Daniel Gallego Vico

Today’s release of Llama 3.1 marks an extremely important point in foundation model development. For the first time, at least in the academic benchmarks, we see an open source model (with a very permissive license - even allowing training data generation) performing on par with the models by OpenAI and Anthropic. Can’t wait to try the models.

38

FOSSA is Buying StackShare, a Site Used By 1.5 Million Developers: Open-source compliance and security platform FOSSA has acquired developer community platform StackShare, the company confirmed to TechCrunch. From a report: StackShare is one of the more popular platforms for developers to discuss, track, and share the tools they use to build applications. This encompasses everything from which front-end JavaScript framework to use to which cloud provider to use for specific tasks. Read more of this story at Slashdot.

1

Day 0 support for finetuned Llama 3 models in Llamas Exchange Added support for Llama 3 8B base model in Llama Exchange. Llama 3 is a very good base model for finetuning use cases considering its performance. Finetuning runs with Llama 3 are already showing impressive results. Finetuning LLMs is already the best way to achieve better performance than generic LLMs like GPT-4 in specific tasks. With a strong 8B base model, it will be a more cost-effective way to utilize LLMs. Calling out to domain experts/developers to create exceptional Llama 3 finetunes and monetize them with Llama Exchange. #Llama3

5

🚀 We are excited about FOSSA's acquisition of StackShare to revolutionize how we approach supporting the developer ecosystem & software supply chain security. 🛡️ 0️⃣ Ground-Zero focus: Reinvesting in our developer community and open source. StackShare is the best place to research developer tools within the community and discuss trade-offs and technical impact. 📢  We're excited to continue to double down on this for developers. 1️⃣ Enterprise security teams, get ready for a game-changer! 🎯 We're supercharging your ability to manage and track developer tools, tackling those pesky supply chain risks head-on. (more to come ⏩ ) 2️⃣ Data transparency is about to reach new heights! 📊 Our collaboration will create the most comprehensive metadata and knowledge base for developer tools & OSS you've ever seen. We've been thankful for the opportunity to work with Yonas Beshawred, the StackShare team and the StackShare community. They've accomplished a lot over their journey and we're excited to join in for the fun: 📈 40M Lifetime Unique Visitors 💻 1.5M Registered Developers 🥞 50K Company Stack Profiles ✍️ 30K Developer Content Contributors 🔒 3K Enterprise Accounts ⚙️ 400M Webhooks Processed 🤖 300K Git Repos Analyzed #TechAcquisition #DeveloperTools #Cybersecurity #SoftwareSupplyChain #FOSSA #StackShare https://lnkd.in/gc8_hACX

16

1 comentario

Llama 3.1 8B is an incredible model for finetuning – it surpassed GPT-4o & almost doubled the accuracy of the base Llama 3 8B model (37% -> 61%) in my tests! I finetuned both Llama 3.1 8B and Llama 3 8B on 261k multi-step math problems. I then tested those 2 finetuned models (along with Llama 3 8B & GPT-4o) on 1,000 new math problems and Llama 3.1 8B vastly outperformed all other models. Love that we now have great small open source AI models that are beating state of the art proprietary models with some finetuning & are much faster + cheaper! I'm writing a blog post (and doing a video) on exactly how I finetuned Llama 3.1 8B to beat GPT-4o and will include the full dataset, evals, and code. Coming soon!

113

9 comentarios

In case you feel the need, the need for speed

20

Here’s the biggest mistake startups make: they build products, then look for people to sell them to. I remember when I met a group of ambitious entrepreneurs in Barcelona. They built an amazing web application that allowed customers to do everything at lightning speed. The problem was that they didn't know if their target audience wanted it. Here's what they could have done: 1. Identify the problem they were solving 2. Survey their target market 3. Create a prototype 4. Iterate based on feedback 5. Then, build the actual product This approach would have saved them thousands of dollars, countless hours, and a lot of stress. When you build a product without knowing what your customers want, you're taking a huge risk. So, if you're building a product, start by understanding your customers. What are their pain points? What are their needs? What are their hidden behaviors? Once you know this, building a successful product is much easier. PS: Go, Go.

16

19 comentarios

🚀 New #Cardano dev video series 🚀 One of the first obstacles faced by developers new to Cardano is learning the Extended UTXO model and understanding how it plays into building transactions. I'm happy to announce the launch of a hands-on video series that addresses this challenge! In this 4-part series, I will teach you how to use the cardano-cli and the EUTXO model to build, sign, and submit transactions. As a developer and Stake Pool Operator, I can assure you that the cardano-cli is the most useful terminal-based tool one can learn. To give you a glimpse into the flexibility and power of the EUTXO model, I will show you how you can transfer ADA across multiple addresses in a SINGLE transaction. And here’s the best part: in order to follow along with the videos there’s absolutely NO LOCAL SETUP required. This is thanks to the amazing people at TxPipe and their fantastic platform, Demeter.run. Using Demeter, I will show you how to set up the cardano-cli and connect to a managed instance of a Cardano Node. Whether you're expanding your blockchain skills or diving into Cardano for the first time, this series is for you! https://lnkd.in/eehXzmet

18

2 comentarios

Top 16 leading TECH & BUSINESS voices in AI you should follow today! These individuals consistently provide valuable technical insights. I've gained a lot of knowledge from them, and I recommend following them: Aurimas Griciūnas T. Scott Clendaniel Armand Ruiz Demetrios Brinkmann Andreas Schwarzkopf Gina Acosta Gutiérrez Philipp Schmid Pavan Belagatti Brij kishore Pandey Aishwarya Naresh Reganti Tobias Zwingmann Pascal Biese Peter Gostev Eduardo Ordax Ashish Patel 🇮🇳 Arpit Singh I've undoubtedly missed other tech thought leaders. Could you name a few? ~~~ If you found this interesting, like and repost to support me. Follow me, Evgeny Krapivin, for more on AI tech, strategy and my startup journey.

54

20 comentarios

From a hackathon project to a full-fledged product! 🚀 Thrilled to unveil the 💡 Smart AI Search 💡 — my passion project over the last few quarters. Explore it here (http://ai.trivago.com) and share your thoughts in the comments! ✨ #AI #Search #trivago #google

32

1 comentario

Curious about the 100x+ economic gain behind AI vs humans? Read on. Vela AI agents are at least 100x cheaper, 5x faster and same quality as a venture capitalist (VC). Let's walk through the VC math. 1) Highly paid knowledge workers such as VCs in the Bay Area costs at least $120 an hour. 2) High-quality startup analysis typically takes 5-20 minutes per lead. This analysis is required to decide if a VC should take the first meeting. So, the cost per analyzing qualified lead ranges between $10 and $40. Let's skew towards cheaper and pick $20. Now, let's look into the AI agent math. 1) It takes 0.5-3 minutes to analyze a company, depending on the model's capabilities. 2) It costs $5-$20 an hour to run a LLM, depending on the infrastructure's sophistication and length of text. It costs $0.04 to $1 for AI agents to analyze a single startup. Let's skew towards more expensive and pick 20 cents. This gets interesting at scale. Humans get bored if they analyze more than 30 companies a week in-depth (at least I do!). These AI agents never get bored, works non-stop and even work on Saturday nights. When you run these models on 1M startups a year continuously, you end up paying $200K a year. I've never met an expert human or army of expert humans who can analyze 1M companies in a year, in which their quality is continuously increasing and the cost is continuously going down. This is a super power. Now imagine what other industries and businesses can use these 100x. If you have not deployed AI agents in your line of knowledge work, please drop whatever you're doing right now. Whatever that is, I can assure you that it won't matter in a few years. If you're a student or someone thinking about next steps, please stop what you're doing. This subject is the only thing matters for your career. Prioritize building, deploying and managing AI agents.

86

28 comentarios

I'm excited to share that I've completed another Module at MIT on "Designing Two-sided Markets, from APIs to Roadmaps." Only one more module to go until graduation day! 🎓 #MIT #ModuleComplete #GraduationCountdown

32

2 comentarios

Don’t Build for Everyone. Build for Someone. Last week at Google For Starups (GFS) Madrid, I saw a sign in the office that read, “Build for Everyone.” It’s a noble idea—probably more about inclusivity than actual startup advice—but here’s the thing: for startups, this approach is a trap🚨 If you build for everyone, you end up building for no one. The path to product-market fit isn’t about being everything to everyone—it’s about finding the smallest, most focused niche that lets you achieve your goals. At my last startup, we built a product to help sports teams organise themselves. In the U.S. alone, our target market could have been over 100 million people. Sounds exciting, right? 👉🏻But here’s what we learned: the broader the audience, the harder it is to solve real problems and build momentum. Instead, we hyper-focused on one segment: adult co-ed, post-collegiate social sports leagues. A small subset, sure, but this focus unlocked incredible traction. Why? 1) Simpler Product Development Fewer edge cases. More clarity on user needs. We could prioritize features that mattered most to our niche. 2) Faster Learning Insights correlated quickly within a focused audience, helping us iterate faster and build the right things. 3) Killer Adoption By understanding this niche deeply, we built a feature that became a game-changer for them—taking us from 250k to 2.5M users in just nine months. Startups often fall into the trap of thinking bigger means better. But scaling comes after nailing your initial market. Starting with a hyper-focused niche doesn’t limit your ambitions—it sets the foundation for broader success. Patagonia didn’t start by selling all outdoor gear; they started with climbing tools. Facebook didn’t launch for everyone; they launched for Harvard students. So, while I love a good motivational sign, here’s my advice: Don’t build for everyone. Build for the smallest “someone” you can serve better than anyone else. 👉🏻🗝🗝 Key Nuance: We must also have clarity on how this niche grows into a scalable market. What’s your take? Do you agree that startups should narrow their focus early on? Let me know in the comments!

26

20 comentarios

This is the feeling I get when my Content Assistant, Beatriz, nails our content plan. We are workings with a lot of founders AND creators right now. Here is how we are doing it.... I am a busy founder … and now a creator also. And I know not all tasks are created equal. My first system, back in my podcasting days was as follows. I record. I send the file. I move on. What would happen after I sent the file is my team (shout out Matías still with us today) would turn that file into. 1 x full YouTube clip with subs. 3 x micro-clips at ~5 mins for main channel. 5-7 x micro-clips for clips channel. 2 x posts for Insta and other socials in portrait. 2 x 60 second clips for stories. 1 x page with show notes. And way, way more. On top of that them did all talent prospecting, communications and bookings. And also did the same to get me on other shows. As I said, I would walk in and hit record. And that’s it. Now, today, when it comes to my social presence and my newsletter, I have Bia working as my right-hand wing-woman. The work Bia helps me with is the following. Deep-dive research including 10-15 articles, quotes, images and assets. Multiple ChatGPT and Bard prepared outline suggestions. Proof-reading and editing of each edition. Cover art and inside art for each piece (I still do some covers). Transfer of draft to beeghiiv editor. Post-release publication plan of social posts, Slack groups, other networks. And much, much more. Add to that the work Bia does helping me with my personal brand in general - I am not about to do 150-175% of the work with around 60% of the personal time investment. Meaning I can do what I should be doing, which is being a CEO. If you are a busy CEO who needs to up their comms game, or a creator looking to do more of the creative and less of the grunt work send me a DM and I will tell you how we can help find your own Content Assistant. That way you can spend your energy on things that really move the needle. --- 📌 If you want to hire your own high performing global team check us out: https://lnkd.in/dW7ggbSb

64

46 comentarios

This is why we need to continue to support open source AI projects... 😳 We have been backing open source projects since the early days of GitLab, and we don't have any plans to stop. 💪🏾 Projects you should consider supporting. Comment with additional ones I am missing. 🙏🏾 Ollama LlamaIndex Predibase Mistral AI OpenAssistantGPT H2O.ai Apache MXNet PyTorch Stable Diffusion AI Art Generator Keras TensorFlow scikit-learn OpenCV FASTAI #Opensource #AI

51

8 comentarios

Extensive #EarlyStage Venture Capital Contacts. Discover a collection featuring 1,250+ VC firms, 7,400+ professionals (Partners and Principals), and 6,400 email contacts. https://payhip.com/b/O49Bv #startup #startups #vc #venturecapital #investmentbanking #privateequity #managementconsulting #b2b #founders #entrepreneurs

1

In this episode, Daniel Palma goes into what makes Estuary Flow unique in the world of generic streaming platforms. Check it out! and subscribe to the channel for more byte sized sessions about streaming data processing.

13

1 comentario

Starting September with 6 consecutive months of GMV and Revenue growth 🎉 And for the first time we know WHY. And it's sustainable When we graduated YC we had similar growth - but it wasn't replicable and it felt like we were constantly scrambling. We compensated poor retention with massive top-of-funnel. But when you do that, you reach the natural limit on top line - as churn cancels out new customers and you end up at net new = 0. This is called fin tailing. And it's why poor retention is the silent killer So what did we do? We started patching leaks in our funnel. It took time. And we almost ran out of water. But now *most* of the leaks are patched. And as we add water to the bucket, even in small quantities. The bucket fills up.

63

6 comentarios