Swaroop Kallakuri

https://lnkd.in/gWUpZECK Mixture of Agents (MoA) is an approach to harness the collective strengths of multiple LLMs to improve state-of-the-art quality. According to Together AI, it achieves a score of 65.1% on AlpacaEval 2.0, surpassing prior leader GPT-4o (57.5%).

5

!!AI Adoption!! In a recent study, over 40% of Japanese firms revealed they lack AI adoption plans, highlighting a significant challenge in harnessing the full potential of artificial intelligence. I wonder how is Croatia doing comparatively? However, both countries face similar obstacles: workforce readiness and integration of AI in traditional industries. Embracing AI can drive innovation and economic growth, making it imperative for businesses worldwide to develop comprehensive AI strategies. #AI #Innovation #BusinessStrategy #Croatia #Japan #FutureOfWork https://lnkd.in/dXSjP5hJ

27

3 Comments

One question I got on the LLM Evaluation framework SECTOR I launched yesterday, is if it will work on summaries which are highly creative in nature also context can be way larger than response Below is sample example with a summary and how output is broken down to individual sentences and compared with context Try it out. Link to the repo - https://lnkd.in/gUBSztpE Again this is early stage and collectively we can make it better by sharing feedback and suggestions #artificialintelligence #generativeai

39

3 Comments

Rethinking RAG: The Hidden Costs and Optimization Challenges of Retrieval-Augmented Generation As AI continues to evolve, Retrieval-Augmented Generation (RAG) has become a cornerstone of many production systems. But are we overlooking critical aspects of its implementation? Here are some thought-provoking insights: The Embedding Paradox: • Cheap to compute, expensive to store • 1 billion embeddings can require 5.5TB of storage • Monthly storage costs can reach $14,000+ for high-dimensional embeddings The Quantization Revolution: • Binary quantization can reduce storage costs by 32x • Only 5% performance reduction in some cases • Potential for 25x speed increase in retrieval Model Size Matters: • Larger embedding models are more resilient to aggressive quantization • Parallels the behavior of quantized LLMs The Open-Source Advantage: • Vector stores like Qdrant now support scalar and binary quantization • Up to 99% accuracy preservation with 4x compression for scalar quantization The Tradeoff Triangle: • Balancing accuracy, speed, and storage costs • No one-size-fits-all solution Key Questions: • Are we overengineering our RAG systems? • Could simpler, quantized models outperform complex ones in production? • How will advancements in embedding compression reshape AI infrastructure costs? As we push the boundaries of AI, it's crucial to look beyond raw performance and consider the full spectrum of tradeoffs in our systems. What's your take on the future of RAG optimization? s/o to Prompt Engineering https://lnkd.in/d5uKryzQ

12

📢 Who Invented the Term QoE (Quality of End User Experience)? 📢 Ever wondered who coined the term QoE (Quality of End User Experience) and pioneered solutions to measure it, relating it to QoS (Quality of Service) for the first time nearly 20 years ago? Discover the innovator behind this groundbreaking concept and the impact it has had on the telecommunications industry. Watch the video below to learn more! 🎥👇 #QoE #drdavidsoldani https://lnkd.in/gcvF8Tux

11

1 Comment

The current trend in LLMs is small models. This easy-to-understand article discusses the various ways to achieve model compression and provides key references.

5

With the advent of LLM’s we now have choices on what modeling techniques we use to build and deploy analytical models #AgileData Do we use the tried and true analytical models that have been around for decades or do we move to the use the newer LLM capabilities? Shane Glass is publishing a series of article’s that compares using purpose built analytical models vs the Gemini LLM. Well worth a read, even if your not a Google Cloud user (which of course you should be). https://lnkd.in/gWgcryM6

1

Texas AG’s Landmark AI Settlement: A Wake-Up Call for Health Tech & AI Companies A recent settlement between the Texas AG and Pieces Technologies highlights the need for transparency and accuracy in AI products, especially in healthcare. The case stresses that misleading claims about AI performance can lead to regulatory action under existing consumer protection laws. Critical takeaways for AI companies: Ensure transparent and honest product marketing. Disclose risks, limitations, and proper use. Adopt robust AI governance frameworks. This serves as a critical reminder for both developers and healthcare organizations. https://lnkd.in/gv7qhpJu #AI #Healthcare #Compliance #Ethics

9

1 Comment

OpenAI recently published their Model Spec, a document that describes rules and objectives for the behavior of their GPT models. The spec is intended for use by data labelers and AI researchers when creating data for fine-tuning the models. #AI #GPT #Data

1

1 Comment

Insightful look into how modern infrastructure and AI are redefining data management.

30

1 Comment

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

Big technology companies, including Microsoft, Alphabet, Meta, and Amazon, are set to spend over $1 trillion in coming years on AI investments- Some investors have raised concerns that big tech's AI-related capital expenditures could disappoint, given long development timelines and limitations on the technology's capabilities. A beautiful framework to understand why these concerns have risen and for understanding the economic and societal impact of technological revolutions. In The Technological Revolution and Financial Capital, Carlota Perez identifies five major revolutions in history: Industrial Revolution, Age of Steam and Railways, Age of Steel and Heavy Engineering, Age of Oil and Automobiles, Information and Telecommunications Revolution (ongoing). Each revolution follows a similar cyclical pattern, which she divides into two key phases: the Installation Period and the Deployment Period. The Installation Period is characterized by the introduction and rapid development of new technologies. During this phase, financial capital plays a critical role in driving innovation and infrastructure expansion. Investors and entrepreneurs, eager to capitalize on the potential of emerging technologies, inject significant funds into new ventures, leading to rapid economic growth and the proliferation of new industries. However, this phase is also marked by speculative excess. Financial markets often become overly enthusiastic about the potential of new technologies, leading to the formation of bubbles. These bubbles are fueled by unrealistic expectations and excessive risk-taking, which eventually result in a financial crisis when the bubble bursts. Despite its destructive effects, this crisis serves as a necessary turning point, signaling the end of the Installation Period and setting the stage for the next phase. Deployment Period Following the financial crisis, the economy enters the Deployment Period, where the focus shifts from speculation to sustainable growth. In this phase, the new technologies developed during the Installation Period are widely adopted and integrated into the broader economy. This leads to a "golden age" of stable economic growth, increased productivity, and improved living standards. The benefits of the technological revolution are more broadly distributed, resulting in greater social and economic inclusivity. The Deployment Period represents the maturation of the technological revolution. The initial excitement and volatility of the Installation Period give way to a more stable environment where the new technologies are embedded in everyday life, transforming industries, creating new markets, and reshaping social structures. Reading in context of the above, my take - AI is still in the installation phase. We still have a few years to go before we reach the Turning Point for the Digital Revolution and then we could see the golden age of deployment over the next few decades. #AI #Investment #Capital #Tech

32

2 Comments

🔝🚨Anomalous emergent behaviors in AI: Hermes 3, has "existential crises" when you use right inputs, a black prompt or ask "Who are you?" or "What do you see?". Here you are some examples of unexpected emergent behaviors at scala. Emergent behaviors are really relevant in AI because thosw appears without especific training and many times only a few times even in the LLM has been trained with the same datasets and procedures. #AI #AGI #LLM #research #artificialIntelligence https://lnkd.in/dyDUEFs4

3

1 Comment

For LLM Optimization, a multi-dimensional approach is key. Here are some possible options to enhance Optimization: - Prompt engineering or Streamline Instructions to LLM for more specificity and control. - Implement RAG or provide supportive knowledge and context to LLM related to the Prompt. - Consider Fine-tuning or training LLM in specialized domains like legal, finance, insurance, or retail. Adopting a diverse strategy can lead to improved outcomes in LLM Optimization. #Strategy #Optimization #Training #Specialization Rewritten by LinkedIn AI LLM

12

💥💥💥 Unintentional Unalignment: Likelihood Displacement in Direct Preference Optimization Noam Razin, Sadhika Malladi, Adithya Bhaskar, Danqi Chen, Sanjeev Arora, Boris Hanin Abstract Direct Preference Optimization (DPO) and its variants are increasingly used for aligning language models with human preferences. Although these methods are designed to teach a model to generate preferred responses more frequently relative to dispreferred responses, prior work has observed that the likelihood of preferred responses often decreases during training. The current work sheds light on the causes and implications of this counter-intuitive phenomenon, which we term likelihood displacement. We demonstrate that likelihood displacement can be catastrophic, shifting probability mass from preferred responses to responses with an opposite meaning. As a simple example, training a model to prefer No over Never can sharply increase the probability of Yes. Moreover, when aligning the model to refuse unsafe prompts, we show that such displacement can unintentionally lead to unalignment, by shifting probability mass from preferred refusal responses to harmful responses (e.g., reducing the refusal rate of Llama-3-8B-Instruct from 74.4% to 33.4%). We theoretically characterize that likelihood displacement is driven by preferences that induce similar embeddings, as measured by a centered hidden embedding similarity (CHES) score. Empirically, the CHES score enables identifying which training samples contribute most to likelihood displacement in a given dataset. Filtering out these samples effectively mitigated unintentional unalignment in our experiments. More broadly, our results highlight the importance of curating data with sufficiently distinct preferences, for which we believe the CHES score may prove valuable. 👉 https://lnkd.in/gxh4-S-5 #machinelearning

3

Read this article to explore several practical use cases of generative AI in data analytics, such as with data augmentation and data processing. #AI #datascience #artificialintelligence https://hubs.li/Q02DT8td0

5

Value capture has always fluctuated between applications and infrastructure. What will we see getting commiditized first in the ML world - compute infra or applications? Read my thoughts on this here : https://lnkd.in/gD3N7dHX #strategy #hyperscale #scaleups #bizgrowth #foundersadvice #bizmodel #innovation #fintech #insurtech #tokenization #productmanagement #web3  #TMinsights

5

Excellent report from General Dynamics Information Technology John P. Curran regarding adoption and trends of AI in the Govt. I used Meta to generate summary below and included link to original article. Main AI use cases are chat and creating/updating computer code. Note also the 23% failure rate of AI projects due to user expectation misalignment. This can be overcome with thorough protoyping and establishing use case metrics. - Successful AI project completion timelines are narrowing, with an average of 14 months from pilot to production. - Future demand for AI projects is building around chat and coding-related functions. - 54% of surveyed AI experts said projects currently under development involve improving computer code. - 59% of projects involve "research and understanding", which may relate to creating chat functions. - 23% of AI projects fail due to misalignment between AI solutions and user expectations. - Partnership efforts are required for successful AI projects. Data diversity is crucial, and incorporating diverse data sources can lead to better AI outcomes. Trends: - AI adoption is increasing across the Federal government, with a growing focus on chat and coding-related functions. - Agencies are leveraging cloud-native services, tooling, and other capabilities to accelerate AI project development. - There is a growing interest in translating older code to newer languages, such as Python or cloud-based analytic tools. - Domain-specific chat functionality is becoming a priority, with many agencies planning to implement chat services in the next two years. Statistics: - 325 AI experts surveyed across the Federal government - 14 months: average time from pilot to production for AI projects - 54%: percentage of projects currently under development that involve improving computer code - 59%: percentage of projects that involve "research and understanding" - 23%: percentage of AI projects that fail due to misalignment between AI solutions and user expectations. Bill Casola John Grim Leslie Anne McIndoe Kim Giffin Michael S. Freeman III Tobias Quaranta https://lnkd.in/eqvmAVFG

3