Treebel Solimani

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

Great discussion coming up regarding under-the-radar opportunities in the AI landscape. Registration link in the comments section👇🏼 #ai #artificialintelligence #nvidia #openai #investing

9

1 Comment

NVIDIA is partnering with Abu Dhabi-based G42 to create a climate technology lab, marking its first partnership the UAE's AI firm. The 2 companies will collaborate on developing AI that can improve the accuracy of weather forecasting & they will rely on Nvidia's Earth-2, an open platform that can make climate & weather predictions. In April, Microsoft announced a $1.5 bn strategic investment in G42. And the investment is supposed to strengthen both companies' collaboration on bringing the latest Microsoft AI technologies & skilling initiatives to the UAE & other countries around the world. On September 17 the companies announced the establishment of 2 new centers in Abu Dhabi focused on helping develop industry standards for the use of AI. According to Statista, the AI market in the UAE is projected to reach $4.2 bn by 2030. The partnership between Nvidia & G42 has the potential to strengthen the UAE for it to be an AI superpower in the Middle East. #nvidia #climatetech #environmentalsustainability #sustainability #abudhabi #technews #middleeast #ai #macroeconomics

1

AI Architecture: Test Time Training (TTT) more efficient than transformers? Worth to do a knowledge deep dive. #artificialintelligence #ai #technology #tech #deeptech #venturecapital #vc #startups #invest #business #funding

4

2 Comments

GitHub is a treasure trove of innovation knowledge, accessible for the world to freely access. At the forefront of driving AI adoption are AI Agents, especially tokenized AI agents. AI Tidbits, a substack blog on AI, provides a non-exhaustive, overview of the open-source AI Agents (non-tokenized) ecosystem, geared for developers creating these AI agents. While numerous market maps exist for AI including ones we at Decasonic have published, they often cater more to venture capitalists than hobbyists, innovators and developers. Developers need actionable tools and frameworks to launch functional AI agents today and it’s great to read and explore these resources on GitHub.

28

3 Comments

Congrats to Alex (Wei) Chen & Zack Li on the $15M round! It's so exciting to see the awesome innovation you are doing (very impressive demos!) with device side AI - and super exciting breakthroughs in performance, multi-modal interactions, and orchestration across functions and apps on the edge. The next big frontier of AI innovation is on the edge (i.e. device side). To me, it's the analogous to the era when compute moved away from mainframe and the server (i.e. today's big AI in the cloud) to a decentralized model.... and Alex, Zack, and the team at Nexa AI are leading the way - going far beyond the basic SLMs and building foundational technology for the future of GenAI, the edge.

8

1 Comment

Here is a really interesting update from LLM360 research group where they Introduce 'K2': A Fully-Reproducible Open-Sourced Large Language Model Efficiently Surpassing Llama 2 70B with 35% Less Computational Power This model, known as K2-65B, boasts 65 billion parameters and is fully reproducible, meaning all artifacts, including code, data, model checkpoints, and intermediate results, are open-sourced and accessible to the public. This level of transparency aims to demystify the training recipe used for similar models, such as Llama 2 70B and provides a clear insight into the development process and performance metrics. The development of K2 was a collaborative effort among several prominent institutions: MBZUAI, Petuum, and LLM360. This collaboration leveraged the expertise and resources of these organizations to create a state-of-the-art language model that stands out for its performance and transparency. The model is available under the Apache 2.0 license, promoting widespread use and further development by the community. LLM360 has provided a robust set of evaluations for K2, encompassing general and domain-specific benchmarks. These evaluations cover medical, mathematical, and coding knowledge, ensuring the model performs well across various tasks and domains. The LLM360 Performance and Evaluation Collection and the K2 Weights and Biases project document a detailed analysis of K2’s performance..... Read our full take on K2 here: https://lnkd.in/gMX5kVFT Model: https://lnkd.in/geFVw8R4 LLM360 #llm #artificialintelligence #opensource

39

2 Comments

Multimodal AI is unlocking incredible new possibilities! 🤖 The convergence of different AI modalities is setting the stage for groundbreaking innovations in the B2B AI SaaS landscape. Exciting to see how these capabilities can transform industries and open new doors for startups. 🚀 #AI #MultimodalAI #Innovation #B2B #SaaS #TechTrends #VentureStudio #AITransformation

5

🌟 We're in the Spotlight! I spoke with Datanami on tackling one of the most challenging aspects of AI today: hallucinations in #LLMs. LLMs inherently face issues due to the high compression rates of data, significantly surpassing the 12.5% threshold identified by Claude Shannon, the Father of Information Theory. This often results in what's known as "hallucination," where models generate incorrect or fabricated information. At Vectara, we're pioneering solutions to reduce these inaccuracies through innovative methods such as #RAG Retrieval-Augmented Generation and our proprietary fact-checking LLM, Boomerang. These techniques ensure that our AI models maintain high fidelity to the facts, drastically reducing the hallucination rates experienced by conventional LLMs like GPT-4. Our mission is to empower businesses with reliable, fact-based AI tools, enhancing the utility of AI without compromising on accuracy. Discover more about how Vectara is leading the charge in creating dependable AI solutions, perfect for any company aiming to integrate advanced AI responsibly. 🔗

51

1 Comment

Congratulations Rana el Kaliouby, Ph.D.! Been following you ever since we worked together for the Cisco Data Symposium 2020. 💙 Rana created Affectiva and has done so much in the field of #AI. So if you're interested in learning from leaders in the #ArtificialIntelligence world, be sure to check out her new podcast - Pioneers of AI!

3

🚀 Revolutionizing High-Speed AI Integration SambaNova and Gradio's partnership is unlocking new pathways for developers, simplifying the integration of high-speed AI models. With just a few lines of code, accessing groundbreaking AI inference platforms has never been easier, helping to accelerate innovation in AI applications. ⚡ Streamlined Development This exciting collaboration empowers developers to rapidly create and deploy web applications using SambaNova’s AI, offering unmatched speed and efficiency while saving valuable time. 🌍 Navigating the AI Landscape As AI technology evolves, companies must consider performance, sustainability, and scalability in their strategies, ensuring responsible innovation. #AIEvolution #HighSpeedAI #AIIntegration #SambaNova #Gradio Read more: https://lnkd.in/eCNYFfpN

1

1 Comment

I tested if Claude 3.5 Sonnet could replace me. It changed my mind on LLMs and fraud detection. The first time I tried testing Claude on fraud detection, I made a mistake: I fed it with a dataset of 10k records. Of course, that was way beyond its limitations. But last week it hit me: What if I don’t feed it with big data like a “regular” Machine Learning model? What if I ask it to operate with very little data instead? So I designed a very simple experiment: • 99 transactions only • No fraud labels • No training data • No data enrichment • A prompt my mother could write After 30 minutes I already had the initial results. It took me 30 more minutes to collect my jaw from the floor. Can Claude replace a veteran fraud investigator? Not now. At least, not yet. But it can: • Score fraud 64% higher than non-fraud cases • Identify fraud/safe signals • Detect unreported fraud And all of that with no expertise, little data, fast and for free. This week I’m going to break it all down in a special “The Saturday Fraud Strategist” edition. If you’re not subscribed yet, the link is under my picture. ⬆️ See you on Saturday! ------------ Enjoy this? ♻️ Repost it to your network and follow Chen Zamir for more. Want to dive deeper? Join "The Saturday Fraud Strategist". Each week I'll be sharing detailed advice on how to grow Fintechs safely. ⬆️ Click "Subscribe to newsletter" under my name ⬆️

65

18 Comments

Unlock deeper insights with AI Key Insights from Jo Lambadjieva, Amazing Wave: Claude Opus offers a superior grasp of semantic contextual meaning compared to ChatGPT, making it more effective for analyzing text-based data. While ChatGPT counts word mentions, Claude intelligently recognizes synonyms and related phrases, aggregating them into meaningful insights. This capability allows users to identify trends in reviews more accurately, such as grouping "cheap price" and "cost efficient" as one concept. Experimenting with different AI models highlights the importance of choosing the right tool for nuanced data analysis. #AI #DataAnalysis #Amazon #ChatGPT #ClaudeOpus #MachineLearning #Sales #Ecommerce #Innovation #Technology

9

6 Comments

🇦🇪🇸🇦🇶🇦 In the final piece of our token use cases series, we highlight how the Middle East's proactive regulatory approach demonstrates that localized efforts and strong public - private partnerships can drive global progress. The Middle East, particularly Dubai, continues to set the pace in global blockchain innovation. It serves as a prime example of how optimism and collaboration between public and private sectors can drive transformative progress. This approach is laying the groundwork for significant advancements in blockchain and Web3, with clear regulatory frameworks that attract institutional interest and accelerate mainstream adoption. We explore how countries like the UAE, Qatar, and Saudi Arabia are leading the way with comprehensive regulatory frameworks and practical use cases in stablecoins, DeFi, and payments. These frameworks are giving institutions the clarity they need to innovate, scale, and lead in areas like e-commerce, gaming, and AI. At Decasonic, we see the Middle East as a pivotal player in the next wave of blockchain adoption, setting standards that other regions can learn from and adapt. My Jordian American colleague, Abdul Al Ali, leads this effort for us. Kudos to the many industry leaders driving this important innovation: Marwan Alzarouni, Mohamed Hamdy, Omar Rahim, Saqr Ereiqat, Amro Awaidah, Sep Alavi, among the many people I have the privilege to call friends. #Blockchain #CryptoRegulation #Web3 #TokenEconomy

26

4 Comments

California's Senate Bill 1047 is deeply flawed. Holding developers liable for misuse of their models won't improve AI safety, but it will negatively impact innovation. Like many other startups, our technology is possible because of access to open source LLMs. Companies of every size, from enterprises to start-ups, benefit from open source innovation. Sharon Zhou, PhD recently spoke on a panel with Matt W., GM of AI at The Linux Foundation, Charles Xie, CEO and founder of Zilliz and creator of Milvus, and Christopher Cuong T. Nguyen, CEO and co-founder AITOMATIC about the importance of an open ecosystem and implications of SB 1047. Thanks to The AI Alliance and Cerebral Valley for bringing more awareness to SB 1047!

153

5 Comments

Acquisitions reveal a lot about ambitions. NVIDIA is actively acquiring AI startups. 2023 saw a surge in Nvidia’s dealmaking activities. The trend continues in 2024. This year, they bought Run: ai and Deci AI. These companies specialize in optimizing AI workloads. Nvidia’s focus on AI performance and optimization is no news. This acquisition spree enhances its AI capabilities. Major corporations acquire startups to: - Access latest technology - Innovate faster - Gain new talent - Strengthen market position PayPal started acquiring BNPL startups before offering the Buy-Now-Pay-Later option on its main platform. Acquisitions are done to drive growth faster. You don’t buy companies; you buy growth. 𝙄𝙣𝙩𝙚𝙧𝙚𝙨𝙩𝙚𝙙 𝙞𝙣 𝙢𝙤𝙧𝙚 𝙈&𝘼-𝙧𝙚𝙡𝙖𝙩𝙚𝙙 𝙘𝙤𝙣𝙩𝙚𝙣𝙩? 𝙈𝙖𝙠𝙚 𝙨𝙪𝙧𝙚 𝙩𝙤 𝙛𝙤𝙡𝙡𝙤𝙬 𝙢𝙚. #entrepreneurship #venturecapital #startup #mergers #acquisitions Serial Entrepreneur & Investor Helping Startups Become Unstoppable – David Hauser

77

4 Comments

OpenAI ‘s sama. the video that’ll sink sama. the look on sama’s face when asked by kennedy if he makes a lot of money. ‘ I have no equity in OpenAI ‘ ‘ I love my job ‘ given the recent emails that are out from years ago between elon, sama, greg brockman… well … it appears that he’s about to confront a different type of cannibal. in the form of the now dubbed by the donald … the great elon musk. it appears that this was never about saving humanity. the threat of a new persuasion machine competing against the other persuasion machines in the form of news, media as we’ve known it. including social media. perhaps there’ll be a polymarket bet. chances that sama no longer is ceo. chances that sama goes to prison as in the case of sam bankman-fried. chances that Microsoft ditches OpenAI. chances that pg is proven wrong that sama left on an island of cannibals would become its leader. chances that OpenAI never becomes for-profit. chances that OpenAI get stiffed as elon was. chances bill gates throws sama under the bus. chances sama, billy gates, greg brockman, reid hoffman, vinod khosla pay a multi-billy penalty to elon. in the recent filing it states that elon invested over 44 millys towards OpenAI and sama misled him. pretty crystal clear breech of trust between them. sum ‘ting tells me that by unravelling this cabal we will achieve ai safety. Ilya will be vindicated. this puts a big smile on our faces. why? no one likes being played. we are not pieces on a chess board. also it appears that sama didn’t put any of his own $ into OpenAI yet single handedly proclaims himself as ceo. with greg brockman as his useful fool. nerd drama. if elon can parallel park a rocket booster. putting sama in his place will be … autonomous. lol. source: https://lnkd.in/e56AuyUF < . >

1

1 Comment

Excited to have Hamza Farooq, AI guru from Stanford University join FastLearner. Hamza’s first course on the platform will cover building your LLMs - meaning building your very own ChatGPT-like systems. It will be live exclusively on FastLearner.ai. Stay tuned! #AI #LLM #GenerativeAI #GenAI #VectorSearch #FastLearner

14

2 Comments

Peteris Erins and I spent time this year diving deep into the Starknet Ecosystem in collaboration with StarkWare and the Starknet Foundation on the future of Ethereum and Starknet and Cairo's unique role in the cutting edge of the rollup-centric roadmap and the validity-proven future of blockchains Here's 13k words from us at Auditless on why you should be taking Starkware and Cairo seriously right now https://lnkd.in/dNh8YD8V

15