Gen AI for Business #10

Hello, and welcome to the 10th edition of our Generative AI for Business newsletter! We're on the brink of hitting 8,000 subscribers, and we couldn't be more excited. Your support, comments, and interactions have been the fuel that keeps us going, and we're incredibly grateful for each and every one of you. Seriously, we might just throw a virtual pizza party when we hit that milestone. 🍕

This week, we've got a fantastic lineup of stories for you. Salesforce has launched a game-changing AI benchmark for CRM—think of it as a dating app for businesses to find their perfect AI match. Target's new AI tool is making store operations smoother than ever, turning staff into inventory ninjas and customer service superheroes.

Meanwhile, McDonald's is shaking things up by dropping IBM's AI order tech in favor of new drive-thru innovations. Maybe next time you order a Big Mac, the process will be faster and smoother than ever. 🍔🍟

Verizon is using AI to boost customer loyalty by matching calls with the perfect agent. It's like a matchmaking service, but for customer service calls. And don't miss out on AWS's whopping $230 million investment in AI startups and the heated LLM price war in China that's making AI as common as your morning coffee.

We are showcasing some amazing Women Leading in AI, sharing prompts from our subscribers, and much more! 

Thank you for being part of this incredible journey. 

We hope you enjoy this month's insights and find the inspiration to drive your own AI innovations. Here's to many more milestones together! Happy reading!

If you enjoyed this newsletter, please leave a like or a comment and share! Knowledge is power. 

Thank you,

Eugina

News about models and everything related to them

Large language models (LLMs) have seen significant advancements in 2024, with new releases and research enhancing their capabilities. These models are becoming more efficient and accurate, benefiting from improved training techniques and larger datasets. Key developments include better alignment methods, enabling models to generate more relevant and contextually appropriate outputs. Additionally, open-source initiatives are making these advancements more accessible to the broader AI community, fostering innovation and collaboration.

In this section, we cover: Anthropic has introduced Claude 3.5 Sonnet, a highly efficient AI model available for free on Claude.ai and via API, featuring "Artifacts" for real-time content creation. NVIDIA launched an open-source synthetic data generation pipeline on its NeMo framework to enhance large language models (LLMs). Galileo's Luna model detects AI hallucinations accurately and cost-effectively. The newsletter also addresses the issue of AI hallucinations and shares a personal experience. Self-improving LLM systems like DrEureka and LLM-Squared are highlighted for their iterative enhancements. MAGPIE, developed by the University of Washington and Allen Institute for AI, generates alignment data for LLMs without manual prompts. ARSIC methodology automates satellite image captioning. The environmental impact of LLMs is discussed, noting their carbon footprint is comparable to other technologies. Meta released new AI models for various applications, and the Kolmogorov-Arnold-Moser matrix is introduced for improving LLM accuracy. Various LLM types and improvements in diffusion models are explored. IBM's InstructLab aims to democratize AI development, and MIT's research on visual knowledge in language models highlights multimodal AI potential and challenges.

• Introducing Claude 3.5 Sonnet \ Anthropic Anthropic has launched Claude 3.5 Sonnet, a new AI model that outperforms its predecessors in intelligence, speed, and cost-effectiveness. Available for free on Claude.ai and via API, it excels in reasoning, knowledge, and coding tasks. Claude 3.5 Sonnet also introduces "Artifacts" for real-time content creation and editing. 

The model maintains a strong focus on safety, privacy, and rigorous testing. Future plans include expanding the Claude 3.5 family and developing new features for enhanced user experience. Have you experimented with it yet? It’s getting some rave reviews with headlines like “best AI model yet”, “Anthropic ups the ante in the AI arms race.  Did you know that it understands humor? Here are the ratings: 

• NVIDIA Releases Open Synthetic Data Generation Pipeline for Training LLMs  NVIDIA has released an open-source synthetic data generation pipeline to train large language models (LLMs). This tool, available on NVIDIA's NeMo framework, allows users to create high-quality synthetic datasets to enhance the performance and accuracy of LLMs. The pipeline addresses the need for vast amounts of diverse data in AI training, making it easier for developers to generate and customize datasets for specific applications. This innovation aims to accelerate AI model development and improve AI capabilities across various industries. Read more in the technical report here: Nemotron-4 340B Technical Report  details a synthetic data generation pipeline designed for training large language models (LLMs). It emphasizes the creation of high-quality, diverse synthetic datasets to improve LLM performance. Key features include the ability to customize data for specific applications and its integration with NVIDIA's NeMo framework, aiming to streamline and enhance AI model development. 

• Galileo Introduces Luna: An Evaluation Foundation Model to Catch Language Model Hallucinations with High Accuracy and Low Cost - MarkTechPost Galileo has introduced Luna, an evaluation foundation model designed to detect language model hallucinations with high accuracy and low cost. Luna uses advanced techniques to identify and correct inaccuracies in AI-generated text, ensuring more reliable outputs. This model aims to improve the quality of AI responses in various applications by minimizing errors and enhancing user trust. Luna represents a significant step forward in the development of more dependable AI systems. My recent ChatGPT fail was when I was trying to get a list of local publications for my husband to send a press release about his new business. It was late at night, and I wanted to do it quickly (which is not always good, as we all know), so I ran a query on ChatGPT and got a table with email addresses for local media. The majority of those emails were hallucinated, not real. So, the AI lied to me. So, the next day, I did my own search and collected the real ones. Visit the section below on Concerns around GenAI to read 32 cases where AI went terribly wrong. Enjoy. 

• And this article explains hallucinations and why models hallucinate: https://www.technologyreview.com/2024/06/18/1093440/what-causes-ai-hallucinate-chatbots/ AI hallucinations, or the generation of incorrect or fictitious information by chatbots, occur because large language models (LLMs) predict words based on statistical likelihood rather than retrieving factual data. This prediction-based process, akin to generating new sequences from scratch, often results in plausible but inaccurate outputs. Improving AI accuracy involves training on more text, using chain-of-thought prompting, and possibly integrating fact-checking mechanisms. However, due to their probabilistic nature, completely eliminating hallucinations remains challenging. 

• Why self-improving LLM systems will be a big trend - TechTalks  Self-improving large language models (LLMs) represent a significant trend due to their ability to autonomously refine their performance through iterative feedback cycles. These models generate hypotheses, test them, and integrate the results to propose better solutions. This process leverages their vast training on diverse data and is scalable, allowing for rapid hypothesis testing and refinement. Examples like DrEureka and LLM-Squared demonstrate superior results in tasks such as robot manipulation and loss function optimization, respectively. Despite requiring well-crafted prompts and verification mechanisms, these systems hold promise for accelerating AI advancements. 

• MAGPIE: A Self-Synthesis Method for Generating Large-Scale Alignment Data by Prompting Aligned LLMs with Nothing - MarkTechPost MAGPIE is a novel method developed by researchers at the University of Washington and Allen Institute for AI that generates large-scale alignment data for large language models (LLMs) without manual prompt engineering or seed questions. 

By leveraging predefined templates, MAGPIE prompts LLMs to autonomously create user queries and corresponding responses, resulting in diverse, high-quality instruction datasets. This automated approach, tested on Llama-3 models, ensures efficient, scalable dataset generation, enhancing LLM performance across various benchmarks by providing extensive, varied instruction data.  

• Towards Automatic Satellite Images Captions Generation Using LLMs: Methodology | HackerNoon  The article on HackerNoon discusses ARSIC, a methodology for generating captions for satellite images using large language models (LLMs) and APIs. This method involves three main steps: analyzing spatial relationships between objects, guiding LLMs with prompts to create captions, and evaluating and selecting the best captions. By automating the caption generation process, ARSIC reduces the need for human annotation, improving accuracy and efficiency in remote sensing image captioning. It explains some prompting strategies as well; take a look at the prompting section in the newsletter to get some additional insights.

• Environmental impact of LLMs – discusses the environmental impact of large language models (LLMs), emphasizing that while their carbon footprint is notable, it is often overestimated compared to other technologies like video streaming and cryptocurrency mining. The operational costs of LLMs are significant, but not disproportionately higher than existing tech infrastructures. The author argues that the real concern lies in the potential industrial-scale deployment of LLMs, which could amplify their environmental footprint, but overall finds that the public's worry might be somewhat exaggerated. 

• Releasing New AI Research Models to Accelerate Innovation at Scale | Meta  Meta has released several new AI research models aimed at accelerating innovation. These include models for image-to-text, text-to-music generation, multi-token prediction, and AI-generated speech detection. The Chameleon model can handle both text and image inputs and outputs. JASCO improves control over AI music generation. AudioSeal is introduced for detecting AI-generated speech. Most models are available under research-only licenses to encourage collaboration and further advancements in AI technology. 

• The Kolmogorov-Arnold-Moser matrix is the basis of new LLMs - TechTalks discusses the Kolmogorov-Arnold-Moser (KAM) matrix as a new basis for large language models (LLMs). Unlike traditional models based on multilayer perceptron (MLP), which use fixed neuron weights, the KAM matrix uses trainable nonlinear activation functions, offering greater accuracy and explainability. This approach allows for better memory retention and reduced need for extensive retraining. The KAM matrix is poised to revolutionize AI by making complex functions manageable and understandable. 

• Large Language Models (LLMs) include various types such as Transformer models like GPT-3, which utilize attention mechanisms for text processing and generation. BERT focuses on understanding the bidirectional context in text. Diffusion models iteratively add and remove noise to generate data. Recurrent Neural Networks (RNNs), including LSTM and GRU, process sequential data. Multimodal models like CLIP integrate text and images. Autoencoders, including Variational Autoencoders (VAEs), are used for data compression and generation, while Generative Adversarial Networks (GANs) consist of a generator and discriminator for creating realistic data. This paper, Simple and Effective Masked Diffusion Language Models explores improving masked diffusion models for language processing. Using an effective training recipe and a simplified objective, these models achieve state-of-the-art performance among diffusion models, approaching autoregressive (AR) model perplexity. The study highlights how these models handle discrete data and proposes techniques to improve their accuracy and efficiency, making them viable for text-generation tasks. This work has significant implications for enhancing the performance and application of diffusion models in natural language processing.

• A practical guide to making your AI chatbot smarter with RAG • The Register provides insights on the practice of stress-testing AI systems to identify vulnerabilities and improve robustness. It emphasizes the importance of adversarial testing, simulating potential threats, and ensuring that AI models can handle real-world challenges. The guide outlines strategies for effective red-teaming, including diverse attack simulations, continuous monitoring, and collaboration with ethical hackers. This proactive approach helps in enhancing AI security, reliability, and trustworthiness. 

• A New Google Study Presents Personal Health Large Language Model (Ph-Llm): A Version Of Gemini Fine-Tuned For Text Understanding Numerical Time-Series Personal Health Data - MarkTechPost  A new Google study introduces the Personal Health Large Language Model (PH-LLM), a version of Gemini fine-tuned for text understanding and numerical time series analysis of personal health data. PH-LLM is designed to enhance the interpretation of complex health information, aiding in more accurate and personalized health management. This innovation aims to improve the integration and analysis of diverse health data types, contributing to better health outcomes and patient care.

• Apple Releases 20 New Open Source AI Models Apple has released 20 new open-source AI models, enhancing natural language processing capabilities for applications such as language translation and text generation. These models are freely available to developers, supporting innovation and integration of advanced AI features into various projects. This initiative highlights Apple's commitment to contributing to the AI research community and fostering technological advancements. Available on Apple’s Hugging Face page here: Apple   

• Democratizing Large Language Model development with InstructLab support in watsonx.ai - IBM Blog IBM's introduction of InstructLab within its watsonx.ai platform is a significant move towards democratizing AI development. This initiative, which facilitates community-driven customization and lowers training costs through synthetic data generation, highlights IBM's commitment to fostering collaborative and accessible AI innovation. However, the true test will be whether this tool can effectively engage the wider developer community and produce meaningful advancements in large language model development. Don’t you think so?  

• Understanding the visual knowledge of language models | MIT News  MIT's latest research delves into how language models understand and interpret visual information. By integrating textual and visual data, these models can achieve a more comprehensive understanding, enhancing tasks like image captioning and visual question answering. The study underscores the potential of multimodal AI systems to improve various applications, from accessibility tools to advanced AI assistants. However, it also highlights challenges in aligning visual and textual information accurately, necessitating further innovation and refinement.

Gen AI news from different industries

GenAI) is enhancing banking with personalized services, improving customer experience. In law, AI significantly cuts task times but widens the gap between large and small firms. Retail rapidly adopts GenAI for better customer engagement and efficiency, boosting revenue. AI impacts education by necessitating adaptive policies for job markets. In travel, AI streamlines processes and enhances security despite privacy concerns. GenAI transforms customer experience (CX) with deeper insights and multilingual support. In asset management, AI aids predictive maintenance and decision-making. HR benefits from AI through task automation and better decision-making. GPT-4 with RAG improves clinical trial accuracy and efficiency. Scientist.com's ELISA AI accelerates drug discovery. The Global Telco AI Alliance's multilingual LLM aims to enhance global telecom customer interactions. Roblox's 4D AI advances interactive 3D modeling, and AI optimizes renewable energy deployment, improving grid management.

Check out this conversation with Hema Kadia . We discussed how generative AI can transform various business operations by providing insights, automating tasks, and enhancing decision-making processes. We also addressed the challenges and potential impacts on industries, emphasizing the importance of strategic implementation to leverage AI effectively for business growth and innovation. Give it a read or a listen here: https://tecknexus.com/podcast-tech/generative-ai-for-business-insights-from-eugina-jordan/ 

Banking

• Unlocking Personalization in Banking with GenAI GenAI enables banks to tailor services and communications to individual customer needs by analyzing vast amounts of data and generating personalized insights and recommendations. This technology can improve customer experience, drive engagement, and foster loyalty by offering more relevant and timely interactions. The use of GenAI in banking also includes creating personalized financial advice, automated customer support, and customized marketing strategies, ultimately transforming the way banks interact with their customers.

Legal

• Can Law Firm Gen AI Really Can Cut Lawyers' Time In Half?  A survey by UK law firm Ashurst demonstrated that generative AI can significantly reduce lawyers' time spent on drafting briefs and other tasks. From November 2023 to March 2024, Ashurst conducted trials across 23 offices, showing AI saved up to 80% of time on corporate filings and 45% on legal briefs. The study found AI-generated content to be nearly indistinguishable from human work, with substantial productivity boosts reported. The report emphasizes that AI should supplement, not replace, human lawyers, requiring a balanced strategy and ongoing digital literacy development. All the lawyers who read this newsletter, do you agree?

• Is Gen AI Creating A Divide Among Law Firms Of Haves and Have Nots? | LawSites  generative AI is creating a stark divide between law firms with the resources to adopt these advanced technologies and those without. Firms leveraging AI see improved efficiency and service, gaining a competitive edge, while smaller firms struggle to keep up due to high costs and technical barriers. This growing disparity raises concerns about accessibility and fairness in legal services, highlighting the need for more inclusive technology solutions. 

Retail 

• GenAI’s impact on retail unfolding faster than expected Generative AI is revolutionizing the retail sector faster than anticipated, enhancing customer engagement, operational efficiency, and resilient business models. Key insights from Bain & Company highlight its role in content creation, personalized shopping, and conversational search. Retailers are leveraging GenAI to improve marketing strategies, streamline operations, and provide personalized customer experiences, potentially increasing revenue by 5-10%. This rapid adoption reflects a shift from cautious exploration to confident integration, driven by the technology's ability to deliver significant value and efficiency. 

Education

• Welcome to the Machine: Opportunities and Risks of Generative Artificial Intelligence for Education | The New Zealand Initiative It examines examines the impact of automation and AI on the workforce. It discusses how technological advancements could displace certain jobs while creating new opportunities and challenges. The report emphasizes the need for adaptive policies and education systems to prepare workers for a rapidly changing job market and ensure equitable growth. REPORT SUMMARY 

Travel

• How Artificial Intelligence Is Impacting How We Travel | Inc.com Artificial intelligence is transforming immigration and travel by streamlining processes and enhancing security. AI is used for visa application processing, reducing wait times, and improving accuracy. In airports, AI-powered facial recognition expedites boarding and security checks. Additionally, AI helps analyze vast amounts of data to detect fraud and manage border control more effectively. These advancements make travel more efficient and secure while also presenting challenges related to privacy and ethical use. I need to renew my passport, do you think AI can help? ;)  

Customer Experience (CX)

• 7 ways generative AI can drive innovation in CX strategies -  Generative AI is revolutionizing customer experience (CX) strategies in seven key ways: providing deeper customer insights, enabling sentiment analysis, predicting call intents, analyzing agent performance, deploying AI-powered virtual assistants, modernizing legacy software, and offering multilingual support. These innovations transform contact centers into advanced tech hubs, enhancing service efficiency and personalization. Embracing GenAI helps businesses anticipate customer needs and stay ahead in a rapidly evolving digital landscape.

Asset Management

• 6 ways generative AI can optimize asset management | Supply Chain Dive These include predictive maintenance by analyzing equipment data to forecast failures, optimizing inventory by predicting demand, enhancing decision-making through data-driven insights, automating routine tasks to improve efficiency, personalizing customer experiences, and improving supply chain resilience by anticipating disruptions. These applications help companies maximize asset utilization, reduce costs, and increase operational efficiency.  

HR

• Article: GenAI in HR: From Buzzword to Business Strategy — People Matters – discusses how generative AI (GenAI) is transitioning from a buzzword to a strategic tool in HR. It highlights GenAI's potential to revolutionize HR practices through personalized employee experiences, efficient talent acquisition, and data-driven decision-making. By automating routine tasks, providing predictive analytics, and enhancing employee engagement, GenAI helps HR professionals focus on strategic initiatives. The integration of GenAI in HR is becoming essential for companies aiming to optimize their workforce management and gain a competitive edge. 

Healthcare

• GPT-4 enhances clinical trial screening accuracy and cuts costs GPT-4, especially with Retrieval-Augmented Generation (RAG), significantly enhances clinical trial screening accuracy and efficiency, reducing costs. Research from Harvard Medical School and others demonstrated that GPT-4V with RAG could efficiently analyze both structured and unstructured EHR data, accurately determining eligibility criteria. The RECTIFIER system showed metrics comparable to or better than human reviewers while being more cost-effective.

Science 

•   Scientist.com Unveils Elisa: An LLM-Powered Research Assistant for Scientists in the Pharmaceutical and Biotechnology Industries Scientist.com has launched ELISA, a new AI-driven large language model (LLM) designed to enhance drug discovery by improving the selection and analysis of antibodies. This tool aims to accelerate research processes and improve accuracy in identifying suitable antibodies, significantly aiding scientists in their work.  

Telco

• Global Telco AI Alliance sign JV to establish multi-lingual LLM designed for telcos The Global Telco AI Alliance (GTAA) has announced a joint venture to create a multilingual large language model (LLM) specifically for telecommunications companies. This LLM will support languages including Korean, English, German, Arabic, and Bahasa. The initiative, involving partners like Deutsche Telekom, e& Group, Singtel, SoftBank Corp., and SK Telecom, aims to enhance customer interactions and provide innovative AI solutions for a global audience. The JV will deploy tailored AI applications, potentially serving around 1.3 billion customers across 50 countries. Who else besides me thinks that access to data to build the models will be a real challenge? I have spent 24 years in telco and know they do not like sharing …  

Gaming

• Roblox’s Road to 4D Generative AI  Roblox is advancing towards 4D generative AI, which incorporates dynamic interactions in addition to traditional 3D modeling. This new dimension involves understanding appearance, shape, physics, and scripts to create fully interactive and functional objects. Early tools like Avatar AutoSetup and Material Generator are already enhancing 3D creation efficiency. The goal is to allow creators to develop complex interactive experiences more easily, addressing challenges in functionality, interactivity, and controllability of AI-generated assets. 

Energy

• 3 ways to harness the power of generative AI for the energy transition The World Economic Forum article explores how generative AI can drive the energy transition by optimizing renewable energy deployment, improving grid management, and accelerating innovation in energy technologies. It emphasizes the role of AI in enhancing energy efficiency, predicting maintenance needs, and enabling more accurate climate modeling. The integration of AI in energy systems can significantly reduce emissions and support global sustainability goals, but requires careful consideration of data privacy and ethical implications.

Regional and regulatory updates

Generative AI is transforming India's SaaS industry, helping startups like Fractal Analytics, Postman, and Freshworks become million-dollar businesses by enhancing software development and customer service. Google launched its Gemini app in India, supporting nine languages to improve user engagement. A new bill, led by Sen. Ted Cruz, aims to combat AI deepfake porn by holding social media companies accountable for removing such content quickly. Meta paused its AI training plans using European user data due to regulatory concerns over GDPR compliance and user consent.

• Generative AI will Help Build $100 Million SaaS Businesses in India  Generative AI is revolutionizing India's SaaS landscape, enabling the rise of 100 million-dollar businesses. Startups like Fractal Analytics, Postman, and Freshworks are at the forefront, leveraging AI for software development, automated customer service, and personalized marketing. This tech-driven growth is powered by India's robust ecosystem and skilled talent pool, positioning the country as a leader in AI innovation. 

• Google brings Gemini mobile app to India with support for 9 Indian languages | TechCrunch Google has introduced its Gemini mobile app to India, featuring support for nine Indian languages. This move aims to enhance accessibility and user engagement by catering to the diverse linguistic landscape of the country. The app leverages AI to offer personalized content and services, making it easier for users to navigate and utilize Google's ecosystem in their native languages. This launch is part of Google's broader strategy to expand its reach and impact in emerging markets.  

• New AI deepfake porn bill would require Big Tech to police and remove images  Lawmakers on Capitol Hill are urgently addressing the surge in deepfake AI pornographic images affecting everyone from celebrities to high school students. A new bill, spearheaded by Sen. Ted Cruz, R-Texas, aims to hold social media companies accountable for policing and removing such content. The proposed Take It Down Act would criminalize the publication or threat to publish deepfake porn, requiring platforms to remove offending images within 48 hours of a victim's request and make efforts to eliminate copies. Enforcement would be overseen by the Federal Trade Commission. The bill, introduced by a bipartisan group, highlights the severe impact of non-consensual AI-generated images on public figures and minors. Despite broad agreement on the need to tackle this issue, there are competing Senate bills on the approach, including one from Sen. Dick Durbin, D-Ill., focusing on victim rights to sue. The new legislation aligns with broader efforts led by Senate Majority Leader Chuck Schumer, D-N.Y., to develop comprehensive AI regulations.

• Meta pauses plans to train AI using European users' data, bowing to regulatory pressure | TechCrunch  Meta has announced a pause on its plans to train AI systems using data from users in the European Union (EU) and the United Kingdom (UK), following regulatory pressure from the Irish Data Protection Commission (DPC) and the UK’s Information Commissioner’s Office (ICO). These authorities raised concerns about data privacy and compliance with stringent GDPR regulations. Meta intended to update its privacy policy to include the use of public content from Facebook and Instagram for AI training, citing the need to incorporate diverse languages and cultural references. However, privacy activists, notably the organization NOYB, filed complaints arguing that Meta’s approach violated GDPR requirements, particularly regarding user consent. Meta’s reliance on the "legitimate interests" provision of GDPR, previously rejected by the Court of Justice of the European Union for targeted advertising, added to the regulatory challenges. This pause is a significant move amidst ongoing debates over data privacy and AI ethics.

Gen AI for Business Trends, Concerns and Predictions: 

Bain & Company's "Automation Scorecard 2024" highlights that companies investing heavily in automation, particularly generative AI, achieve significant cost savings and technology adoption. Microsoft, OpenAI, and Apple face challenges in dominating the generative AI market due to competition and integration issues. Comprehensive security frameworks are vital for AI deployment, emphasizing data and model security. EY, Unilever, and Ocado stress ethical AI practices. Generative AI drives broader AI adoption in enterprises, though trust and bias concerns remain. AI failures underscore the need for careful management. New AI chips from Oregon State University promise significant energy savings. Debiasing AI reduces creativity, posing challenges for marketers. Ilya Sutskever's Safe Superintelligence Inc. aims for safe AI development. Addressing AI biases requires ethical data practices and transparency. YouTube commits to responsible AI innovation, focusing on user experience and ethical standards.

• Automation Scorecard 2024: Lessons Learned Can Inform Deployment of Generative AI | Bain & Company  The "Automation Scorecard 2024" by Bain & Company highlights that companies heavily investing in automation outperform others in cost savings and technology adoption, particularly in generative AI. Leaders, defined as those allocating over 20% of their IT budget to automation, achieved an average of 22% in cost reductions compared to 8% by laggards. Successful automation strategies include enterprise-wide implementation and engaged staff. Generative AI investments are prioritized for new use cases, technology replacement, and enhancing existing processes, driving substantial value across industries like finance and pharmaceuticals.

• Why Microsoft, OpenAI, And Apple Likely Will Not Dominate Generative AI  Microsoft and OpenAI, while leaders in the field, face stiff competition from other tech giants and a rapidly evolving market. Despite their advancements, they may struggle with the integration and scaling of AI across diverse industries. Apple's approach is more cautious, focusing on integrating AI into its ecosystem rather than leading the generative AI market. Apple has also been slower in releasing AI products compared to its competitors, which could limit its influence in the generative AI space. Do you agree with this assessment? 

• Generative AI security requires a solid framework It outlines five key areas for securing the AI pipeline: data security, model security, usage security, infrastructure security, and AI governance. The article stresses the importance of comprehensive security measures, including data encryption, vulnerability scanning, API protection, and continuous monitoring to mitigate risks associated with AI deployment. 

• EY, Unilever, Ocado Share Strategies for Ethical AI Deployments  EY, Unilever, and Ocado have shared their strategies for deploying ethical AI, emphasizing the need for responsible AI practices to ensure trust and minimize risks. EY highlights a comprehensive framework for AI governance, stressing the importance of aligning AI systems with ethical norms and regulatory requirements. Unilever focuses on integrating AI ethics into its operations, using tools and partnerships to ensure AI systems are transparent and accountable. Ocado, a leader in online grocery, implements AI with a strong focus on ethical considerations to enhance efficiency while maintaining customer trust. These strategies underscore the critical balance between innovation and ethical responsibility in AI deployment. 

• Gen AI Inspiring Greater Enterprise Adoption of Other AI Types, Says Research A study by Pegasystems revealed that the rise of generative AI (Gen AI) is driving greater adoption of other AI types among enterprises. Key findings include 95% of businesses crediting Gen AI for their AI adoption, with 44% using it for creative tasks like content creation. While 92% plan to increase AI use, many overestimate their understanding of AI, with 80% unaware of its long history. Trust in AI is complex, with 51% concerned about transparency and bias. The demand for AI skills is high, highlighting the need for expertise in AI implementation.  

• You are going to love this one: 32 times artificial intelligence got it catastrophically wrong | Live Science Examples include AI chatbots dispensing incorrect and dangerous medical advice, facial recognition systems misidentifying individuals leading to wrongful arrests, recruitment algorithms displaying bias against certain demographics, and autonomous vehicles causing traffic accidents. These instances reveal the critical risks associated with AI, showcasing how errors can lead to severe social, ethical, and legal repercussions when AI systems malfunction or are improperly managed.

•  New computer chips show promise for reducing energy footprint of artificial intelligence | Oregon State University Researchers at Oregon State University have developed a new AI chip that significantly enhances energy efficiency, potentially reducing AI's energy consumption by six times compared to current standards. These chips use memristors made from entropy-stabilized oxides (ESOs), which allow for efficient in-memory computation, reducing data transfer energy loss. This innovation, mimicking biological neural networks, optimizes energy use for AI tasks and is particularly effective for time-dependent data like audio and video. The research, supported by the National Science Foundation, involves collaboration with several universities.  

• Creativity Has Left the Chat: The Price of Debiasing Language Models  The paper examines the impact of alignment techniques, specifically Reinforcement Learning from Human Feedback (RLHF), on the creativity of large language models (LLMs). Focusing on the Llama-2 series, the study reveals that aligned models exhibit lower entropy, form distinct clusters in the embedding space, and tend towards "attractor states," indicating reduced output diversity. This reduction in creativity has significant implications for marketers who use LLMs for creative tasks, suggesting a careful balance between consistency and creativity is necessary.

• Ilya Sutskever is back with Safe Superintelligence Inc. and is focused solely on developing safe superintelligence, combining rapid advancements in AI capabilities with a commitment to safety. Based in Palo Alto and Tel Aviv, SSI aims to attract top talent to tackle this critical technical challenge. Their mission is to build a safe superintelligence, free from short-term commercial pressures, and advance both safety and capabilities simultaneously. 

• How to Fix “AI’s Original Sin” – O’Reilly discusses the historical context of AI biases, stemming from biased training data. To address these biases, the article recommends several strategies: using ethically sourced and diverse datasets, ensuring transparency and accountability in AI development, and respecting copyright laws to support content creators. It highlights the importance of community engagement and ethical practices to build fairer AI systems. I want to address a critical flaw in the AI development process: the backward approach of using unethically sourced data for training models, which leads to lawsuits and reactive solutions. It seems illogical to steal data, face legal consequences, and then scramble to rectify the situation through licensing agreements. This process not only damages trust but also highlights a fundamental disregard for ethical standards from the outset. Instead, the industry should prioritize ethical data sourcing and transparency from the beginning, ensuring that AI development respects both legal and moral boundaries. This proactive stance would prevent many issues and build a more trustworthy AI ecosystem.

• Our approach to responsible AI innovation - YouTube Blog YouTube's latest blog post outlines a commitment to responsible AI innovation, emphasizing transparency, fairness, and privacy. The platform's focus on building AI systems to improve user experience while safeguarding against harmful content and biases is commendable. However, the true effectiveness of these initiatives will depend on their implementation and ongoing updates to address evolving challenges. YouTube's engagement with external experts and stakeholders is a positive step, but continuous scrutiny and adaptation will be essential to maintaining ethical AI practices. 

News and updates around  finance, Cost and Investments

AWS is investing $230 million in generative AI startups through its Generative AI Accelerator program, offering credits, mentorship, and technical training to 80 startups. In China, companies like Alibaba and Tencent are in a price war over large language models to increase accessibility. Rising AI development costs, estimated at billions, may concentrate power among well-funded firms. Nvidia's market value has surpassed Microsoft due to high demand for its AI and chip technologies, emphasizing generative AI's impact on industry transformation.

• https://www.crn.com/news/cloud/2024/aws-commits-230m-to-ai-startups-to-drive-genai-innovation-here-s-why Amazon Web Services (AWS) has announced a $230 million investment in generative AI startups as part of its global Generative AI Accelerator program. This initiative will support 80 startups globally, providing them with AWS credits, mentorship, and technical training to accelerate the development and deployment of generative AI applications. The program aims to foster innovation in sectors such as healthcare, finance, media, and climate change. Each startup will receive up to $1 million in AWS credits and access to industry experts and resources, including partnerships with NVIDIA for advanced AI tools. Learn more in the official PR: AWS Announces $230 Million Commitment for Generative AI Startups 

• LLM price war sees cream rise to top - Chinadaily.com.cn Remember a few weeks ago, I shared that it started, now as you can see, it’s heating up! China's leading AI companies, including Alibaba, Tencent, and Baidu, are engaged in a price war over large language models (LLMs). This competition is expected to accelerate the development and adoption of advanced AI technologies in China. By reducing prices significantly, these companies aim to make LLMs as ubiquitous as essential utilities like water and electricity, promoting widespread use across industries and driving further innovation in AI applications. 

• Disregard the title, it’s a bit of misleading as it provides a really good analysis of Gen AI European startup funding. The best-funded generative AI startups in Europe have something in common: Big Tech experience.

• The Billion-Dollar Price Tag of Building AI | TIME Artificial intelligence development is becoming increasingly costly, with significant financial investments needed to train next-generation AI systems. Dario Amodei, CEO of Anthropic, estimates the cost for upcoming AI systems to be around $1 billion, with future generations potentially costing $10 billion. Similarly, Microsoft and OpenAI plan to build a $100 billion supercomputer for AI development. A new study by Stanford University and Epoch AI researchers reveals that the costs of training advanced AI systems have been doubling every nine months due to the growing computational power required and high employee compensation. This trend suggests that only well-funded companies, like Google, Amazon, Microsoft, Meta, and their backed startups, will be able to compete, potentially leading to a concentration of power in the AI industry. The study highlights the need for responsible development and deployment of AI, urging developers and policymakers to consider the implications of such concentrated power. For me, this consumption raises critical concerns about the long-term viability of such energy-intensive AI development. The industry must grapple with whether the marginal benefits of increasingly large models justify the hefty environmental costs. 

• Understanding the Cost of Generative AI Models in Production blog post provides a comprehensive analysis of the true costs of deploying generative AI models, highlighting that raw compute pricing is just the tip of the iceberg. The total cost of ownership (TCO) includes infrastructure costs like container services, networking, load balancing, and logging, as well as significant hidden costs in development time and maintenance. Schmid emphasizes that opting for managed services often proves more cost-effective in the long run, as they reduce the burden on internal resources and expedite time-to-market.

• Gen AI Boom Drives Nvidia Value to Overtake Microsoft | Technology Magazine Nvidia has surpassed Microsoft in market value, driven by the booming demand for its AI and chip developments, reaching a market cap of $3.3 trillion. This surge is largely due to the popularity of its high-end GPUs, essential for AI development. Nvidia's rapid growth reflects the significant role of generative AI in business transformation, with many companies integrating these capabilities for a competitive edge. The company's advancements in AI technology, such as the new Blackwell GPU, continue to set industry standard.  

What/where/how Gen AI solutions are being implemented today?

Salesforce launched a generative AI benchmark for CRM to help businesses select suitable LLMs based on real-world data. Target introduced an AI tool for employees to improve inventory management and customer service. McDonald's dropped IBM's AI order tech to explore new drive-thru innovations, highlighting AI integration challenges. Verizon uses generative AI to enhance customer loyalty by predicting intent and matching calls with suitable agents, aiming to retain 100,000 customers.

• Salesforce com : Announces the World’s First LLM Benchmark for CRM - MarketScreener  Salesforce has launched a generative AI benchmark for Customer Relationship Management (CRM), aimed at helping businesses choose the most suitable large language models (LLMs) for their applications. This benchmark considers critical factors like compliance, security standards, cost, accuracy, trust and safety, and speed. Unlike academic benchmarks, Salesforce’s tool evaluates LLMs using real-world CRM data across metrics such as factuality, completeness, conciseness, instruction-following, operational cost, response speed, and data security. This comprehensive evaluation helps businesses optimize their AI technology stack, ensuring the selected model aligns with their specific CRM goals and use cases. The practical utility of this benchmark is emphasized for diverse CRM scenarios.

• Target leans in on AI with new tool for store employees Target has introduced a new AI companion tool for its store employees, designed to assist with various tasks. This tool aims to enhance efficiency by helping staff manage inventory, provide better customer service, and streamline operations. The AI tool leverages advanced algorithms to offer real-time insights and support, ultimately improving the shopping experience for customers and the working experience for employees. Official press release: Target to Roll Out Transformative GenAI Technology to its Store Team Members Chainwide  

• McDonald's Drops IBM's AI Order Tech, Seeks New Drive-Thru Tech McDonald's decision to drop IBM's AI order technology in favor of exploring new drive-thru tech options highlights the fast food giant's pursuit of innovation in customer service. The move raises questions about IBM's technology efficacy and McDonald's strategic direction. This shift could signal a broader industry trend towards more advanced, efficient AI solutions. However, it also underscores the challenges companies face in integrating cutting-edge technologies effectively into their operations. Now I want a Big Mac. ;) With a side of chicken nuggets. Mmmm. 

•  Verizon uses GenAI to improve customer loyalty | Reuters Verizon's innovative use of generative AI to enhance customer loyalty highlights both the potential and complexity of integrating advanced technologies in customer service. By predicting customer intent and matching calls with the most suitable agents, Verizon aims to retain 100,000 customers and improve service efficiency. While this approach showcases the benefits of AI in reducing churn and personalizing customer interactions, it also raises questions about data privacy and the long-term effectiveness of such large-scale AI deployments. Balancing technological advancement with ethical considerations will be crucial for sustained success. 

Women Leading in AI 

New Podcast:  Learn about Shaping the Future AI, VC, Boardroom Dynamics with Sarah Benson-Konforty on the latest Women And AI podcast. This episode covers the role of AI in healthcare and the challenges faced by women in the VC space. Sarah shares her insights on the state of AI in venture capital and offers advice for startups seeking funding. Tune in anywhere you get your podcasts. To catch the full episode now to learn more about AI’s intersection with VC investing and the boardroom. 

Featured AI Leader: 🏆Women And AI’s Featured Leader - Ewa Ding 🏆

Join us in celebrating Ewa her work as a leader in the space with UX4AI. Learn how Ewa is using AI and how she sees it impacting UX. 

Learning Center

• https://github.com/microsoft/generative-ai-for-beginners/tree/main The GitHub repository "Generative AI for Beginners" by Microsoft offers a comprehensive introduction to generative AI, covering fundamental concepts, techniques, and practical applications. The content is structured into modules that include theoretical explanations, hands-on tutorials, and coding exercises. It aims to equip beginners with the knowledge and skills needed to understand and implement generative AI models. Topics include neural networks, GANs, and transformer models. This resource is ideal for those new to AI or looking to expand their expertise in generative models. 

• Everything you wanted to know about ChatGPT and its evolution: ChatGPT: Everything you need to know about the AI-powered chatbot ChatGPT has seen major updates, including the integration of GPT-4 for more natural responses and the launch of ChatGPT Enterprise for advanced business use. Key features include internet access via plugins, text-to-image generation with DALL-E 3, and educational tools for teachers. The timeline highlights these enhancements and shows ChatGPT's growth, from the introduction of AI tools in Opera GX to expanding capabilities in education and enterprise settings.  

• NVIDIA Introduces Five New Free Technical Courses for Developers - Blockchain.News NVIDIA has introduced free technical courses to help developers, data scientists, and IT professionals enhance their skills in AI, data science, and accelerated computing. These courses feature hands-on labs and real-world examples, making it easier for participants to apply what they learn. This initiative aims to support ongoing education and innovation in the tech industry. Which one will you be taking first? 

Question(s) of the week

There was a comment last week on the newsletter, “I am working on LLM guardrails (firewall) and some of your posts are very relevant… I’m looking forward to gaining more insights generally on LLMs/GenAI, more specifically the insights related to safety/security and associated guardrails.”

So, I reached out to my trusted network to get the answer. I have known Amit Ghadge for over 16 years, he is an engineer by trade, an exceptional product managing leader, and now a Gen AI consultant to some big global brands. Here is his answer: 

You are a chatbot designed to assist users with information about products. Your primary functions include providing detailed product descriptions, answering questions about product features, availability, pricing, and guiding users through the purchasing process. Your interactions should always be helpful, respectful, and adhere to the highest standards of professionalism.

 → This is a system instruction to the LLM which is not visible to the user. The last sentence in the instruction above is about the guardrails.

Do you have any other questions we can help you answer? 

Prompt of the week

• Talila Millman , Managing Director, MillmanTech, and Author, The TRIUMPH Framework: 7 Steps to Leading Organizational Transformation, suggested these prompts. “If you are writing in Python and you want a piece of code that will print the current date. Just write this comment: # Print the current date. Another example: you want to create a Python function that generates the sum of two numbers, just create a python comment that says that, and the AI assistant will create a code snippet with the function. Here is how the comment will look like: # Create a function that calculates the sum of two numbers. And Talila also provided an example of other useful prompts: Find any problems with this code. Write unit tests for the following function. The examples provided work specifically within an Integrated Development Environment (IDE) with an AI assistant.

• [2406.06608] The Prompt Report: A Systematic Survey of Prompting Techniques  The paper "The Prompt Report: A Systematic Survey of Prompting Techniques"(download here: arXiv:2406.06608v1 [cs.CL] 6 Jun 2024)  presents a comprehensive analysis of prompting techniques in generative AI. It establishes a structured taxonomy of 58 text-based prompting methods and 40 for other modalities, alongside a vocabulary of 33 terms. The study highlights the varied approaches and conflicting terminologies in prompt engineering, aiming to clarify and categorize these methods for improved understanding and application in AI systems. 

• And if you are using ChatGPT to help you create things, this is a good prompt guide:

I have personally built a lot of different business and technical materials in ChatGPT from different business plans, roadmaps, and test plans by asking to think “as a business leader or product manager or an engineer” to do a task and have provided clear parameters, dates, and requirements for a roadmap or the plan or a roadmap. You will need to reiterate the output and check the variables until you get the desired results. 

Tools and Resources

• 21 No-Code AI Tools to Streamline Your Workflows  
• Newsletter by Gen AI for Telco & Tech by Hema Kadia  that covers news and developments of Gen AI in telco and tech. Enjoy!

• GitHub - abgulati/LARS: An application for running LLMs locally on your device, with your documents, facilitating detailed citations in generated responses. LARS (Layer-wise Adaptive Rate Scaling) is a repository on GitHub providing an implementation of an optimization algorithm designed for training large-batch machine learning models. LARS improves the convergence rate by adjusting learning rates for each layer individually based on the layer's weight norms and gradient norms, promoting stable and efficient training. This technique is particularly useful for scaling up the training of deep neural networks and achieving better performance on large datasets. 

• Step by Step guide to develop AI Multi-Agent system using Microsoft Semantic Kernel and GPT-4o provides a step-by-step guide for developing AI multi-agent systems using Microsoft Semantic Kernel and GPT-4.0. The guide covers the integration of Semantic Kernel with GPT-4.0 to create intelligent agents capable of complex tasks. It details setting up the development environment, creating and configuring agents, and deploying the system for practical applications. This approach aims to leverage advanced AI models to enhance automation and efficiency in various domains. 

If you enjoyed this newsletter, please comment and share. If you would like to discuss a partnership, or invite me to speak at your company or event, please DM me.