Eric Siegel

This paper is right on cue with predictions of Makimoto’s wave. MatMul wave started in 2012 and it should’ve peaked in 2022 which saw the standardization of transformer+attention. Now those basic blocks can be revisited to be more efficient and rethink the fundamental operations such as MatMul and memory requirements. I would expect this trend to continue and innovations of the next 10 years to drive towards customization. Importance of the Compiler in the software tools stack has never been more critical to get sustained performance out of changing hardware. https://lnkd.in/gT7rtY54

2

3 Comments

The Economist's article “Large language models are getting bigger and better” on April 17, 2024 on the challenges of scaling large language models (LLMs) rightly identifies the limitations of current AI architectures. While the relentless pursuit of larger LLMs suggests a future of limitless progress, a deeper dive reveals significant flaws. The "scaling hypothesis"—that more data and power equals human-like intelligence—misses crucial aspects. LLMs face serious hurdles: the dwindling supply of high-quality data, the inefficiency of scaling, and the simple fact that size doesn't equate to cognitive complexity. Here, neurosymbolic AI presents a promising alternative, blending the data-handling prowess of neural networks with the logical precision of symbolic AI. This combination not only promises enhanced efficiency and structured reasoning but also aligns closer with human cognitive processes. Neurosymbolic AI excels in areas where traditional models falter, offering a path to more sustainable, intelligent systems capable of real-world applications like airplane design and financial risk assessment. It’s time for the AI community to pivot towards technologies that not only scale but think smarter. https://loom.ly/96rDCD0

21

4 Comments

Ray Kurzweil's Bold Predictions: Will AI Transform Our World by 2045? Article: Ray Kurzweil on How AI Will Transform Energy, Manufacturing, and Medicine Summary: For over a decade, I've followed Ray Kurzweil's work. This renowned futurist predicts that around 2045, we'll reach a "singularity"—when technology advances so rapidly that it transforms human life in ways we can't control or reverse. Kurzweil believes this could lead to a merging of human and machine intelligence. In addition, according to Kurzweil, AI will revolutionize energy, manufacturing, and medicine. In this article, he's very optimistic about overcoming regulatory hurdles in medicine. What are your thoughts about his predictions listed in this article? https://lnkd.in/g_-jJ32w #ArtificialIntelligence #RayKurzweil #Singularity #AIFuture #TechInnovation United States Artificial Intelligence Institute United States Data Science Institute

6

Estimating the uncertainty in LLM outputs is vital to deploying Generative AI for high-stakes applications across medicine, law, finance, insurance, and data enrichment/extraction/validation. But strategies like relying on the LLM to evaluate its own output or the token probabilities fail to reliably catch untrustworthy LLM outputs. Today my team is excited to announce the Trustworthy Language Model, which provides a reliable confidence score for every LLM response. Catch incorrect responses and hallucinations by upgrading your LLM into a TLM. By sampling multiple LLM responses and returning the most trustworthy one, our TLM can improve the accuracy of responses from GPT-4 by 10% and from GPT-3.5 (ChatGPT) by 20%. Check out benchmarks of TLM performance here: https://lnkd.in/gkJfXUX8

41

4 Comments

By combining the generation capabilities of large language models (LLMs) with a retrieval component typically using a vector or semantic search, RAG chatbots can provide informative and personalized responses backed by evidence from a supplied corpus of documents. However, the performance of these models hinges on properly processing and indexing the document corpus for efficient retrieval. If the retrieval is a failure, then the chatbot responses will be incoherent to the users. In this article, we’ll explore the key steps involved in preparing documents for RAG models.

2

MIT’s New AI Model Predicts Human Behavior With Uncanny Accuracy TOPICS:Artificial IntelligenceCSAILMIT By ADAM ZEWE, MASSACHUSETTS INSTITUTE OF TECHNOLOGY APRIL 19, 2024 Abstract Human Behavior Art Concept Researchers at MIT and the University of Washington have developed a new method for modeling the decision-making behaviors of agents, taking into account computational constraints. This model, which can predict future actions from past behavior, aims to improve AI systems’ collaboration with humans by understanding and adapting to human irrationalities and decision-making processes. Credit: SciTechDaily.com A new technique can be used to predict the actions of human or AI agents who behave suboptimally while working toward unknown goals.

1

#RAG is a popular paradigm in #genai applications. Yet, the interplay between how #llms and #slms use the information provided via a retrieval system and its intrinsic knowledge is poorly understood. This paper sheds some insights into the use of these dual sources of knowledge and joins a short list of mechanistic interpretability work that is critically needed to advance the science of #ai applications.

45

3 Comments

How to Create Powerful AI Representations by Combining Multimodal Information #AI #AIio #BigData #ML #NLU #Futureofwork http://ow.ly/vx8u30sEUKA

6

So glad that this important paper by David Gray Widder is at #FAccT2024! Draws upon 75 interviews with technologists & activists to analyze how value-laden notions of objectivity, quantification, and codification are used to build epistemic power structures in AI ethics practices. These practices legitimize the perspectives and lived experiences of some stakeholders while marginalizing others. Concludes by imploring a turn to "humble technical practices": "I ask those practicing “AI,” and other related quantified sciences widely practiced as value neutral, to help level epistemic power differentials by underscoring the partiality of their work, constructing what I call humble technical practices, which specifically seek to make their epistemic limits clear." Read the full paper here: https://lnkd.in/emxDF2XW #ai #aiethics #techethics #techpolicy #aipolicy #aigovernance

51

6 Comments

One of the emergent capabilities in LLMs is Theory of Mind (ToM) which is a psychological framework used to understand how humans communicate, each forming their own state machine of the other’s state of mind. The idea that AI could really exhibit ToM has been highly debated ever since Michal Kosinski wrote about it. This new study in Nature Portfolio digs in deeper, teasing apart various elements and evaluating LLMs against human baselines. Reading this study - including differences between the models’ behavior - I am wondering if LLMs may appear as “humans” from a different culture, and if that metaphor “normalizes” human machine interaction vis-à-vis human perception of the machine. #llm #AI #openai #llama

112

15 Comments

Knowsaic clients have an advantage because we enrich your content via semantic modeling and human curation to equip your knowledge domain for AI competitive advantage. #contentcuration #ontologies #metadata #humanintheloop #hitl #knowledgegraphs #complexity #semanticweb #promptengineering Elizabeth McLean, MSLS CKS

4

In conjunction with Biden’s AI Executive Order from last fall, the President's Council of Advisors on Science and Technology (PCAST) just released a study and report exploring the implications of AI for scientific research. A key focus is on how a well-designed, equitably shared, and responsibly used data and AI infrastructure could empower scientists and others to tackle some of our greatest societal and global challenges. Executive summary is below.  Special thanks to Eric Horvitz, Chief Scientific Officer at Microsoft for his work on this committee. #ai #responsibleAI #aiinresearch

75

3 Comments

🚨BREAKING BRAINS: the lobotomizing of large language models and the paradox of control. https://lnkd.in/eBtUfM3G #LLMs #AI #AGI #GPT4 #GPT5 #Claude3

1

I remember learning category theory in undergrad and wondering: when would this ever be useful? Turns out it's very useful for program synthesis and software modeling -- thanks Kristopher for the great talk!

9

🚀 Coming to you from New York City! 🌆 I just watched the interview with Dr. Simon Prince on Machine Learning Street Talk, and I'm thrilled to get his new book, 'Understanding Deep Learning.' 📚 Dr. Prince's point about Large Language Models (LLMs) is crucial. LLMs can't have internal consistency. They've learned everything from the internet, which means they believe the earth is both flat and round in different statistical proportions. 🌍 While they hopefully conclude it's mostly round, somewhere in their data is the notion of a flat earth. When humans decide to believe something, we rely on an internal, self-consistent model of the world. LLMs, however, lack this kind of model. They operate on something much more complex and difficult to analyze. 🤯 Understanding this distinction is key to navigating their challenges and potential🌟 #Pangeon #ThePangeon #AI #AGI #OpenAI #FutureTech #GPT #ChatGPT #MachineLearning #DeepLearning #TechInnovation #AIResearch #StartupJourney

11

4 Comments