Nicko Trataris

A few, short video clips made with Luma from Luma Labs. As with all AI, the outcomes are highly dependent upon the prompt. Here are the results of 3 different prompts in an attempt to get a group of superheros working together at a conference table. Perhaps this is some hint of an upcoming meeting? You know it is.... While the content results need work, and there are improvements to be made in my prompting, the ability of AI to conceptualize a business meeting with people around a table, movement, conversation and the context of superheroes is something that was not possible 6 months ago with most genAI without rendering wildly distorted body parts/movement and wildly different outcomes from similar prompts. Tremendous progress in a very short amount of time. This is more than a panda riding a skateboard, or a gen ai beauty shot or landscape. It's complex scenario... And I'm here for it. Teaser: Wait for the final frames of the 3rd video. You gotta see it.... #genai #filmai #contentai #videocontent #promptengineering

9

2 Comments

Patronus AI Introduces Lynx: A SOTA Hallucination Detection LLM that Outperforms GPT-4o and All State-of-the-Art LLMs on RAG Hallucination Tasks Patronus AI has announced the release of Lynx. This cutting-edge hallucination detection model promises to outperform existing solutions such as GPT-4, Claude-3-Sonnet, and other models used as judges in closed and open-source settings. This groundbreaking model, which marks a significant advancement in artificial intelligence, was introduced with the support of key integration partners, including Nvidia, MongoDB, and Nomic. Hallucination in large language models (LLMs) refers to generating information either unsupported or contradictory to the provided context. This poses serious risks in applications where accuracy is paramount, such as medical diagnosis or financial advising. Traditional techniques like Retrieval Augmented Generation (RAG) aim to mitigate these hallucinations, but they are not always successful. Lynx addresses these shortcomings with unprecedented accuracy. One of Lynx’s key differentiators is its performance on the HaluBench, a comprehensive hallucination evaluation benchmark consisting of 15,000 samples from various real-world domains. Lynx has superior performance in detecting hallucinations across diverse fields, including medicine and finance. For instance, in the PubMedQA dataset, Lynx’s 70 billion parameter version was 8.3% more accurate than GPT-4 at identifying medical inaccuracies. This level of precision is critical in ensuring the reliability of AI-driven solutions in sensitive areas. Read our take on this: https://lnkd.in/gKamTdSU Paper: https://lnkd.in/gCEGBmkn Details: https://lnkd.in/gDpdCgHj Download Lynx-70B on Hugging Face: https://lnkd.in/g7JJpwjq Download quantized Lynx-8B on Hugging Face: https://lnkd.in/g7wsV8fp Anand Kannappan Patronus AI

62

💡Seeking efficiency in an AI-driven world, we’re witnessing a game-changing strategy: chunking. It seems we could be on an exciting threshold where complex problems are untangled more easily with technology.🚀 Spearheaded by notable quantum intelligence expert — Dr. Han Heloir, the principle of chunking has been ingeniously applied in Generative AI applications. Here's a key insight: ✔️Just as splitting a book into chapters aids easier comprehension & navigation, 'chunking' simplifies AI processing by breaking down complex information into smaller, more manageable parts, thereby enhancing the AI performance. Forecasting 📊: ⏩ Such an innovative approach in AI could very well redefine how we perceive problem-solving and efficiency. And that's no small feat. ⏩To future-proof, businesses will need to keep pace and consider integrating chunking mechanisms to efficiently execute Information Retrieval on a profound scale, revolutionizing performance and accuracy across AI networks like nothing before. Commentary 🗒️: 💬 This rise of 'The Art of Chunking' in AI indicates that 'Smarter' might not necessarily mean 'Harder'. Our definition of smart is evolving - it's those who can simplify complexity and close the gap between efforts and results, by leveraging AI via novel means like chunking. 💬 Consider this: If a smarter AI is a lazy AI, using 'minimum effort for maximum output,’ corporations globally wanting to climb the efficiency ladder could have a dynamite digital partner in AI-powered by chunking. Let's toast ☕️ to the beginning of an exciting new chapter in the advancement of AI. Focus, pragmatism, and innovative thinking will likely serve us well, anchoring our future in an AI reality where our ambiguity towards complexity gets replaced with curiosity, and efficiency becomes the cornerstone of our AI experience. You can read further about 'The Art of Chunking' here: [Link](https://lnkd.in/d2ugi-xw) Let's revolutionize our perspective of what it means to be 'smart' in the AI realm! #AI #Chunking #Innovation

1

AI developers need specialized AI hardware because general-purpose CPUs cannot handle the highly parallel computations required for deep learning models... Most companies purchase AI chips or use cloud GPU services to develop these advanced models. The AI chip market is significant and growing rapidly. According to various market research reports, the global AI chip market was valued at around $15 billion in 2021 and is projected to reach over $70 billion by 2027, with a compound annual growth rate (CAGR) of more than 30% during this period. This growth is driven by the increasing adoption of AI across various industries, advancements in AI technologies, and the demand for more efficient and powerful computing hardware. We have all heard of NVIDIA, the biggest AI chip maker. The company had excellent results in 2023, reached a trillion in valuation and solidified its status as the leader of GPU and AI hardware markets. Despite the big players like NVIDIA, Intel and others, there is also room for startups in the AI Chip market. Let’s look at the top 5 of them. 1 SambaNova Systems Country: USA | Funding: $1.1B Focuses on building machine learning and big data analytics platforms. 2 Cerebras Systems Country: USA | Funding: $720M Cerebras is a computer systems company dedicated to accelerating deep learning. The pioneering Wafer-Scale Engine (WSE) – the largest chip ever built – is at the heart of their Cerebras CS-1 deep learning system. 3 Graphcore Country: UK | Funding: $692M Graphcore is a semiconductor firm specializing in AI and machine learning accelerators. Their goal is to create a highly parallel Intelligent Processing Unit capable of housing the entire machine-learning model within the processor. 4 Lightmatter Country: USA | Funding: $422M Lightmatter is creating engines that will fuel discoveries, drive progress, and minimize our environmental impact. 5 @Groq Country: USA | Funding: $362.6M Groq significantly streamlines computing to boost performance in artificial intelligence, machine learning, and high-performance computing. Despite the big players in the market, investing in an AI chip startup offers the potential for groundbreaking advancements in computing power, enabling faster processing of complex tasks and driving innovation across various industries. Read more: https://lnkd.in/de67rZpq -------- • Follow for more • Contact me or QLA for exclusive deals

5

#Ai is full of hard parts: “The hard part of fine tuning is building the training dataset. The hard part of RAG is creating the knowledge repository and optimizing the the search mechanism. The hard part of using long context windows is paying the huge cost.” The magic is learning a layered approach that uses the right parts for each use case. Great viewpoint, Mike McUne! #ResponsibleAi #SecureAi #VerifiableAi #PrivateAi

1

The Stacks. TLDR. (<2 min read) Sandboxes and bad questions We opened this month's Stacks by sharing our new AI integration, why Sandboxing is the right choice for our database and why this is the exact feature that you didn't ask for, but needed. Our second AI editorial shares the importance of asking good questions when doing research and how better search strategies won't completely save us. Fibro: Make a Decision Already Our fibromyalgia content gives you the confidence push you need to stop referring out these patients and making the freaking diagnosis yourself. Standards of care guidelines call for primary practitioners to just call it already. If you use the current guidelines, you're not likely to flag the wrong person. We promise. ED in men (and not the one you're thinking) Eating disorders are maybe more common in your patients with ovaries, but the consequences of missing it in the other 50% of your patients is just as serious. We share how to flag your patients with eating disorders regardless of gender, and why you need to be more careful looking at every patient that sits across from you. Thumbs down for B3 on cholesterol In our sneak peek content (typically for members' eyes only), we share the importance of clinically relevant outcomes for cholesterol care and how niacin doesn't fit the bill for cholesterol lowering. Hashimoto's and Ovarian Reserve A few important papers were published in the last 18 months creating even more of a concrete connection between HT and POI. If you're working with people TCC who have a positive TPO, you may want to read this ASAP. Want to read the whole thing? Click here: https://joom.ag/hStd

2

🆕 Benchmarks 201 with Clémentine Fourrier of the Hugging Face Open LLM Leaderboard! Why Chat Arenas are **OVERRATED**: - they bias sycophancy/assertiveness over factuality - annotators are not representative of humanity - models are argmaxxing public arena data - not reproducible - can't give fine grained test of model capabilities Why you should STOP using LLMs as Judges: - Position bias (prefer first answers) - Self bias (prefer their own answers) - Length bias (prefer long answers) - Discrete bias (can't do continuous ranges) - Reproducibility (GPT4 closed source - prefer Seungone Kim's Prometheus or Beijing Academy of Artificial Intelligence(BAAI) JudgeLM) One of our best ones 🔥

30

1 Comment

AI systems from Stability AI and Shutterstock transform 2D images into 3D meshes in seconds Key points: ✅ Image-to-3D magic: Now, you can create stunning 3D meshes directly from a single image. 📸✨ ✅Two new tools: Stability AI's SF3D and Shutterstock's new service are making 3D generation more accessible than ever. ✅How it works: These tools use advanced AI algorithms to analyze an image and create a detailed 3D representation.🧠 ✅ Fast and efficient: SF3D can generate a 3D mesh in just half a second! ⚡️ ✅ Endless possibilities: Use 3D meshes for gaming, animation, product design, architecture, and more. 🌎 Don't miss out on this game-changing technology! Explore the future of 3D creation today. GitHub: https://lnkd.in/dQGd3Kzn HuggingFace: https://lnkd.in/dswU57HK Research Paper: https://lnkd.in/dRCG5Kib #ai #3d #aitools #tech #trends #animation #productdesign #architecture

4