👩‍💻 When AI is fed AI-generated data; Amazon races Nvidia

This week: 

• 40% of global companies use AI, with India leading adoption at 59%; the global AI market is projected to reach $1.85 trillion by 2030, with significant usage in customer service, cybersecurity, and digital assistants. 

• Amazon is developing custom AI chips to compete with Nvidia, aiming to reduce costs, improve performance, and maintain competitiveness in AI and cloud computing. 

• 65% of nonprofits are open to AI, but only 12% use it; key uses include unlocking big data for climate action, creating collaboration tools, and leveraging data sets for efficiency and impact, with notable examples like Earth Species Project and WattTime. 

Read below. 

💼 AI in Business

Extent of AI Usage in Companies

Some key statistics: 

• Global AI Usage: 40% of global companies currently use AI, and 82% are either using or exploring AI technologies. 

• Regional Adoption: India leads in AI adoption with 59% of companies implementing AI, followed by the UAE at 58% and Singapore at 53%. The United States has a lower adoption rate at 33%. 

• Market Projections: The global AI market is projected to reach $1.85 trillion by 2030, growing at a 37.3% compound annual growth rate (CAGR) from 2024 to 2030. 

• Enterprise Adoption: Larger enterprises are twice as likely to use AI compared to smaller businesses, with 42% of companies with over 1,000 employees using AI. 

• Usage Trends: Common uses of AI in business include customer service (56%), cybersecurity and fraud prevention (51%), and digital assistants (47%). 

• Employment Impact: AI saves employees an average of 2.5 hours per day and 66% of business owners have hired employees specifically to implement or leverage AI processes. 

Amazon races Nvidia 

Amazon is developing its own AI chips to compete with Nvidia, aiming for cheaper and faster solutions. 

• The initiative is driven by the high cost of Nvidia chips and a desire to reduce dependence on Nvidia for AI cloud services. 

• Amazon’s custom chip development has dual goals: providing affordable options for complex calculations and maintaining competitiveness in the AI and cloud computing industry. 

• Amazon’s AI chips, including Trainium and Inferentia, are in early stages but show promise for significant cost-performance improvements. 

• AWS, a major growth driver for Amazon, relies heavily on these advancements, with AWS making up nearly a fifth of Amazon's total revenue. 

• Amazon's Graviton chips, now in their fourth generation, and custom AI chips were crucial during the recent Prime Day event, supporting high levels of online activity. 

When AI is fed AI-generated data 

 In a recent study published in Nature, researchers found that training AI models on AI-generated text leads to "model collapse," where the models produce nonsense. This phenomenon occurs as AI-generated text contaminates the training data, hindering the improvement of large language models (LLMs). 

• Training AI on AI-generated text leads to "model collapse," producing nonsense outputs. 

• The problem arises due to AI-generated text polluting the training data. 

• This issue can affect all AI models using uncurated data, including image generators. 

• AI models trained on synthetic data forget less common information, becoming more homogeneous. 

• This affects AI fairness, as rare events often relate to marginalized groups. 

• The phenomenon is compared to inbreeding, amplifying errors with each generation. 

• Solutions include using more real data, filtering AI-generated content, and watermarking AI outputs. 

• Collaborative efforts from tech companies are necessary to mitigate this issue. 

🌐 From the Web

AI is complicating plagiarism. How should scientists respond? 

Generative AI complicates plagiarism detection in academic writing, raising ethical and procedural concerns. Researchers debate its permissible use, disclosure requirements, and potential for increasing academic dishonesty. 

Method prevents an AI model from being overconfident about wrong answers 

MIT researchers developed the "Thermometer" technique to efficiently calibrate large language models, preventing overconfidence in incorrect answers and improving trust in AI predictions. 

Microsoft sinks, chipmakers climb as AI rally faces divide 

Microsoft shares fell 2% due to slow cloud growth, while Nvidia and other chipmakers rose, highlighting a divide in the AI market's impact on investors. 

🏳️ AI for Good

AI-Powered Nonprofit Highlights

While 65% of nonprofits are open to AI, only 12% currently use AI, indicating a gap between interest and implementation in the nonprofit sector. Nonetheless, some nonprofits are making great strides in utilizing AI. 

• Major Themes in AI Use: Nonprofits use AI to unlock big data for climate action, start small with specific use cases, create tools for collaboration, leverage robust data sets, and use smaller models for efficiency and impact. 

• AI for Interspecies Communication: AI-powered nonprofits like Earth Species Project and Project CETI have discovered sophisticated communication patterns in sperm whales, potentially enabling interspecies communication, which could lead to better environmental protections. 

• Climate Change Mitigation Potential: A 2023 BCG report found that AI could help mitigate 5% to 10% of global greenhouse emissions by 2030, showcasing AI’s significant potential in combating climate change. 

• WattTime and Climate TRACE: WattTime, an AI-powered nonprofit, co-founded Climate TRACE, which uses AI and satellites to monitor global greenhouse gas emissions, helping identify high-polluting facilities and reduce CO2 emissions by shifting to cleaner capacities. 

• Digital Green's Farmer.Chat: Digital Green’s AI assistant, Farmer.Chat, supports agricultural extension workers in helping small-scale farmers with climate-smart practices, leading to increased productivity and reduced costs in agricultural extension programs. 

🤖 Prompt of the week

Act as a data scientist and explain the model's results. I have trained a decision tree model and I would like to find the most important features. Please write the code. 

See you next week,

Mukundan

The Data Talent Pulse is brought to you by TeamEpic, a trusted global AI Talent provider. Learn more about us here.