Ray Leihe

Example of how increased inference speed creates qualitatively new opportunities: Have a conversation with Vapi at https://vapi.ai. Thanks to ultra-fast inference of things like Llama-3 running on Groq, you can have a back-and-forth conversation that feels very human. You can even interrupt Vapi mid-conversation and it will just handle that seamlessly. Pretty awesome.

4

2 Comments

Sometimes, software innovation doesn’t mean reinventing a category. Sometimes it means taking legacy software that works and making it better, faster, and more affordable. Take LiveRamp, for example. - We’ve had conversation after conversation with data and advertising teams who complain about the same issues with their data onboarding software: - “I’m tired of uploading CSVs and maintaining custom data pipelines.” - “Onboarding new audiences takes days, sometimes weeks.” - “I can’t move fast enough to experiment with new audiences.” - ”LiveRamp is overpriced, especially for the value we get.” We saw a market with only one major software player delivering software that customers were clearly getting value from but had a lot of fundamental issues with. So we built Match Booster. It delivers the same (and often better!) match rates across paid media channels, in a modern way. Better, faster, and more affordable. And our customers love it. Every customer who has switched has eliminated manual work, sped up campaign launch times, and significantly reduced their vendor costs. What software or market do you think needs this kind of innovation next?

126

2 Comments

AI is not slowing down, and the scaling laws continue to hold. Claude 3.5 Sonnet is now probably the most powerful model available, but Sonnet isn't even Anthropic's biggest model—Opus is. A particularly impressive number here is 59.4% on GPQA diamond questions. GPQA is a benchmark of PhD-level questions for which you can't Google the answers, and they don't make the data available online for people to easily train on, so the contamination risk is lower. In November 2023, GPT-4 scored 38.8% on GPQA with few-shot prompting. Claude 3.5 Sonnet gets 59.4% with 0-shot prompting (while GPT-4 has improved, with GPT-4o, to 53.6%).

2

I shared a new Reforge artifact on Amplitude's PLG use cases 🔗 Use this template to: › Define your PLG use cases. › Evaluate them against 4 key factors (willingness-to-pay, motivation, ability, permission). › Outline the Growth Model Mix across acquisition, activation, and monetization. 👉 Bookmark it and share it with friends: https://lnkd.in/e9KB2EGz

36

2 Comments

For me, yesterday marked a milestone in AI agents with the release of Replit Agent. That might sound like hyperbole, so let me explain. I've long been a hobbyist programmer, and have worked professionally as an AI engineer—albeit admittedly for just 1.5 years before rejoining Klick to focus on generative AI strategy. My biggest frustrations with coding often aren't related to writing code at all. They're usually more related to setting up environments, managing source code, deployment, monitoring, secrets management, and so forth. (One exception is learning new languages that aren't as intuitive as Python, like Swift—so convoluted!) When there's so much friction to create an idea and deploy it to the world, you often just don't even work on the idea. You think of all the downstream work and you're like, "why bother?" It's demoralizing. Reducing the friction for bringing ideas to life and sharing them will lead to a lot more things in the world that solve a lot more problems. And that's what Replit has long done. It makes it very easy to create a project, write code, and deploy it, without having to worry about all the little details that tend to bog you down. (And I get that this makes you beholden to Replit's environment, but it also lets you move faster, so I think it's worth the tradeoff, at least for prototyping ideas.) Recently, they also started rolling out AI features, including coding assistance like GitHub. But yesterday, they took a big leap by adding Replit Agent, which lets you describe an idea, get a plan from the agent, and then have the agent build things while you watch and give feedback. There are, of course, other tools like this, such as Devin. But the fact that Replit is such a complete package—write, test, deploy from one interface, easily—makes Replit's agent much more convenient and impactful than anything else I've seen. After seeing this, and feeling an urge to experiment with ideas more, I slapped down the ridiculous low price of $10/month for Replit Core. That's about what GitHub Copilot alone costs, so it's great value. I immediately tried it, and it helped me create a simple app, "Story from Image," that takes a picture of you from your webcam, and writes a story based on the details in the image. (Not sharing the link to avoid draining my Replit and OpenAI credits, but attached is a screenshot.) You can learn more about Replit Agent here: https://lnkd.in/gSAuqryN PS: I have no financial incentive to say any of this or share the link. There's no affiliate code or anything. I just think this is pretty awesome and useful.

22

This is going to enable stunning levels of survey fraud. AI can now literally just look at the computer screen and take the survey.

5

3 Comments

Explainer for what Retrieval Augmented Generation (RAG) is. Something you’ll likely hear come up in conversation if you’re talking about AI, language (small, medium, large) models, or using as an natural language processing (NLP) technique for getting more context on data. RAG addresses the limitations of existing language models and paves the way for more intelligent and context-aware AI interactions. Its the process of optimizing the output (answer you read on ChatGPT, Claude, or Llama 3 when you’ve asked it a question) of a large language model, so it references an authoritative knowledge base outside of its training data sources before generating a response. But that also creates an ethics based question, how to define an ‘authoritive knowledge base’ or what data is acceptable / accurate to use to form a response. This is where you might hear the term where LLMs hallucinate when you ask it a question. Which for some reason is the word people have chosen instead of saying what it puts out is not accurate, factual, or true giving you a wrong and misleading answer where you might then go tell a friend, colleague, customer, engineer on a project… you get the picture. Credit: Steve Nouri

1

This runs long, but well worth a watch if you want to gaze into the crystal ball of what next with GenAI, agentic workflow, and ultimately AGI. Hype-free! #Sequoia #ReflectionAI #genai

3

Perfecting your product and GTM strategy are surely important. But Box CEO and co-founder Aaron Levie argues there's one factor that outweighs all others in startup success: recognizing and capitalizing on tech and macro tailwinds. Fighting against them? You're setting yourself up for failure. Our full conversation here: https://lnkd.in/gEAPFR-A

12

High-intent VERBAL buying signals for SEs: - "I need to run this by my CFO" - "How soon can we realize value?" - "What does the implementation look like?" High intent PHYSICAL buying signals for SEs: - *I'm smiling*  - *I'm nodding*  - *I shift to edge of my seat* The vocal cues are more obvious but so many miss the non-verbals. I'd argue they're more critical for that same reason. #presalesevolution #solutions #saas

29

1 Comment

In round 2 of our AI series, Mitch gets into building out the vision behind your organization's AI Strategy Framework.

2

I had a 30-minute meeting with Daniel Saks, co-founder and CEO of Landbase.com. He discussed their Agentic AI product for Sales and marketing, specifically the GTM-1 Omni solution. He also touched on the opportunities available for AI and ML engineers. This conversation likely gave insight into how their AI technology transforms go-to-market strategies and the growing demand for talent in AI and machine learning fields. Daniel Saks discusses the evolution of Landbase’s Agentic AI, its applications in sales and marketing, and the challenges faced in leveraging AI for lead generation. Highlights 🌍 Daniel Saks shares his entrepreneurial journey from a family furniture business to co-founding AppDirect and Landbase. 🤖 Landbase utilizes Agentic AI to intelligently automate go-to-market strategies for businesses. 📈 The Agentic AI market is predicted to be five times larger than the current SaaS market in the next 10-15 years. 🔍 Landbase offers insights through a Digital Trust Score, helping businesses improve their credibility. 📊 Their AI model has been trained on extensive data, optimizing lead generation and conversion strategies. 📚 Landbase provides a suite of products including outbound, nurture, and inbound solutions for effective lead management. 💪 The importance of assembling a skilled team and the challenges of data scarcity in AI development were highlighted. Key Insights 🌟 Transformative Potential of Agentic AI: Agentic AI represents a significant shift in how businesses approach lead generation, enabling them to automate and optimize strategies effectively. This could redefine market dynamics. 📊 Market Growth Prediction: The potential growth of the Agentic AI market suggests that startups can disrupt traditional software markets by offering innovative solutions. This emphasizes the need for adaptive business models. 🔑 Importance of Credibility: The Digital Trust Score reveals that a strong digital presence and credibility are crucial for successful marketing campaigns. This insight underlines the significance of brand reputation in lead generation. 💡Data-Driven Decision Making: The success of Landbase’s AI model relies on extensive data training, highlighting the necessity of having quality data for effective AI applications in businesses. 🛠️ AI Product Suite: Landbase’s offerings—outbound, nurture, and inbound solutions—illustrate a comprehensive approach to lead management, catering to various stages of the sales funnel. 🌍 Global Talent Scarcity: The challenges in finding skilled talent for AI development indicate a competitive landscape, suggesting that companies should invest in talent development and training. 🚀 Entrepreneurial Spirit: Saks encourages aspiring entrepreneurs to leverage the current era of technological advancement, emphasizing that great companies can emerge from anywhere with the right vision and conviction. #landbase #agenticai #ai #GTM #danielsaks #srinivasanramanujam

7

3 Comments

Good summary here of a new demo from OpenAI at VivaTech. Some things we've learned: • Using the desktop app, ChatGPT can interact with users directly via the camera in a real-time, interactive way that feels very human. For example, in the demo the user shows it some pictures, and a map, and gets directions. • Directions introduce a new use case that might become more common when ChatGPT can see everything you do in real-time. For example, imagine you're out and about, and need directions. Instead of opening up Google Maps to talk you through it, you can open ChatGPT, show it where you are, tell it where you want to go, and have it navigate you there—showing it landmarks along the way so it knows you're on the right path. • ChatGPT will be able to create voiceover scripts for videos. They did a demo where they showed it a Sora-generated video, and had it write a voiceover script. They also used 20 seconds of audio of the speaker to imitate his voice, then used that to record the actual voiceover. And they could change the language of the voice on a whim. The big question for me with all of this is: When will we get it? My hunch is we'll see a full rollout in mid to late summer, with new features in general rolling out slowly until we get more information and possibly a release date for OpenAI's next model possibly only by their November developer conference. After all, it will probably take people a few months to adjust to the reality of what the new ChatGPT voice mode can do. I don't think OpenAI will drop a more powerful model along with the new voice mode simultaneously. I honestly think that will create too much future shock for many people. Article here: https://lnkd.in/gVuc6ADh

6

1 Comment

Former Benchmark GP Bill Gurley found strong evidence that ChatGPT is churning a staggering 65% of their users annually. 📉 Putting this in perspective, Netflix, the gold standard for consumer subscription, has an annual churn rate of just 9%. 🍿 OpenAI's consumer churn rate suggests that their future might be in the enterprise, and also explains their recent friendly deal with Apple. 🍏 If these churn rates persist, OpenAI or another LLM could end up paying Apple for the privilege of being the default LLM, similar to how Google pays to be the default search engine on Apple devices. 🔍 What are your thoughts on this shift? Do you think LLMs can lock in users the same way Google did with search? #OpenAI #ChatGPT

3

3 Comments

There are probably 20 good pieces written on product management for every 1 written on analytics. I'm not sure why that is, because there is incredible leverage in getting your analytics org right at every stage of growth. Jessica Lachs wrote the best new thing on analytics in a long time: 𝐒𝐭𝐚𝐫𝐭𝐢𝐧𝐠 𝐚𝐧 𝐀𝐧𝐚𝐥𝐲𝐭𝐢𝐜𝐬 𝐎𝐫𝐠 𝐅𝐫𝐨𝐦 𝐒𝐜𝐫𝐚𝐭𝐜𝐡 — 𝐋𝐞𝐬𝐬𝐨𝐧𝐬 𝐅𝐫𝐨𝐦 𝐚 𝐃𝐞𝐜𝐚𝐝𝐞 𝐚𝐭 𝐃𝐨𝐨𝐫𝐃𝐚𝐬𝐡. Full essay is below - it is well worth the read.

221

19 Comments

My thoughts on Meta's ROBYN MMM model. For context, Robyn is an open source MMM product that Meta has developed in the last few years. After working on a no. of these models, here are my takeaways: 1. Its not perfect, but its much better than its previous versions A new feature is calibration frame which accommodates incrementality tests and spend, should users want to add any tests. 2. It is time consuming - There are a no. of PIPs and integrations that are needed for it to initially run, and can take more than couple of hours to just set up these integrations with #nevergrad - Even though most of the code is written, there is heavy lifting and decision making that is needed from data scientists. A no. of models are output and needs a seasoned analyst/statistician to know which models to pick and test - It does not have any EDA. You have to do it separately 3. Modeling issues - Like most MMMs, this one is also biased by spend. It heavily takes into account the spend and calibrates contribution proportionately - The optimizer reccomendations are below average. - You need to understand how to use hyper parameters and tweak them based on each model. Needs fundamentals around these 3. Documentation on Robyn's github is pretty extensive There is a facebook group of active MMM users who are helpful, vs the Github page. All in all, I think there are opportunities to improve these outputs, but for a model with less than 10 media variables, I think it does a decent job. What has been your experience? #mmm #modeling #robyn #roi Naavics

66

24 Comments

I keep hearing from our customers that Zapier has shipped some of the most interesting agentic behaviors out there. Our customers want to have Zapier Central navigate to a website, scrape the content, and send the data to an app like Google Docs. For most people, this feels like science fiction, but this is live today. With Central, you can easily have your agents search Google or go to any website, scrape its contents, and take screenshots. You can do this in real time in your Chrome browser or while you sleep. You can also use our 7,000+ app integrations and automatically integrate the data into your workflows. So next time you’re heading to a conference and need a list of the attendees in advance or are looking for a way to automatically create an advanced leads list, there’s Central. Try Zapier Central for free today. zapier.com/central. Let me know what you think! CC: Harrison Jackson who led the charge here.

134

16 Comments