John Milinovich

Their goal is to create 1 billion personas as synthetic characters for various scenarios. I'm going through the first release. The released data is all generated by public available models (GPT-4, Llama-3 and Qwen), and I have been using them simultaneously along with Gemini, Claude and Mistrel. Synthetic Data Samples: 50,000 math problems 50,000 logical reasoning problems 50,000 instructions 10,000 knowledge-rich texts 10,000 game NPCs 5,000 tools (functions) A subset of PERSONA HUB: 200,000 personas

4

Stanford University's latest Foundation Model Transparency Index (FMTI) includes Writer LLMs, and the results are impressive! This was the first time Stanford Institute for Human-Centered Artificial Intelligence (HAI) evaluated Writer’s Palmyra-X model in FMTI, reflecting our growing impact to building customer trust through transparency, especially in enterprise applications where security, accuracy, and data protection are paramount. Palmyra-X achieved some of the highest marks for enterprise-grade models based on a set of 100 indicators. But in so many ways, our commitment goes far beyond FTMI expectations. At Writer, we understand that enterprises need top-notch performance paired with solid assurances about data management and protection. Our strategy isn't just to meet industry standards — it’s to surpass them. 🔍 Here are just some of the additional ways we're setting higher standards with our customers: Zero data retention: We never store data longer than is needed to provide our service, and give customers the right to delete data at any time. No training on customer data:
We don’t use the data you share with us to create, modify, or train our models. Zero model distilling & quantization: We do not distill or quantize our models, so customers can achieve peak reliability of accuracy, quality, and performance at all times. Toxic filter transparency: We share the mechanisms of our toxic content filters, ensuring our customers understand the safeguards we have in place to prevent harmful content generation. Bias mitigation: We’re transparent with our customers about our ongoing efforts and methodologies to identify and mitigate bias in our models. We’re providing our customers with technology that's not only powerful but also secure, and empowering them with the knowledge to make informed decisions. Our participation in this research is just the first step. We’re on a journey to raise the bar on transparency and redefine industry standards. Stay tuned for more to come! Link: https://lnkd.in/g8sMU2Qx

218

7 Comments

Y Combinator, an influential early-stage startup accelerator once led by OpenAI's Sam Altman, has notably increased its focus on AI, reflecting a broader trend where AI integration is becoming crucial for startups. Despite the more selective nature of its recent cohorts, the 2024 winter batch saw 64% of startups focusing on AI-related themes. Renowned for its impact in shaping the tech landscape, Y Combinator has nurtured successful companies like Airbnb and Coinbase. Its success formula includes supporting tech-savvy founders within a collaborative environment and making strategic investments across diverse, innovative projects. The evolving focus suggests that AI is increasingly seen as essential for new startups to gain traction and compete effectively in the rapidly evolving tech ecosystem, thereby extending YC's influence across many sectors and pushing the boundaries of AI applications. 🌐⇢ https://lnkd.in/dB7562fc

42

2 Comments

I'm doing a free webinar: Crafting your Product Vision in the Age of AI https://lnkd.in/eEwfp4Bm Tuesday, Oct 29th at 4pm CET It's a short half hour session where there will be time for questions and I can even stay later to talk about Product Vision if folks have questions. We'll be talking about how the speed with which AI enables people to launch products, run experiments, and A/B test their way to awesomeness impacts Product Vision. Do you still need one? Is it better to be a startup or an incumbent when things are moving this fast? So stop on by! #productvision #freewebinar #learning

27

2 Comments

Together AI Introduces Mixture of Agents (MoA): An AI Framework that Leverages the Collective Strengths of Multiple LLMs to Improve State-of-the-Art Quality Quick read: https://lnkd.in/gejHtYDp Paper: https://lnkd.in/g8sbqAWn GitHub: https://lnkd.in/gYsmsZa6 Together AI

38

2 Comments

AI isn't just about replacing humans—it's about empowering them. 💪 AssuranceIQ, an innovative insurance startup, is pioneering a groundbreaking approach to leveraging generative AI: 1. They're using Large Language Models (LLMs) to score customer conversations. 2. This AI-powered system structures unstructured data from interactions. 3. The goal? Enhance the performance of human sales and support teams. 4. It's not about AI vs. humans, but AI amplifying human capabilities. 5. This could revolutionize how we train and develop customer-facing teams. This use case showcases the true potential of AI in business: not as a replacement, but as a powerful tool to augment human skills and drive performance. What other innovative ways could we use AI to empower rather than replace human workers? #AIInnovation #CustomerExperience #FutureOfWork #BusinessAI #TechTrends Share your thoughts on AI augmentation in the comments below!

10

1 Comment

As academic based LLM benchmarks are losing their relevance -- there are 2 types of Benchmarks that will matter more in the B2B space: 👉 Crowdsourced user votes to rank LLM output (like Chatbot Arena) OR where the test DATASET is kept secret by the evaluator (e.g GSM1k from scale_AI) 👉  Enterprise use case specific LLM benchmarks (e.g produced by CRM, ITSM, HR SaaS companies) evaluated on real data. Because you can not game real customer use cases. #llm #benchmarks #llmeval #AI

11

1 Comment

📚 Book of the Week (Oct 17-21, 2024): Next week, we're featuring "Build a Large Language Model (From Scratch)" by Sebastian Raschka, PhD. Participate and get a chance to win a free copy! 👇🏼 How to participate: 🔸 Register on our Slack 🔸 Join the book-of-the-week channel 🔸 Ask questions to the author from Monday through Thursday 🔸 Win a free copy of the book! The author will choose the winners on Friday What you'll learn from this book: 🔸 The inner workings of LLMs through code, diagrams, and examples 🔸 How to create your own small but functional model 🔸 Insights into training methods behind large-scale models like those powering ChatGPT 🔸 Fine-tuning larger pre-trained models with the provided code Don't miss this chance to deepen your understanding of LLMs and engage directly with the author. https://lnkd.in/eDTqBnzv

414

5 Comments

The Benchmarking Challenge ‼ As generative models like LLMs demonstrate remarkable capabilities in language understanding and complex reasoning, the need for evolving our evaluation methods becomes crucial. How do we effectively measure the progress of these increasingly sophisticated systems? Comprehensive benchmarking of LLMs at scale presents obstacles that must be overcome: 1️⃣ Cost Efficiency: Large benchmarks demand substantial energy and compute. How can we make benchmarking faster and more affordable without compromising quality? 2️⃣ Evaluating Complex Tasks: LLMs excel at multi-task, open-ended problems beyond traditional narrow benchmarks. How do we assess these broader capabilities? 3️⃣ Specialized Domain Performance: As LLMs advance in areas like code generation, how can AI community ensure benchmarks stay relevant and accurately assess evolving performance in specialized tasks? While benchmarks play a crucial role in evaluating AI systems, it's important to remember that they aren't the be-all and end-all of GenAI progress. They're an important factor in assessing AI capabilities, but not the only one. ❗ They're like academic grades in the real world - important, but not everything. We say "grades don't define you," yet we can't help comparing them. Similarly, the AI community both critiques and relies on benchmarks. 🔑 Real-world application, ethical considerations, and adaptability to diverse scenarios are equally crucial in determining #GenAI system's true value and impact. What are your perspectives on this rapidly evolving landscape? Keen to hear from researchers, developers and leaders tackling these issues.

5

1 Comment

Med42-v2 Released: A Groundbreaking Suite of Clinical Large Language Models Built on Llama3 Architecture, Achieving Up to 94.5% Accuracy on Medical Benchmarks Researchers from M42 Abu Dhabi, UAE, have introduced the Med42-v2, a suite of clinical LLMs built on the advanced Llama3 architecture. Developed by the team at M42 in Abu Dhabi, these models are meticulously fine-tuned using specialized clinical datasets, making them particularly adept at handling medical queries. Unlike generic models, which are often preference-aligned to avoid answering clinical questions, Med42-v2 is specifically trained to engage with such queries, ensuring that it can provide relevant and accurate information to clinicians, patients, & other stakeholders in the healthcare sector. The development of Med42-v2 involved a two-stage training process designed to optimize the models for clinical use. The Llama3 models were fine-tuned in the first stage using a curated dataset that included medical and biomedical information, chain-of-thought reasoning, and conversational examples. This stage accounted for 26.5% of the final training dataset and enhanced the models’ ability to understand and generate responses relevant to clinical contexts. The second stage focused on preference alignment, ensuring the models’ outputs aligned with human expectations and ethical standards. This stage utilized datasets such as UltraFeedback and Snorkel-DPO, allowing the models to be iteratively refined to meet clinical requirements..... Read our full take on Med42-v2: https://lnkd.in/gcz8fg-X Paper: https://lnkd.in/gDtsJvst Model Cards: https://lnkd.in/ggKmkheg M42 Health Clément Christophe, PhD Praveenkumar Kanithi Tathagata Raha Shadab Khan

43

good vibes are all you need

22

There’s no single UX approach for using LLMs today. I see companies trying 3 main approaches: 1) Chatbots, 2) Integrating with existing interfaces, and 3) Agent-driven workflows. Let's explore. 1.) Chatbots Imagine you have tons of data and simply add a chatbot. I see many founders wanting to go this route: they’ve got a successful legacy business and now want users to query a chatbot powered by LLMs plus their own data. It’s usually the first thing people try, but it’s unlikely to be the long-term answer for most cases. Do we all want endless chatbots? Probably not. It makes users think too much before they get clear value from the product itself—they’re left wondering, “What am I supposed to prompt this thing with?” 2.) Integrating into Existing Interfaces This approach is closer to how we use products today. Here, you upload data and receive a tailored analysis. It takes more planning to decide what those outputs should be, but it’s in line with existing experiences. Think of LinkedIn’s “Get an InMail drafted” feature—GenAI integrated directly into familiar workflows. It’s new yet feels familiar, and the product does the heavy lifting in the background, creating the right prompts to make things seamless for the user. 3.) Agent-Driven Workflows Then there are agents that run workflows on behalf of users. I haven’t yet incorporated an agentic workflow into my daily routine, but this seems to be where we’re headed. It’ll likely take some time before this becomes a standard way of interacting with these tools, but the potential is there. MindStudio, Cassidy, and Zapier are doing a great job in powering these use cases. The best products will likely blend these methods; for example: - If you’re onboarding someone in India, maybe lean into voice features, knowing voice notes are widely used there. -For someone over 50, keep it super simple—they don’t need to know it’s AI working behind the scenes. -Or, if you’ve got a tech nerd, let them explore agentic workflows. Our team at Casper Studios is spending a lot of time thinking critically about these with customers. What else are you thinking about from a UX x AI perspective?

17

6 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

Just published a #neurotech newsletter, it became kind of quarterly rather than monthly 🤦‍♂️. There's a lot of cool stuff there. Let me know what did I miss. Link: https://lnkd.in/etwKkA54 I. Reading List 1) Techstack/Hardware/Wetware * Non-Invasive - two studies on how non-invasive (fUS and VNS) stimulation improves cognitive performance; closed-loop TMS; fUS to increase local attention and filter-out external information; * Invasive/Minimally Invasive - false memory in hippocampus & neural prosthetic to facilitate human memory encoding and recall; inner speech decoding; self-stiffening intracortical microprobe; flexible electrodes implantable via <2mm hole; flexible electrodes spanning multiple brain areas; <15 µm electrodes in visual cortex; directly printing electronics on the cranial surface; optical probes; spinal cord stimulation to convey functionally unrelated sensory input to the brain; 2) Data Science - a model for intrinsic and input-driven neural population dynamics; a foundation model for SEEG/EEG; mental imagery decoding from fMRI; LLMs applied to EMG data for silent speech reconstruction.   3) Misc Neurotech Reading - feedback loops between BCI commercialisation and research acceleration and between AI and neuro research fields; Petavoxel fragment of human cortex and brain organoids; A blue-print of how neurtoech may move to the clinical frontline, based on the TMS' adoption trajectory; Payers, social consensus, and non-tech BCI challenges. Neuroethics & neurorights. II. Startup/Corporate News 1) Business Reading - Neurotech and international security, what at the macro/country level affect neurotech adoption; a story of a bionic eye startup; vertical integration in the BCI world. 2) Funding - EEG-powered for hearing test, consumer/productivity applications; Electrical stimulation against depression, gait and balance deficit, headaches, for better sleep, and menstrual pain. EMG for consumer devices and prosthesis. Light and sound for depression and sleep. #bci #neurotech #venturecapital #frontiertech #deeptech

35

2 Comments

𝗪𝗵𝗲𝗻 𝗔𝗜 𝗠𝗲𝗲𝘁𝘀 𝗕𝗶𝗼𝗹𝗼𝗴𝘆: 𝗔 𝗡𝗼𝗯𝗲𝗹 𝗣𝗿𝗶𝘇𝗲-𝗪𝗶𝗻𝗻𝗶𝗻𝗴 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻 🧬🤖 𝗧𝗟;𝗗𝗥: 𝗔𝗜 𝗷𝘂𝘀𝘁 𝘄𝗼𝗻 𝗮 𝗡𝗼𝗯𝗲𝗹 𝗣𝗿𝗶𝘇𝗲. 𝗪𝗲𝗹𝗹, 𝘀𝗼𝗿𝘁 𝗼𝗳. 𝗟𝗲𝘁'𝘀 𝘂𝗻𝗽𝗮𝗰𝗸 𝘁𝗵𝗶𝘀 𝗴𝗮𝗺𝗲-𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝗺𝗼𝗺𝗲𝗻𝘁 𝗳𝗼𝗿 𝘀𝗰𝗶𝗲𝗻𝗰𝗲 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵. Two researchers from Google DeepMind and one from the University of Washington have snagged the prestigious award for their groundbreaking work in protein prediction and design using AI. AI is rewriting the rules of scientific discovery. 🚀 In my GenAI workshops with brands, I often get asked: "Is AI really a big thing? Will it actually impact what I do?" Usually, I focus on how AI can tackle small but crucial routine tasks. Start small, I say. But this? This is BIG. This is the other end of the impact spectrum. 𝗔𝗜 𝗶𝘀 𝘀𝗼𝗹𝘃𝗶𝗻𝗴 𝗮 𝗿𝗲𝗮𝗹 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 Predicting protein structures used to be a painstaking process, taking months or even decades. With AlphaFold, the same task is accomplished in hours or minutes. That's not just an incremental improvement; it's a revolutionary leap forward. 💡 𝗧𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 𝗶𝘀 𝗯𝗶𝗴 AlphaFold has been cited in over 20,000 scientific studies. It's not just a niche tool for a handful of researchers - it's transforming the entire field of biochemistry. This AI is accelerating the discovery of new medicines, potentially saving countless lives. We're talking about impact on a global scale. 🏆 𝗧𝗵𝗲 𝗻𝘂𝗺𝗯𝗲𝗿 𝗼𝗳 𝗽𝗲𝗼𝗽𝗹𝗲 𝘄𝗵𝗼 𝘄𝗶𝗹𝗹 𝗯𝗲𝗻𝗲𝗳𝗶𝘁 𝗶𝘀 𝗵𝘂𝗴𝗲 Dr. Jumper, one of the laureates, put it perfectly: "We can draw a straight line from what we do to people being healthy." Beyond human health, this tech could lead to the development of enzymes that efficiently break down plastic bottles, making recycling easier and more effective. The beneficiaries? Potentially every single one of us and our planet. This is technology that could improve human health and combat climate change simultaneously. 🌍 Now, let's zoom out for a second. This Nobel Prize is more than just a pat on the back for some smart folks. It's solid proof that AI is already having a material and meaningful impact on the world. And that makes me happy. 😊 This feels like the validation we've been waiting for. Slowly but surely, we can confidently talk about meaningful progress made possible by AI. It's not just hype or potential anymore - it's Nobel Prize-winning reality. What do you think? Are you excited about the potential of AI in scientific research? Have you seen AI make a difference in your field, big or small? Drop your thoughts in the comments - let's get this Nobel-worthy discussion going! 👇 ----------------- Hi! 👋 I'm Patrick, founder of MVP Studio. We turn Tech dreams into storytelling realities. Follow me for more insights on the intersection of AI, science, and storytelling. Ready to craft your own Nobel-worthy narrative? Let's connect!

25

3 Comments

What does our future actually look like if we realize the full potential of regulated data? As Integral powers more and more varieties of regulated data across different use cases, I’m seeing more and more that regulated consumer data is going to be impactful in so many ways: - AI-powered early disease detection: Imagine new AI systems like @Paige-AI analyzing your routine medical scans in real-time, potentially catching cancers or other diseases at their earliest, most treatable stages. - Precision medicine breakthroughs: With rapid access to vast pools of genetic and health data, researchers could develop treatments tailored to your specific genetic makeup, dramatically improving efficacy and reducing side effects. - Predictive mental health interventions: AI could analyze patterns in your digital behavior, sleep, and biometric data to predict and prevent mental health crises before they occur. - Automated dietary management: Imagine an AI nutritionist who instantly adapts your meal plans based on your real-time blood sugar levels, activity data, and long-term health goals. - Personalized environmental health alerts: Systems could combine your health data with real-time environmental monitoring to provide personalized air quality or allergy alerts specific to your sensitivities. - Smart pharmaceutical management: Your smart pill dispenser could communicate with your health records and current biometrics to adjust medication dosages in real time, ensuring optimal efficacy and minimizing side effects. At Integral, we’re excited to build the infrastructure necessary to get the right data on people like you and me so companies can create these seamless experiences without sacrificing privacy. https://lnkd.in/e8xgbxap #DataCompliance #DataRegulation #DataPrivacy

26

3 Comments

Are LLMs better at Deductive (start with a general formula/relationship and compute outputs for different inputs) or Inductive (start with examples of inputs and outputs and figure out the generalized relationship) reasoning.... The answer is they are better at Inductive reasoning... Based on this what kind of applications are they better suited for???? More details in this amazing study....https://lnkd.in/gjKYUraW

13

2 Comments

Microsoft'𝐬 𝐕𝐀𝐒𝐀-𝟏 𝐰𝐢𝐥𝐥 𝐛𝐞 𝐭𝐡𝐞 𝐧𝐞𝐱𝐭 𝐬𝐭𝐞𝐩 𝐢𝐧 𝐜𝐨𝐧𝐭𝐞𝐧𝐭 𝐂𝐫𝐞𝐚𝐭𝐢𝐨𝐧!! Microsoft Research recently unveiled their groundbreaking AI framework, VASA-1, described in a detailed paper released on April 16 on arXiv. This innovative technology pushes the boundaries of digital interaction by generating hyper-realistic talking faces that convey a wide range of emotions through visual affective skills (VAS). 🤖 What is VASA-1? What makes VASA-1 standout from previous photo to video technologies, is the level of customisation that it brings in. Previous popular methods would articulate just the face regions to lipsync to an uploaded audio clip to fit the face to the audio output. While Vasa-1 allows for a lot more fine control thus providing realistic outputs with the same photo input. 👁️ Core Innovations Explained The framework is built on a diffusion-based model that refines images from a blurry start to crystal clarity. It uses a holistic approach to facial dynamics and head movements, operating within a specially developed 'face latent space' and 'expressive latent space' that captures the essence of facial features, facial expressions and emotions separately yet comprehensively. Thus providing the ability to move the photo around, change gaze location, change expression, change scale and much more in the photo while maintaining consistent and reliable outputs. 📈 What more? Now, Imagine All of this running realtime on a Home PC!! Thats exactly what VASA-1 is capable of. A single Nvidia 4090 GPU is capable of running VASA-1 realtime producing video output at 40fps, with a 512x512 resolution !! So, as expected VASA-1 is in research preview and not available outside of Microsoft research team. That being said, if released to public the ramifications would be far reaching. We at SoftServedWeb operate at the cutting edge of AI, AR/VR while developing our solutions. Being upto date with the latest happenings in the industry helps ensure that we provide the latest and best when it comes to meeting our client requirements. Source : https://lnkd.in/g-qJ5R9a Video Source : Microsoft VASA-1 anouncement. #AI #MicrosoftResearch #Deepfake #DigitalIdentity #Innovation #SoftServedWeb #Solutions

7

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