Abhishek Sharma

The work that Anthropic is doing to unravel the black-box nature of #LLMs is brilliant, foundational, and highly commendable. They are addressing a key challenge in any neural network -- explainability. Not only can this research lead to safer and more-accurate models, but the ability to explain WHY a model generated a specific response is critical in many regulated industries. Bravo!

19

The Tech sector has amassed a significant, above benchmark position in the sector (5 times larger than the next largest holdings in financials). Given the large size of the overweight and spectacular performance we were not surprised to see selling ahead of the previous earnings season. The selling was reversed as #IT companies exceeded high analysts’ expectations. Now we see minimal selling ahead of the next set of results. #STTInsights

2

Perplexity's answer to my question "Actor Scarlett johanssan and Anil Kapoor are included in Times 100 AI list but why not Elon Musk? Answer: Scarlett Johansson and Anil Kapoor have been recognized in TIME's 100 Most Influential People in AI list for their advocacy against unauthorized use of their likenesses in artificial intelligence applications. Their inclusion highlights the growing concerns regarding the ethical implications of AI technology, particularly in relation to deepfakes and the protection of personal identity. Elon Musk, despite his high profile in the tech industry and contributions to AI through companies like OpenAI, is not included in this list. The criteria for inclusion appear to focus on individuals actively advocating for ethical standards and protections against AI misuse, rather than merely those who have made advancements in AI technology. Thus, while Musk is a significant figure in AI development, his absence may reflect a lack of direct involvement in the ongoing discourse about the ethical implications of AI and the protection of personal rights, which is the central theme of this year's TIME 100 AI list 🤨

13

I came across this study from McKinsey that states that AI adoption is surging to a record high, and Gen AI is performing well, but data barriers and risk need to be addressed (maybe not such a big surprise). What I like in this survey is the message that AI and Gen AI have begun to create real business value and don't only exist as garage projects in an IT organisation; I notice that Risk, Legal, and Compliance reports increased revenue and that HR reports the most significant cost decrease. I'm realistic, but this is a good way to start the value conversation. That is a strong signal that business adoption is moving in the right direction. Unsurprisingly, the industry technology is most likely to invest in AI. I think financial services are forced to adopt AI to be cost-effective and comply with regulations. Interestingly, energy and materials are more prominent than media. Given the insights from the McKinsey study, I'm keen to explore the AI landscape in the Nordics. Localised research could shed light on any potential lagging of AI adoption in our region.

8

Dealing with difficult stakeholders? For those of us responsible for managing portfolios, programs, projects or leading teams, there are a core set of skills we need to develop in order to achieve success. Manage on-time product launches, handle delivery, people, conflicts effectively and so on. For doing so, developing relationships across the organization and beyond is essential. One vital skill is the ability to influence. Regardless of the role, entrepreneurs, project managers, leaders are constantly faced with situations where they need to influence team members, bosses, customers, partners, investors and other stakeholders. And sometimes stakeholders can be difficult. One of the reason, I find stakeholders are difficult is they do not feel heard. Their needs are not listened to. And if they think you're not listening, the situation gets worse. The trick is to slow down and pay your full attention. Actively listen. You don't have to agree with them but appreciate their perspective. Making them feel heard and valued will help you regain control of the situation. Dealing with difficult stakeholders is inevitable. Let's get better at it. Recall some challenging interactions you recently had and reflect on your listening. How might have the other person perceived your listening. What did you do (or not do) that might have hurt or benefitted the situation? #listening #difficultstakeholders #highimpactleaders #influence #emotionalintelligence

21

32 Comments

⬇️ Organizations are increasingly turning to artificial intelligence (AI) to improve mainframe operations, enhance transactional workloads, accelerate modernization and close skill gaps. Revolutionizing how IBM Z® users engage with and manage the mainframes is what the IBM watsonx™ Assistant for Z is all about.

11

Last week I was fortunate to speak on behalf of Capital One at two timely conferences at the intersection of AI and finance, the Evident AI Symposium and the FinRegLab AI Symposium. In each forum it was clear that the pace of AI advances in financial services is moving rapidly. At the same time, the industry remains in a test-and-learn phase with a focus on thoughtful, risk-centered and well-managed AI development. I shared my strong belief that one of the noble aims of AI is to transfer cognitively burdensome tasks from the human to the system and deliver enhanced experiences for employees and customers alike. Achieving this in a durable way, at scale, requires a shift in focus from siloed use cases to an emphasis on the far-reaching technical and cultural underpinnings that make organizations AI-ready. This includes platformization, testing and benchmarking, monitoring and observability, and instrumentation to continually refine and improve AI post-deployment to name a few. All of this in turn enables collective learning and continual improvement for the long-term. Some of my thoughts here were captured on Bloomberg during the #EvidentAISymposium: https://lnkd.in/g8Ba5cPB Andrew Buemi Milind Naphade Abhijit Bose Aparna Sinha Maulin Patel John Morgan Sambit Sahu Giri Iyengar Stephen Rawls Stephen Potter Jennifer Borden Ilene Eng Rob Pulciani Micaela Klein Daben Liu Bayan Bruss Robert Alexander

331

17 Comments

🌟 New from #NVIDIAResearch: Weight-Decomposed Low-Rank Adaptation (DoRA). A groundbreaking advancement in fine-tuning technology that's set to revolutionize how we optimize pretrained models without increasing inference costs. 👀 https://dy.si/HXvUD9

8

Talent knows no boundaries, but opportunity should. 🌟 Meet Kartik Kansal, an ISRO scientist with muscular dystrophy who cleared the prestigious UPSC exam four times, yet wasn't allocated a service. My daughter and I are passionate advocates for Duchenne muscular dystrophy (DMD) in the US. Kartik's story resonates deeply, highlighting a crucial question: How can our collective efforts extend the support Let's brainstorm ways to: 1. Raise awareness about Muscular Dystropy diseases in India 2. Advocate for equal opportunities in recruitment processes 3. Support initiatives that bridge the gap between talent and opportunity Together, we can make a difference across borders. Share your ideas below! #EqualOpportunities #MusculardystrophyAwareness #GlobalImpact Anoushka Dhawan

52

"The U.S. sorely needs a coordinated national research strategy", says Marcia McNutt, president of the U.S. National Academy of Sciences. In a first-ever “State of the Science” address, she warned the U.S. was ceding its global scientific leadership to other countries, China in particular. This slippage could make it harder for the U.S. to maintain the strength of its economy and protect its national security. U.S. has a declining share of the most cited science papers. It still spends the most money out of any other country on research and development although China's investment growth rate is double that of the USA. China currently files more patents than the U.S. and hosts more than a quarter of the world’s clinical trials, as compared to only 3% in 2013. The US accounts for nearly 60% of all Nobel Prizes awarded since the end of World War II. This dominance is now being challenged. As federal investment dipped in the 1980s, the scientific landscape grew more complex. Companies and philanthropies began funding and conducting greater portions of research and development. The private sector tends to pursue narrower, applied interests, and to keep its findings to itself. When companies take over entire fields as they did with genetically modified organisms and now artificial intelligence, public’s distrust is increased. Math and science proficiency fell during the 1980s too, with fewer Americans entering the STEM workforce. Of utmost importance, McNutt emphasised, was improving K–12 STEM education. Foreign-born individuals play a key role in maintaining U.S. STEM prowess, accounting for 19% of the nation’s STEM workforce and 43% of its STEM Ph.D. population. However, decline in Chinese students enrolling in US universities in recent years, represents risk to the US's ability to attract top talent. In 2022 Congress passed a law in line with many of McNutt’s recommendations: the CHIPS and Sciences Act. The statute which McNutt called a “good start”, promised not only to revive homegrown semiconductor production for computers but also to invest $170 billion in broad research funding over the next five years, as well as to expand STEM education on the K–12, college and graduate levels. The outcome of the upcoming presidential election may shape the future of federal scientific program as well. During the past four years, the Biden administration has supported modest increases to several scientific agencies’ budgets, including those of the National Institutes of Health and the Environmental Protection Agency. In 2020, the last year of his term, former president Donald Trump proposed widespread slashes, although those were dismissed by Congress. See articles from Scientific American and Tech Times. https://lnkd.in/gZdSNe8M https://lnkd.in/gK866QQp #Science #Geopolitics

37

9 Comments

Where there’s a Will, there’s always a way—this has been my guiding principle throughout a career dedicated to driving value-based leadership, transforming organizations, and embracing the power of AI and innovation to overcome challenges. Whether it’s delivering $200M+ annual ROI at a fortune 500 Fintech through AI/ML adoption, helping found AI Ready RVA to advance AI education and workforce development, or improving customer success by 557% through strategic initiatives, my passion lies at the intersection of strategy, technology, and impact. But success isn’t just about the numbers—it’s about fostering cultures of integrity, accountability, and creativity. I’ve been privileged to lead teams, drive social enterprises, and inspire others to embrace innovation while staying rooted in purpose. As we look ahead to 2025, I’m opening my calendar for keynote engagements and presentations. I’m excited to collaborate with leaders and organizations to discuss topics that matter most: ✅ AI/ML integration and workforce innovation: How to harness AI for industry transformation and sustainable workforce growth. ✅ Leadership and organizational transformation: Creating cultures of resilience and collaboration in dynamic environments. ✅ Value-driven social impact: Building initiatives that align business success with positive societal change. ✅ Overcoming perceived disadvantages to truly find your superpower and accelerate your impact. If you’re looking for a speaker to energize your event, let’s connect. Together, we can explore actionable insights and strategies to shape a more innovative and inclusive future. Have you attended a talk or workshop I’ve presented? Drop a comment below to share your experience—I’d love to hear your thoughts! Let’s break boundaries and shape the future together.

29

1 Comment

Now that we know there is only so much we can do with #PromptEngineering with these #Large and #Larger #LLMs here comes the phase of rationalization. Going by the literature some of these are much simpler and would turn out to be much cheaper than the other lazy alternatives that we are getting comfortable with. I don't want to undermine the effort for #DataPreparation and #Evaluation required to ensure the results are worth the effort. Sure, it means a lot of hard work! - Improve #Retrieval: Use contrastive loss to train a dual encoder embedding model for both questions and answers - Improve #Generation: Finetuning with use-case specific instructions and responses #weekend #learnings - https://lnkd.in/g93svug3 - https://lnkd.in/g3wYxJGR

24

Most common recurring issues between Production support and development teams Using an app, John reports to Rajesh that he is unable to access a feature of the app. John is notified by his production support analyst colleague Rajesh that he is investigating the matter. Rajesh begins investigating the problem's underlying cause. After fifteen minutes John: Are you looking at the issue? How much more time is required? Rajiv: I'm looking into this for you, but I need more time to fully investigate. When Rajesh examines the logs, he finds no problems and raises the matter with a member of the development team. Rajesh: Hello, Ramesh John can't see a feature on the application. I went over the logs and didn't notice anything odd. Ramesh: Even though I addressed this topic during our handover conversation, you continue to contact me about it. It is a time and effort waster. Ramesh explains the underlying issue to Rajesh and how to allow access to John was given the feature. Conflicts and frustrations arise between the development and production support teams as a result. Rajesh is frustrated since he wasn't involved in the early stages of development and finds it difficult to understand the feature list of the new services. He expresses dissatisfaction about the development team's rapid release schedule and the support team's inability to monitor every modification that is made before it is put into production. A production team's worst nightmare is trying to comprehend these services and identify the underlying causes of the problems. Ramesh is upset. I work as a developer, yet I spend a lot of time and energy explaining same problems to the support team again. During the handover call with the developer, they don't pay close attention, thus we have to spend time answering their questions. Because of the longer response times, all of these could cause dissatisfaction from customers point of view.Attrition may result from disagreements and frustrations between the development and support teams. Collaboratively addressing such situations is crucial. In software development, documentation is essential. It not only helps to resolve the challenges mentioned above related to knowledge transfers, but it also plays a big part in assisting with the onboarding of new team members and reducing risk by monitoring possible problems. Making a creative recording of the handover sessions is one such option. Support team should ensure they have gone through it a few times before it is actually live. Take the time up front to produce documentation, record your sessions and dont fall into the trap that docummentaiton can be done later.

12

Struggling to find use cases for GenAI in Financial services? Here are some that may help. Some maybe hype, others could add value. You just need to begin somewhere. #generativeai #datascience #machinelearning #fintech #payments #innovation

34

1 Comment

Grateful for the enlightening discussions on Data Lake vs Data Warehouse during our panel session at #DES2024 by AIM Events! Exciting times are ahead as AI and ML advancements promise to revolutionize storage practices.  Here are some key points and future predictions from our talk. 1️⃣ We talked about how to achieve incremental ROI with both storage approaches, considering specific goals. The rise of ML and AI will dramatically change data storage practices. With data volume, user base, and use cases expected to grow, big changes are coming. 2️⃣ AI capabilities are changing data engineering, making data preparation, pipeline management, and data movement faster and more automated. Future strategies need to focus on cost-effectiveness, flexibility, security, and governance. Hybrid approaches like the Data Lakehouse, which combine the best of Data Warehouses and Data Lakes, are becoming popular. 3️⃣ Edge computing will increase the need for better compute power in storage management. Real-time analytics will continue to improve storage practices. AI and ML integration will make data modeling and discovery easier. Additionally, data virtualization and federated queries will allow seamless data access across different sources without needing to move data physically. As we navigate this evolving landscape, it's important to stay updated on these changes and embrace innovative solutions for success. 🚀 How do you see AI and ML influencing your data management strategies? Let me know in the comment section below! #DataManagement #AI #ML #DataStorage #Innovation #DES24 #DES #AIM #ThinkAIThinkAIM #Datewarehouse #DataLake #DataLakehouse #GENAI

15

A Fully Automated Learning Algorithm Applicable to Board Game Playing Programs By Shree Ganesh Keith Ragoonaden, 1995. This UK SERC-funded research project was initiated in 1987 with the idea of implementing an unsupervised learning algorithm that would self-correct its assessment (if deemed necessary) of its response to a given situation based on the opponent's counter-response. In other words, the algorithm would self-learn from the knowledge of its opponent without any human intervention/supervision. A semantic network of multi-layered interconnected knowledge representation of the environment in which the self-learning algorithm would evolve was imperative for its application. The game of GO was the environment used to demonstrate how this unsupervised learning algorithm would evolve with time. The document describing this research work together with the fully implemented unsupervised learning algorithm was completed in 1994.

2

In a recent Stanford Online talk, Anima Anandkumar, Bren Professor of Computing at Caltech, explores the potential of generative AI for science. The session is filled with good questions and interesting answers that provide valuable insights. Watch the talk here: https://lnkd.in/gGHBubex

7

To me, one key difference in the "intelligence" of humans and LLMs of today is that: humans can update their world-representation (parameter weights) during inference, whereas for LLMs, once training + fine-tuning is complete, the world-representation is frozen. An extreme example of this: Einstein running thought experiments (twin paradox, elevator free-fall) in his head as part of a reasoning task, and using the outputs of these thought experiments to change his world-representation of concepts like space-time and gravity in his head (model weights updated). He then goes on to use this modified world-representation as a basis for further reasoning. Obviously, it is not just Einstein; this is the way we humans often reason about things. And learning happens not just through thought experiments: we use information gained during actual experiments and experiences during the reasoning task to update our knowledge representation. I think of this less as meta-cognition and more as dynamic training + inference. How can such a dynamic reasoning/thinking process be mimicked using an LLM? > Can a feedback loop driven RAG + N-shot achieve something similar? > Will future AI models allow fine-tuning on the go, during inference? > Or does the sheer capacity of the LLMs to capture rich semantic relationships and compute against it outweigh the loss from the inability to perform human-like learning during reasoning? Tagging some of my AI expert friends for fun 😉 Lavanya Tekumalla Mohit Kumar Kaustubh Joshi Santosh Waddi Jingchen Wu #ai #generativeAI #llm

53

7 Comments