Madhusudan Pai

Great session with Eric Schmidt on the "near" future of AI. From the next steps of AI becoming a ubiquitous utility, evolving from an "assistant" to a polymath "partner", the black box (non)issue, etc... Take some time to listen to one of the OG's of the last 50 years of computer science, the internet and oh yes...Google https://lnkd.in/e3bKw_yq

2

From UC Santa Barbara & Caltech Towards Large Language Models as Copilots for Theorem Proving in Lean P Song, K Yang, A Anandkumar [UC Santa Barbara & California Institute of Technology] (2024) https://lnkd.in/eQA7bWGX - The goal of this work is to explore LLMs as copilots to assist humans in proving theorems, rather than proving theorems fully autonomously without human intervention. - Existing LLM-based provers struggle with truly novel or challenging theorems. Human insights can be critical for such theorems. - The authors introduce Lean Copilot, a framework for running LLM inference in Lean. It allows building various LLM-based proof automation tools like tactic suggestion, proof search, and premise selection. - Lean Copilot enables fast inference of LLMs locally or remotely. It provides general text-to-text generation and text-to-vector encoding interfaces. - Experiments show the effectiveness of tools built using Lean Copilot in assisting humans and automating proofs compared to existing proof automation in Lean. - The work is open-sourced to facilitate research on human-AI collaboration for theorem proving.

47

2 Comments

A couple of recent college graduates from India against AI finetuning giants like Anyscale and Run:ai? I met Arko C at a NASSCOM event last October. He was building Xylem AI (YC S24) , a platform that allows you to train and deploy LLMs in production. So naturally most Indian VCs rejected him. Because how could a team of three recent college grads compete with the likes SV startups like Anyscale that have raised hundreds of millions of dollars, have teams of experienced AI researchers and a stash of H100 GPUs? Arko hustled and closed two Fortune 500 companies for paid contracts. I personally know unicorns that haven't managed to do this. He then got into YC in the last round in May. There's a lot to still be proved out, but Arko exemplifies how valuable grit is in building startups. I’m doing a webinar with him tomorrow on challenges of using LLMs in production. Link to register in the comments. Disclaimer: I'm a small investor in Xylem.

148

4 Comments

Hallucinations in LLMs are inevitable, no seriously! Josh Fritzsche and I have worked hard at Voze to find a way to tune, prompt and even plead 🙏 with LLMs to prevent them from the occasional 🍄 hallucination. It seemed like the better we have done the harder the edge cases have been to solve and eventually we realized that they tend to frequently happen when LLMs predict or make anticipations on "nothing". I remember at one point nearly going insane with frustration because when LLMs had a hallucination, they were EPIC! 😱 Then we pushed through it and benefited for the effort! How did we solve this common problem? We didn't (Not completely), we got better at identifying the issues as they happened. Once we identified in the data we were able to start looking for patterns. The same catalysts that cause it to happen once if left unchanged caused it over and over again. Our solution is astoundingly simple. Once you know that it's likely to happen..... let it! Accept it will 🍄, but also be preemptive in not focusing on it and use alternative data sources, generators or providers. Change the conditions that cause it but continue to monitor it as it happens. As Dan Caffee is fond of saying to me on a regular bases. "Dusty we need the system to be adaptive and provide mechanisms to learn from it's past." Well he is right! All we needed was data and the people that contribute to it's quality. Huge props 👏 to strong support from Kathy and Janelles audit teams. We were armed with a knowledge and a desire to learn from past mistakes. Armed with confidence that we could correct these problems we could focus on the data to prevent their impact going forward. When doing millions of LLM (Generative AI) calls regularly Josh and I have learned that hallucinations will happen. We also know exactly how we will handle and prevent it going forward! Failure is not doing anything and accepting defeat.

4

2 Comments

Ai : Hunanity's greatest creation for Good and Bad. To secure ourself #zerotrustai In the journey, MS brings up VASA-1 will AI model which essentially spins a selfie image and turn it into a talking clip of you. All you have to do is upload a photo along with a voice note and let the AI model do the talking for you.

2

Going for a hat trick! Next AI/ML Lab cohort kicks off September 28th https://lnkd.in/gBhkYij9 Our AI/ML Lab for Engineering and Science is hitting its 3rd iteration! With Lab 3.0, we are adding to an already packed content curriculum to bring further use cases and case studies of AI in the science and engineering community. Bringing in lessons and hands-on practice in Python and PyTorch, this 40-hr live Lab brings lessons from the field and how you and your organization can learn from best practices in data science, ML, generative AI, NLP, distributed computing and deployment of AI. Caltech Center for Technology and Management Education (Caltech CTME) #ai #ml #datascience #python #pytorch #genai

22

Microsoft researchers Sebastien Bubeck and Ronen Eldan have developed #PHI-3, a groundbreaking family of small yet powerful language models. Designed for efficiency and high performance, PHI-3 models can be deployed on everyday devices, making advanced AI more accessible than ever. #PHI-3 models offer significant improvements in speed and efficiency, enabling AI integration into more consumer devices and enhancing user experiences. Compared to larger models, PHI-3 demonstrates competitive capabilities while being more resource-efficient. The future looks bright for PHI-3, with ongoing research promising even more powerful capabilities for a broader range of devices. This innovation is set to reshape the AI landscape! More in the Turing Post - PHI-3 Article: https://lnkd.in/dX4XXFMC #AI #MachineLearning #LanguageModels #Microsoft #Innovation #Tech #ArtificialIntelligence #PHI3 #SebastienBubeck #RonenEldan

5

1 Comment

Quite good article on impact on future of AI by major decision of US and UAE collaboration in AI space! What the future holds? A lot of new opportunities for manufacturing, engineering and ultimately innovating faster! #us #uae #abudhabi #ai

15

1 Comment

NVIDIA NIM™ — inference microservices that provide models as optimized containers — to deploy on clouds, data centers or workstations, giving them the ability to easily build generative AI applications for copilots, chatbots and more, in minutes rather than weeks. Available to download today! #TDSYNNEX #NVIDIA #DESTINATIONAI

14

1 Comment

Integrating LLMs with tools is essential for creating novel and valuable products that are AI-enabled. I noticed a lot of repetitive code for each new project so I’ve created #AutoTools, a framework that takes any SDK in python and allows it to be used by AI as a tool. Easily control the AI’s access using whatever access controls exist within the SDK’s authentication system. You can also remove any functions that you don’t want to give access through access and deny lists. Importantly, this doesn’t magically give AI arbitrary code capabilities — the AI can only provide inputs for functions you’ve already written, tested, and used extensively within your existing code base. Give it a whirl and let me know what you think! #ai #agents #langchain #python #sdk

15

2 Comments

Totally agree with you, Anthony. The name of the game is to get the right context as tight as possible to the question or task on hand. Graphs are one good way to do this, but you should use all the tools in your belt. With our solution, we use graphs, tables, key value pairs, geospatial information, regular search, vectors, and so on. When you combine all these things, the probability of getting the right context as much higher.

5

Meta’s artificial intelligence chief, Yann LeCun, has said that large language models (LLM) such as #ChatGPT would never achieve the ability to reason and plan like humans. He also said that depending on LLMs to reach human-level intelligence is not right as these models need the right training data to answer human prompts correctly, and hence they are ‘intrinsically unsafe’. LeCun shared his thoughts with The Financial Times and said these LLMs, “do not understand the physical world, do not have persistent memory, cannot reason in any reasonable definition of the term and cannot plan hierarchically.” #artificialintelligence #largelanguagemodel

4

1 Comment

A Seismic Shift in Drug Discovery⚕ I recently heard a talk from NVIDIA CEO Jensen Huang that made me realize – the tech transformation of pharma over the next decade may be one of the most seismic shifts we will see in any category. It’s an industry ripe for disruption. It takes 10+ years and costs billions of dollars to bring a drug to market – with over 90% of drugs failing in clinical trials. While tech has made other industries more efficient, drug discovery has become slower and more expensive over time. But now, the scale of data and compute in drug discovery is reaching an inflection point. “We understand biology,” NVIDIA CEO Jensen Huang said in conversation with Recursion CEO Chris Gibson, “but it’s so hard to understand interrelatedness. You need AI. You’re producing intelligence that leads to the production of drugs.” What does this mean? Scientists will no longer have to rely on physical lab experiments alone to unlock new insights – we can train AI models to make predictions that allow scientists to make more informed decisions about experiments to run and targets to go after.. Jensen said he once thought ray tracing – modeling light in 3-D graphics – would take decades to achieve with existing computing power. But they only had to simulate one pixel – AI could guess the other 64. In a similar way, he says “You don’t have to brute force simulate the physics for every single protein and every single cell.” This new category, TechBio, looks to turn drug discovery into a search problem where connected datasets train AI models to make accurate predictions. To achieve this, scalable, relatable, and reliable data are critical. For example, Recursion has spent years building its proprietary dataset – including images of billions of cells – amassing over 50 petabytes of biological, chemical and patient data. It’s one of the largest datasets of its kind in the world. Experiments and data collection happen in a highly automated wet lab nearly round the clock and this data is optimized to train AI models. (For context, the company does a PhD’s worth of lab work in about 15 minutes.)  The second part of the equation to analyze the data is compute power. Jensen said NVIDIA was “the first chip company to build supercomputers for our own use” while Tesla was the “first car company to build supercomputers.” Now, with NVIDIA, Recursion has built the largest supercomputer in pharma. https://lnkd.in/gtv9_3Dq A decade in, with compute, data, and foundation models, Recursion continues to industrialize across the drug discovery and development chain – getting to lead drugs 3x faster than industry average at 2.5X less cost. And this is only the beginning. Watch this space. #technology #innovation #biotech #pharma #drugdiscovery 

600

35 Comments

The Coming Singularity: Are We Ready for ASI? The race towards Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI) is accelerating at an unprecedented pace. A recent paper, "Situational Awareness in AI," underscores this rapid advancement and its far-reaching consequences https://lnkd.in/gVGjf48p This paper serves as a wake-up call. Unlike past technological advances, the shift towards AGI and ASI will be exponential, impacting every sector and potentially outstripping our ability to adapt. It echoes Alvin Toffler's concept of "Future Shock," where rapid change disrupts societal structures. Key Takeaways: >Exponential Growth: The transition from narrow AI to AGI and ASI will be swift. This technology will not only match but surpass human cognitive abilities, transforming our world. >Economic Redefinition: The race is on to create trillion-dollar AI clusters, reshaping global economic landscapes and potentially leading to intense competition for dominance. >Ethical Imperative: Robust ethical frameworks are crucial to mitigate the potential risks of advanced AI, from job displacement to biased decision-making. Aligning with the Future: The critical question is control. Current legislative bodies lack the expertise to regulate this emerging technology. While intellectual property rights receive much attention, we urgently need to establish alignment principles that hold AI labs accountable. Without clear liability for AI misalignment, there's little incentive for prioritizing security and safety. The current state of security in leading AI labs is concerning. While immediate danger may be low, it will swiftly become a major issue. Furthermore, the national security implications are substantial. Advanced AI extends beyond civilian applications, significantly impacting defense and international stability. We must understand and control these technologies to prevent misuse. The paper's "Project" concept highlights the need for coordinated efforts aligning AI development with national security interests, preventing rival advancements that could jeopardize global safety. The Coming Window: The paper suggests a timeline of 2027-2029 for rapid change. This period will likely be more disruptive than the COVID-19 pandemic. We need leaders prepared for a new "Manhattan Project" for ASI. Falling behind by just six months due to hacking or intellectual property theft could have dire consequences. The next three years are crucial for positioning ourselves, with the next three determining democracy's role in the coming AI era. A Hopeful Future: Despite the concerns, AI holds immense potential for positive change, especially in medicine. The journey ahead will be challenging, but with a realistic and proactive approach, we can navigate it successfully. As an AI Realist, I advocate for a balanced view that recognizes both the opportunities and potential risks of this transformative technology.

11

5 Comments

Big Thinking. AI and Computing in the Future. WEC Yann LeCun . https://lnkd.in/gfzTHhSH Abundance. Open Source. Emergence. Assistants. Efficiency. Sustainability. A new human renaissance. #ai #compute #memory #energy #knowledge #evolution

2

Google DeepMind Introduces WARP: A Novel Reinforcement Learning from Human Feedback RLHF Method to Align LLMs and Optimize the KL-Reward Pareto Front of Solutions https://lnkd.in/gTC54E3H

1