Ahmed Abdelhadi Metwally

OpenAI is at it again. Yesterday, it released new models that are a significant step forward, enhancing the models’ ability to “think” through complex problems in science, coding, and math. While still in the very early days, this could be a huge opportunity to fast-forward innovation with fewer resources.The video below shows how these new models can build a simple app in just a few minutes. If NGOs can handle some of their smaller projects independently, it could open up possibilities for them to explore more advanced options in their matches with tech companies. By making these tools more accessible, tech companies can also push their commitment to social responsibility even further. It’s no longer just about offering support—but rather helping nonprofits become more independent and capable. This update from OpenAI isn’t just about advanced tech; it’s about creating stronger partnerships where both sides contribute to driving innovation. It’s a step toward reshaping what social responsibility looks like—moving from temporary fixes to long-term, sustainable impact.

22

1 Comment

🌟 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

Ilya Sutskever confirmed at NeurIPS '24 pre-training scaling law will not be the same. Internet data, the AI fossil fuel, is running out. It is not the end of the AI hype, the rapid rise of foundational AI has outpaced our ability to fully grasp their long-term societal impact. AI agents and applications are poised to transform every facet of our lives, and we will see it happening in coming few years. There will still be improvements in model architecture, and synthetic data creation, so models will improve, just not at the lightning speed they did till now. Looking forward, the single most important milestone in AI is creating systems capable of advancing themselves—AI researchers that can innovate and improve foundational models autonomously. We need to assess how far are we from this stage.

9

The zoom meeting is starting now: 🚀 Paklaunch Webinar (23rd May) Join us for a Fireside Chat between Amir Husain, Founder and CEO SparkCognition, (unicorn AI startup) and Faisal Aftab, Founder Zayn VC. We will discuss SparkCognition, AI tech, investment landscape, global macros, and more. Our guests will also take select questions from the live audience. ➖TIME: 7am (US Pacific), 7pm (Pakistan) ➖ZOOM: https://lnkd.in/dh_Jjy5J GUEST’S BIO: Amir Husain is the Founder and CEO of SparkCognition, AI based platform offering predictive analytics solution. The product offers machine learning based analytics solutions to predict defects on a combustion turbine and optimizes asset operations. It also detects and protects malware attacks, provides visibility on the process to reduce errors, and business analytics solutions. The product is used in industries such as manufacturing, defense, aviation, maritime, financial services, oil, and gas. SparkCognition has raised a total funding of $300M over 6 rounds. It's latest funding round was a Series D round in January, 2022 for $123M at post money valuation of $1.4B. Investors include March Capital, Doha Venture Capital, B Riley Financial, Temasek, Alan Howard, AEI HorizonX, JC2 Ventures, John Chambers, Dalus Capital, Founders Equity Partners, National Grid Partners, Verison, Boeing, CME Group, Pankaj Patel, and Energize Capital. Amir is the Chairman of the Board of Directors for WorldQuant Predictive that is building a portfolio of exponential technology companies underpinned by a core platform for critical business forecasts and predictions, Chairman of the Board of Directors for Navigate that is building World’s largest decentralized intelligence platform. He was also the Founding Member of Board of Advisors for IBM Watson that helped drive the new era of Cognitive Computing. Amir has been named Austin’s Top Technoloy Entrepreneur Of the Year the year, listed as an Onalytica Top 100 AI Influencer, among several other accoldaes. He serves on the Board of Advisors for the University of Texas at Austin Department of Computer Science and ont he NATO Maritime Unammed Systems Innovation Advisory Board. He is also a member of the Council on Foriegn Relations. Husain has been awarded 30+ patents and his work has featured in world-leading outlets such as Foreign Policy, Fox Business News, and Proceedings from the U.S. Naval Institutie. He is the auther of the best-selling book “The Sentient Machine: The Coming Age of the Artificial Intelligence” and a co-author of the compliation “Hyperwar: Conflict and Competition in the AI Century”.

2

From my conversations, I'm noticing different trends between how AI engineers deploy evaluation models (model as a judge) in production monitoring vs application testing. 💡 Value - Monitoring: provides real time alerts of performance or used to preemptively block problematic outputs from surfacing to the user -Testing: regression testing applications to establish baseline performance before pushing to deployment ⚡ Speed - Monitoring: near real time, requires extremely fast token processing - Testing: batch, async, often processing large test sets, lower token processing speed required 🎯 Accuracy - Monitoring: overindex on false positives, better to flag something as problematic and be wrong than to let something slip between the cracks - Testing: high correlation with human graders, be as accurate and robust as possible 💰 Cost - Monitoring: prefer cheaper models, otherwise the additional inference cost per generation will add up exponentially - Testing: willing to take on higher costs, calculation is accurate testing leads to better performing outputs 🛑 Risk - Monitoring: if your monitoring judge gets something wrong, there is a high consequence because it directly impacts the end user - Testing: if your testing judge gets something wrong, async review gives some potential to remediate the issue At the moment, it seems using model as a judge for testing is a little more popular due to the high cost constraints of constant monitoring, but that is subject to change as more specialized models come out. As the industry matures, I strongly believe we're gonna see specialization between models for monitoring and models for testing. If you're currently using model as a judge techniques, would love to chat more! Let me know what you think. Twitter Handle: https://lnkd.in/eAsQvtiu

26

Optimize for 🚀 Bio Cores First, Silicon 🤖 Cores Second David Heinemeier Hansson makes a compelling case: prioritize human programmers (bio cores) over computer processing power (silicon cores). Here’s why it matters: 👉 Investing in programmer efficiency is far more valuable than solely optimizing for computer speed. 👉 Human labor costs far exceed computing costs. Boosting human productivity should be the main focus. 👉 In the age of AI and high-level programming languages, the bio cores drive your real gains. Bottom line: Make your programmers more productive, and watch your team (and tech) thrive. Read the full article here: https://lnkd.in/gK8HEGtk P.S. : The image below is brilliantly summarized using Napkin.ai for a quick visual breakdown.

6

Over the past period I had an immense privilege of working with Andrej Karpathy (ex director of AI at Tesla, founding member of OpenAI, and our AI sensei :)), and a few cracked GPU engineers (Arun Demeure ex GPU Platform Architect at Apple, ngc92 et al) on the llm.c project. And I must admit I was pleasantly surprised yesterday getting a shout-out from Andrej and seeing my Twitter blow up with notifications 😅 (see the attached screen shot) As someone I've been looking up to since i started my ML journey up to a point where i'm casually chatting with him on a daily basis it all feels very unreal. Arun & ngc92 are extremely knowledgeable, and i learned a ton from them building llm.c. They're our CUDA whisperers. :) But, before continuing, what is llm.c? from Andrej: "...Because llm.c is a direct implementation of GPT training in C/CUDA, the requirements are minimal - there is no need for conda environments, Python interpreters, pip installs, etc. You spin up a cloud GPU node (e.g. on Lambda), optionally install NVIDIA cuDNN, NCCL/MPI, download the .bin data shards, compile and run, and you're stepping in minutes. You then wait 24 hours and enjoy samples about English-speaking Unicorns in the Andes." original tweet here: https://lnkd.in/dU93zyZU I've quickly contributed 113 commits (+ 40 pending once i merge my muP PR), among those: * I added multi-node support with NCCL initialization going via MPI/TCP/FS so that we can support both slurm and non-slurm scheduling envs * Helped make our code bitwise deterministic - so that after the training crashes we have guarantees that starting from the last checkpoing will give **precisely** the same loss & gradient numbers! * Implemented our random shuffling logic in the data loader, added various lr schedulers, and finally caught many bugs (e.g. periodic loss spikes, distributed norm computation bug, memory leaks), etc. I have learned a ton about CUDA, contributed to our layernorm kernel, and finally I now feel comfortable with basically every single part of LLM training since llm.c is such a "bare metal" project. I really feel i can build anything at this point. But I also learned you can't do it without a cracked team. :)) And, of course, it's more fun, and you make friends along the way. Over the past 2 weeks I mostly worked on adding maximum update parametrization support to llm.c: https://lnkd.in/dB2MuCH3 ran a ton of experiments & ablations and we're getting close to merging it in. The issue is that mup's benefits will become more apparent at bigger scales, while currently we're mostly in the ~1B regime. Today we celebrate llm.c milestone: 5 years after OpenAI announced GPT-2, with all the surrounding safety talks and staged release -> today you can now train your own GPT-2 in ~650$ on a single 8xH100 nodes in 24 hours. See this post for more info: https://lnkd.in/dD6Du4se Extremely excited for the future!

931

35 Comments

Sharing my thoughts on How Waymo (formerly known as the Google Self-Driving Car Project) is Pioneering Autonomous Driving with AI. Waymo has become a trailblazer in self-driving technology, and at the core of it all is AI. Leveraging advanced machine learning and deep neural networks, Waymo’s AI-powered system continuously improves through real-world data, making autonomous vehicles smarter, safer, and more reliable. Perception: AI allows Waymo’s cars to "see" and understand their surroundings by processing data from sensors like cameras, LIDAR, and radar. This helps detect pedestrians, other vehicles, and objects in real time. Prediction: The AI system predicts the behavior of everyone around the vehicle—whether it’s cars changing lanes or pedestrians crossing the street—ensuring smooth and safe driving decisions. Planning & Control: AI makes split-second decisions about accelerating, braking, and steering based on the analysis of complex traffic patterns, enabling Waymo vehicles to safely navigate roads. With millions of autonomous miles driven, Waymo’s AI is constantly learning and improving, helping create the future of safer and smarter transportation. The road ahead is AI-driven! www.rxsmeta.com #AI #AutonomousVehicles #Waymo #SelfDrivingCars #MachineLearning #SmartMobility #TransportationTech #FutureOfTransport #AIinAction #Innovation

2

I read this blog a few years back and it resonated quite strongly with me as a founder of a #SaaS company providing AI services to our customers. Thought it was worth highlighting and sharing it again in post-GPT world. AI companies often struggle with lower gross margins of 50-60%, contrasting sharply with traditional SaaS businesses reaching 60-80%+, due to hefty cloud costs and ongoing human support needs. This underscores the reality for #AI startups! Overcoming these hurdles could redefine economic strategies in tech, prompting new approaches that blend software efficiency with service-oriented flexibility to build resilient AI enterprises.

25

1 Comment

I had the pleasure of sitting down with Michael and Sharon from Builders to discuss my journey as a co-founder of orq.ai and the painful lessons I learned over the pas 20 years. What are some of your biggest learnings while building? Links also in the comments

68

10 Comments

My end-of-the-year reading? LLM Reliability by Rush Shahani. I've recently started a new job working on search at Causaly. Causaly is a platform for research in the life sciences domain. A bit different from my most recent experience focused largely on building out search for areas like e-commerce and media. Perhaps the biggest difference is how important precision is in this new domain. With e-commerce search, if a user is looking for sporty shoes and gets a result or two for socks, well, it's not that bad. But a biomedical scientist doing research trying to find a cure for the diseases that plague humanity can't afford to get inaccurate information. At Causaly, we use LLMs for a lot. Our users search through natural language and we translate those questions into queries. We summarize research papers individually and as a group. We let scientists ask questions about that research. We do a lot of work to make sure we're returning accurate answers, and there's always more we can do, and ways to increase our efficiency. Shahani's book goes in depth on how to approach these problems. Solutions like red-teaming, monitoring platforms, and human evaluation techniques. It also talks about techniques to minimize hallucinations and leveraging RAG to ground responses. The book is in early release and is still just about 1/3 complete, but I'm already learning things I'll be taking back to my job. And it looks like this will be my start-of-the-year reading, as well. https://lnkd.in/eFZfJSyp

31

3 Comments

Two weeks back went to visit one of my researchers, Ali Yassine in downtown Toronto. Yes, in 2024, we're all quite adept at working remotely, (virtually) meeting asynchronously (Gary Sorrentino - still remember our convo at the Canadian Ambassador's residence in Riyadh) and even managing, as Johnny Ho, co-founder at Perplexity, explains organizing his teams like slime mold - all that to say, still an unwavering proponent of meeting in person and seeking alignment over R&D in person all the same. Researchers are the unsung heroes of our AI profession - spearheading new ways of doing things, trial and error, research papers, developing, and experimentation are all par for the course for researchers. So a big shout out to Researchers everywhere for making the field of responsible AI, and especially applied AI built upon RWD and RWE (real-world data & real-world evidence) for certain fields such as healthcare a paramount need. The biggest shout-out to my immediate research team: Ali Yassine, Kriti S., Mahmoud El Hazzouri. I'd love to tell you what we're working on LinkedIn fam, but we're under NDA, still, can at least tell you working on a foundational AI model of a different kind, not what you'd expect. Closing comments: my working definition of the term "responsibility" is how you respond with your ability ~ openly encouraging all my peers and contemporaries in the field of AI to challenge themselves to do better when it comes to responsible and trustworthy AI for good.

59

4 Comments

Every time we sit down with enterprise CIOs and AI teams, the top of mind concern we hear is "Reliability" - how do we go from our use case to reliable AI as fast as we can? In our experience, reliable AI is extremely hard to engineer. There is no silver bullet - it is about the many long tail problems that you need to grind at for weeks and months. But, this is what makes enterprise AI extremely resource intensive today. We are building Awen to change this. We are building the core agentic infrastructure that helps AI teams ship reliable AI 100x faster - from data curation to running evals at scale. Experience more in our playground: https://lnkd.in/gFGnpcdR Abhishek Borar Himanshu Muchhal

9

Interesting talk from Prof Subbarao of ASU where he discusses if GPT like models can reason or not? Like the simple language in which he explains stuff, something which only great profs can do :) https://lnkd.in/gb2XP2uX

12

5 Comments

It was a pleasure to hear Dr. Ayesha Khanna, the resident guru of #ai, talk about the future of the space. There is not one industry that won’t be disrupted by real artificial intelligence. I don’t mean the stuff that has AI on the surface and is still powered by humans below the hood. Real AI will transform the nature of life and work whether we like it or not. It’s times like these that it is refreshing to hear a presentation at a business conference end on a note revolving around responsibility. Building Large Language Models powered by data from a biased world, will produce biased results. The question is whose vision of the future should we use to guide us? What does a fair world look like? Who gets to decide what “responsible” means? I don’t have the answers, but it’s great to know smart people are trying to tackle the problem. #echelon #e27 #publicpolicy

48

1 Comment

📚 "LLMs Will Always Hallucinate, and We Need to Live With This" by Sourav Banerjee and the team. 🧠 This is a foundational paper - if you're an AI / LLM practitioner or champion (or naysayer), this is worth reading - many of you will know this but the evidence is vital. 🔍 Summary: As Large Language Models become more ubiquitous across domains, it becomes important to examine their inherent limitations critically (hence my work on the AI Trust / Verisimilitude Paradox). 🤖 This work argues that hallucinations in language models are not just occasional errors but an inevitable feature of these systems. 🎭 The researchers demonstrate that hallucinations stem from the fundamental mathematical and logical structure of LLMs. It is, therefore, impossible to eliminate them through architectural improvements, dataset enhancements, or fact-checking mechanisms. 🧮 As I have said before - we are playing a game of error minimization, so we need to understand risk and risk mitigation. 🎯 There is still utility in LLMs, but they need to be handled and managed with care. ⚠️ We can save you time, money and help you safely navigate the ‘Age of AI’ #AI Risk Guy Digital Human Assistants Paul Edginton Ricky Sydney https://lnkd.in/gRPDET6x

3

2 Comments

Interesting thread on Twitter from Derya Unutmaz and Nick Dobos who think that medicine is going to crumble much faster to AI than any other industry. And they have good points: 1. Huge amount of gatekeeping and refusal to empower patients to make their own decisions. 2. Not an optional thing in many cases, even when you can’t afford it. So you are forced to use the next best thing. 3. Overpaid admin nightmare to get simple things diagnosed suddenly vanishes. What % of patients is that? -- "This is the final warning for those considering careers as physicians: AI is becoming so advanced that the demand for human doctors will significantly decrease, especially in roles involving standard diagnostics and routine treatments, which will be increasingly replaced by AI. This is underscored by the massive performance leap of OpenAI’s o-1 model, also known as the “Strawberry” model, which was released as a preview yesterday. The model performs exceptionally well on a specialized medical dataset (AgentClinic-MedQA), greatly outperforming GPT-4o. The rapid advancements in AI’s ability to process complex medical information, deliver accurate diagnoses, provide medical advice, and recommend treatments will only accelerate. Medical tasks like diagnosing illnesses, interpreting medical imaging, and formulating treatment plans will soon be handled by AI systems with greater speed and consistency than human practitioners. As the healthcare landscape evolves in the coming years, the number of doctors needed will drastically shrink, with more reliance on AI-assisted healthcare systems. While human empathy, critical thinking, and decision-making will still play an important role in certain areas of medicine, even these may eventually be supplanted by future iterations of models like o-1. Consequently, medicine is becoming a less appealing career path for the next generation of doctors—unless they specialize in intervention-focused areas (such as surgery, emergency medicine, and other interventional specialties), though these, too, may eventually be overtaken by robotic systems…maybe within a decade or so." Link to the thread in comments #healthcare #ai #innovation #technology #diagnostics #medical

1

2 Comments

Want to build foundational #AI models using frontier-labeled data? Kiva AI is hiring an AI Researcher (AI Scientist). Meet the hiring manager and hear about the role directly from the source. If interested, please apply directly. Kiva offers high-quality, scalable, cost-effective data labeling, annotation, and reinforcement learning from human feedback. #KivaAI blends human expertise with innovative tech to generate better outcomes for AI models. #AIforAI #AIResearch #Hiring #MachineLearning

9

Discovering the power of #Google_Colab has been a game-changer! The AI code autocompletion feature is incredible—it predicts my code with just an initial prompt, saving me tons of time and effort. While it's not always perfect, a bit of fine-tuning here and there turns it into a speedboat for my ML classes, especially with small datasets. AI is overtaking everything, making me efficient everywhere, from coding to daily tasks. As someone who loves efficiency, this tool has truly transformed my coding experience. If you haven't tried it yet, you're missing out on a powerful way to streamline your work! #GoogleColab #MachineLearning #AIEfficiency #TechInnovation

54

2 Comments