enable customers to leverage large language models (LLMs) & build applications. Both structured data & unstructured data (extracting information from presentations & pdfs) can be built using Cortex. 2. Developing open source initiatives including a data catalog, Polaris, & an open LLM Model Arctic which focuses on SQL performance. Open source is a new motion for Snowflake. 3. Securing data sharing to bring data into customers’ accounts for AI. Data sharing across customers enables a marketing organization to use data from a credit card company to build a model for customer marketing. 4. Developing a platform atop Snowflake with native apps to enable a retailer to provide a search experience to their wholesalers, for example. We also chatted about the importance of data access security for AI applications. For many data leaders, this is one of the top priorities to understand before deploying AI. Thanks, Baris, for sharing your perspectives!

31

4 Comments

Lin Guan and Yifan Zhou have led a very interesting study investigating the effectiveness of Vision Language Models in critiquing robot behaviors in terms of the desirability of their "style." If you are wondering how our work on LLM's inability to verify correctness squares with this study on using them as style critics, see my earlier post https://lnkd.in/gBYAr5qq ! This is the reason why being a style critic is one of the roles LLMs can play in LLM-Modulo framework: https://lnkd.in/gyAyKx4E Check out the details of the work @ https://lnkd.in/gfVZ4EYW ( Lin Guan also has a nice twitter thread on the work here: https://lnkd.in/gnieU-qM )

89

2 Comments

This new little product demo is awesome! Our new Rev AI Transcript Assistant will save you TONS of time. Easily pull out nuggets of info from your transcripts, making valuable content more accessible than ever. https://lnkd.in/gzzFpcDa #transcription #asr #accessibility #transcript #summarization #ai #assistant

31

2 Comments

In the next episode of The Eric Ries show, I talk with Yancey Strickler, co-founder and former CEO of Kickstarter, and founder of Metalabel, a platform for releasing collective work. He’s also the author of This Could Be Our Future: A Manifesto for a More Generous World. Yancey, who started out as a music journalist, believes that creativity and humanity are implicitly connected. He’s been a forerunner in thinking about how to build companies that bring good things into the world and are also successful without devolving into extractive behaviors. He’s also had the rare experience of, as he says, “watching an idea go from impossible to explain to ubiquitous and instantly understood.” We talked about the founding and growth of Kickstarter, which has been profitable since it’s 14th month, the power of humility, past mistakes and future hopes, why he started Metalabel, and more. You can listen to or watch the episode here: - YouTube: https://lnkd.in/g76Ez6B5 - Spotify: https://lnkd.in/gSwg9kKb - Apple: https://lnkd.in/gJk5ng6H Meanwhile, here are a few takeaways from our conversation: 1. There are downsides and upsides to originating a new category: Crowdsourcing seems totally normal to us now, but only because Kickstarter led the way. At the beginning, there were years of struggle to even get people to understand the idea of conditional purchasing. But once they got the site built and people could see what it was, the excitement and adoption of this genuinely new platform were: “instantaneous – people knew they got it, they got what it was, they got how to use it.” 2. Maximizing growth isn’t always the right tactic. “We always had this thought of ‘we're not trying to be as big as possible. We're trying to be what feels right, what feels faithful in some nebulous way’,” Yancey says of Kickstarter. Staying true to their passion and their mission at the start ultimately led to success. 3. Crisis can be good. In many cases, it leads to uncovering the truth about what’s really going on. Instead of just trying to figure out how to make whatever’s going wrong go away, disaster can bring about real change. 4. Organizations are their own entities. They exert force on the people who found them and run them within them just as much as those people guide the organization. Learning to co-exist in that relationship is an art. “[Metalabel] exists beyond us without question. It has its own physics, its own personality that you need to listen to and be humble before. You're supporting it, you're serving it. And there's just so much art and humility and really hard work that comes with seeing it through.” — Lastly, a huge thank you to our sponsors. Without them, this podcast wouldn’t be possible: Mercury DigitalOcean Neo4j

44

2 Comments

I've adapted Exa's semantic search API for Google Apps Script since I found myself frequently combining both tools. ▪️ Matches official SDK interface ▪️ Works natively in Google Workspace ▪️ Supports all API features e.g. neural search, content categorization, domain filtering, and text summarization, publish date. ▪️ Ready to use in Sheets, Docs, and Forms (Google Apps Script is not the first choice when we think about building integrations but after working with it on several projects, I found it quite powerful for specific use cases - simple, almost free, easy) Link to Github documenation and examples: https://lnkd.in/dXeAHhBy Tweet with 8 usecases e.g. Google Calendar brief generator, Google Form response enrichment (when someone submits a company name or topic in a form, automatically enriches the response with latest news, social media presence, and key information): https://lnkd.in/d_KSHgHN  

11

1 Comment

We at WhyHow.AI are big proponents of ‘small knowledge graphs’ as the atomic unit for structured knowledge representation. We explore how modular small knowledge graphs contrast with monolithic large knowledge graphs, especially in the context of the rise of multi-agent systems. We explore how we think about the relevance of smaller, scoped-in graphs that have the advantage of easier graph construction, scaling, orchestration, retrieval, maintenance and debugging. Chris Rec Thomas Smoker https://lnkd.in/eJSznveD

95

4 Comments

That's a good summary Manny Bernabe - Reflecting on an insightful article from O'Reilly about LLMs, I was struck by this point: "LLMs are also broadly accessible, allowing everyone, not just ML engineers and scientists, to build intelligence into their products. While the barrier to entry for building AI products has been lowered, creating those effective beyond a demo remains a deceptively difficult endeavor." This democratization reminds me of the printing press's impact on biblical texts, which expanded access but also introduced new interpretive challenges. As we navigate these advancements, it's crucial to balance accessibility with responsible use. At Ushur, we've been at the forefront of leveraging LLMs in our backend/knowledge work automation as well as carefully introducing them on the conversational side via our Chatbots and other modes. More will be coming out on that. Read more here: O'Reilly Article #AI #LLM #Technology #Innovation #History #Democratization #ResponsibleAI #Ushur

12

1 Comment

Found this to be an extremely interesting read from Philip Rathle, Emil Eifrem and the team at Neo4j. I have been seeing more and more companies taking customer data, enriching it, creating graphs and then using those for RAG across many domains. Grounding, accuracy and next level insights from the Q&A have been the main reasons for going this route vs Vector only - think making next level connections by turning your data into its natural graph form. From this piece I learned more about the development velocity uplift and the ability to audit. Interested in hearing from folks who have had success or failure from this approach.

16

1 Comment