Welcome to your weekly AI Tech Circle Newsletter!

I’m building and implementing AI solutions and sharing everything I learn along the way…

Check out the updates from this week! Please take a moment to share them with a friend or colleague who might benefit from these valuable insights!

Today at a Glance:

• Insights from the 2025 AI Index Report by Stanford HAI
• Generative AI Use case: Summarization of Legal Documents using Generative AI
• AI Weekly news and updates covering newly released LLMs
• Courses and events to attend

Insights from the 2025 AI Index Report by Stanford HAI

If someone wants to get up to stay about the affairs of AI, then the must-read report is “The 2025 AI Index Report from Stanford University Human-Centered Artificial Intelligence“. If you are interested in the 2024 report and the highlights, refer to the earlier edition,’Exploring Global AI Trends: Insights from the AI Index 2024 Report by Stanford HAI.’​

This year’s report has 12 key takeaways:

Economy: Growing record investment and usage:

• Total corporate AI investment hit $252.3 billion, driven by a 44.5% increase in private funding and a 12.1% rise in M&A activity compared to 2023.
• Over the past decade, the sector has expanded rapidly, with overall investment growing 13 times since 2014.
• In 2024, private AI investment in the U.S. surged to $109.1 billion, outpacing China’s $9.3 billion nearly 12-fold and the U.K.’s $4.5 billion by 24 times.
• Generative AI drew significant attention, accounting for $33.9 billion in global private funding, marking an 18.7% rise over 2023.
• Business adoption is also rising rapidly: 78% of organizations reported using AI in 2024, up from 55% in 2023.

​

• Governments worldwide are ramping up AI infrastructure spending:
• Canada unveiled a $2.4 billion initiative,
• China introduced a $47.5 billion fund for semiconductor development,
• France committed $117 billion to AI infrastructure,
• India pledged $1.25 billion,
• Saudi Arabia’s Project Transcendence features a massive $100 billion AI investment.

AI is starting to generate Financial Value across business areas

• Many organizations report some financial benefit from AI adoption, but the impact remains modest.
• Among users, 49% see cost reductions in service operations,
• 43% in the supply chain, and
• 41% in software engineering, yet most savings are under 10%.
• On the revenue side, 71% of companies use AI in marketing and sales report gains,
• followed by 63% in supply chain and
• 57% in service operations, though increases typically stay below 5%.

AI model Usage price is continuously reducing:

• Across various tasks, inference costs for large language models have dropped by factors ranging from 9x to 900x annually.
• The cost to run a model performing at GPT-3.5 level (MMLU score: 64.8) has plummeted from $20.00 per million tokens in November 2022 to just $0.07 per million tokens by October 2024 using models like Gemini-1.5-Flash-8B.
• That’s a reduction of over 280x in less than two years.

​

AI Infrastructure – becoming faster, more cost-effective, and energy efficient

• Recent findings show that ML hardware performance, measured in 16-bit FLOPs, has grown at 43% yearly, doubling every 1.9 years.
• At the same time, hardware costs are falling by 30% annually, and energy efficiency is improving by 40% yearly.

AI is surpassing new benchmarks with the highest speed.

• In 2023, researchers launched advanced benchmarks like MMMU, GPQA, and SWE-bench to push the boundaries of AI capabilities.
• By 2024, the model performance had surged in accuracy on MMMU and GPQA, which rose by 18.8 and 48.9 percentage points, respectively.
• SWE-bench saw the most dramatic leap: AI systems went from solving 4.4% of coding tasks in 2023 to 71.7% just a year later.

​

Responsible AI: Organizations recognize responsible AI risks, but action remains limited

According to a McKinsey survey, while many organizations are aware of critical RAI risks, such as inaccuracy (64%), regulatory compliance (63%), and cybersecurity (60%), proactive mitigation is still lacking across the board.

​

Responsible AI: Transparency in foundation model research is improving, but gaps persist

The latest Foundation Model Transparency Index update shows that average transparency scores among leading developers rose from 37% in October 2023 to 58% by May 2024.

Despite this progress, significant areas still lack adequate disclosure.

AI exceeds physician performance on several critical clinical tasks

• Recent research shows GPT-4 outperforming doctors, whether assisted by AI or not, in diagnosing complex medical cases.
• Additional studies highlight AI’s superiority in detecting cancer and flagging patients at high mortality risk.
• Still, early evidence suggests that combining AI with human expertise often leads to the best outcomes, pointing to a promising area for continued investigation.

Awareness of AI’s influence on everyday life is growing globally

• Roughly two-thirds of people now expect AI-driven products and services to significantly affect daily life within three to five years, up six percentage points since 2022.
• This perception has increased in nearly every country, with the sharpest rises seen in Canada (17%) and Germany (15%), except for slight declines in Malaysia, Poland, and India.

​

Call for Action:

I suggest spending some time reading the ​2025 AI Index report​. It will give you an overview of what is happening and how it shapes the world and industries.

​

Weekly News & Updates…

Last week’s AI breakthroughs marked another leap forward in the tech revolution.

1. Tencent Hunyuan’s deep reasoning model T1 has been significantly upgraded, including enhanced project-level code generation skills, improved text generation and writing quality, and Better multi-turn text comprehension, instruction-following, and word-level understanding.
2. ​PaperBench released by OpenAI, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. PaperBench evaluates replication attempts using detailed rubrics co-developed with the original authors of each paper. These rubrics systematically break down the 20 papers into 8,316 precisely defined requirements that an LLM judge evaluates.
3. ​Llama 4 models from Meta: These models are re-designed to use state-of-the-art mixture-of-experts (MoE) architecture and natively trained with multimodality. Three types are introduced: 1- Llama 4 Scout is a small model with 17B activated parameters with 16 experts. It achieves an industry-leading 10M+ token context window and can also run on a single GPU. 2 – Llama 4 Maverick is the best multimodal model in its class, beating GPT-4o and Gemini 2.0 Flash across a broad range of widely reported benchmarks while achieving comparable results to the new DeepSeek v3 on reasoning and coding – at less than half the active parameters. It offers a best-in-class performance-to-cost ratio with an experimental chat version scoring an ELO of 1417 on LMArena. It can also run on a single host. 3 – Llama 4 Behemoth outperforms GPT4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks. Llama 4 Behemoth is still training.
4. ​TxGemma is an LLM for multiple therapeutic tasks for drug development 2B, 9B, and 27B. It is fine-tunable with transformers. TxGemma models, fine-tuned from Gemma 2 using 7 million training examples, are open models designed for prediction and conversational therapeutic data analysis. These models are available in sizes 2B, 9B, and 27B. Each size includes a ‘predict’ version, tailored explicitly for narrow tasks drawn from Therapeutic Data Commons.

The Cloud: the backbone of the AI revolution

• Industrial Ecosystem Adopts Mega NVIDIA Omniverse Blueprint to Train Physical AI in Digital Twins link​
• Resell Oracle Database@Google Cloud through the Google Cloud Marketplace link​
• Google Cloud Next 2025 – The opening keynote​

Generative AI Use Case of the Week:

Several Generative AI use cases are documented, and you can access the library of generative AI Use cases. Link​

Summarization of Legal Documents using Generative AI

Generative AI automatically converts lengthy legal texts into clear summaries. These summaries present essential details, critical clauses, obligations, and important dates. The process enhances legal teams’ review efficiency and helps management quickly understand key points.

Business Challenges

• Lengthy legal texts require significant time for manual review.
• Legal teams spend excessive hours analyzing dense contracts and compliance documents.
• High risk of human oversight leading to potential errors or missed obligations.
• Manual summaries vary in quality and consistency across teams.
• Increased operational costs due to intensive human resource use.

AI Solution Description

Generative AI analyzes and identifies essential information in legal documents. Using large language models trained in legal language, the system creates accurate summaries highlighting obligations, deadlines, risks, and key clauses. The AI-generated summaries provide legal professionals with concise, reliable insights for informed decision-making.

Expected Impact / Business Outcome

Revenue: Faster document review allows legal teams to handle more cases. Improved turnaround time supports quicker business decisions, accelerating deal cycles.

User Experience: Legal professionals receive concise summaries, reducing workload stress. Decision-makers quickly grasp essential points, improving satisfaction with legal processes.

Operations: AI-generated summaries improve consistency and accuracy in reviews. Automated summarization reduces manual review bottlenecks and resource allocation issues.

Process: Streamlined document processing shortens review cycles significantly. Improved accuracy decreases rework, enhancing productivity across teams.

Cost: Significant reduction in operational expenses through decreased reliance on manual labor. Lower risk of costly errors due to oversight in manual document review.

Required Data Sources

• Historical legal documents: contracts, agreements, policy documents.
• Compliance and regulatory standards data.
• Past manual summaries and legal review records to refine AI accuracy.
• Databases of relevant laws and regulations for contextual analysis.

Strategic Fit and Impact

• Aligns directly with operational efficiency goals.
• Enhances the ability of the legal team to manage high-volume document processing.
• Supports strategic objectives around cost control, accuracy, and rapid decision-making.
• Strengthens compliance capabilities through accurate and timely reviews.
• Positively impacts overall organizational agility and responsiveness.

Favorite Tip Of The Week:

Here’s my favorite resource for the week.

​Cohere has reported the results of the survey that 20% of the AI projects face delays during deployment and highlighted these 5 points which are imperative to address before starting any AI Project:

1. Costs: AI can get expensive fast – API-based service fees mount up, while private deployments require significant upfront investment. Smart budgeting is a must.
2. Security and compliance: Regulations are tightening, and AI systems handling sensitive data need bulletproof safeguards to stay compliant.
3. Performance and scale: Capacity issues and computing bottlenecks can cripple performance without the correct optimizations.
4. Integration: AI’s efficacy is limited if it can’t talk to your systems. Smart use of data connectors and retrieval-augmented generation (RAG) help avoid AI silos.
5. Customization: Off-the-shelf AI may not cut it for complex needs. Companies must fine-tune models with high-quality data and expert tweaks.

Potential of AI:

​GitHub releases the GitHub MCP Server. Model Context Protocol (MCP) is an AI tool calling standard that has quickly gained traction in recent months. MCP tools provide LLMs with a uniform method to invoke functions, retrieve data, and engage with the environment. Anthropic developed the protocol and established the initial GitHub MCP server, which has become one of the growing ecosystem’s most widely used MCP servers.

Things to Know…

Google has introduced the Agent2Agent (A2A) protocol, an open standard designed to enable AI agents from different vendors and frameworks to collaborate and exchange information across enterprise platforms. This initiative aims to foster a future of seamless AI agent interoperability and enhanced automation.​

3 Key Takeaways:

1. Open and Secure Communication: A2A is built on existing standards like HTTP and JSON-RPC, ensuring easy integration. It emphasizes enterprise-grade security, supporting robust authentication and authorization mechanisms.​
2. Dynamic Agent Collaboration: The protocol allows agents to discover each other’s capabilities through “Agent Cards” and manage tasks collaboratively. This facilitates complex workflows, such as candidate sourcing, by enabling agents to coordinate actions and share information effectively.​
3. Broad Industry Support: Over 50 technology partners, including Atlassian, Box, Cohere, and Salesforce, contribute to A2A’s development.

I liked this simple explanation of Agent2Agent vis MCP:

Avi Chawla @_avichawla Agent2Agent Protocol vs. Model Context Protocol, clearly explained (with visual): – Agent2Agent protocol lets AI agents connect to other Agents. – Model context protocol lets AI Agents connect to Tools/APIs. Both are open-source and don’t compete with each other! 10:56 AM • Apr 10, 2025 198 Retweets 831 Likes Read 31 replies

​

The Opportunity…

Podcast:

• This week’s Open Tech Talks episode 156 is “Building and Growing an AI Startup with Sébastien Night” he is a Co-Founder of OneTake AI.

Apple | Amazon Music​

Building and Growing an AI S… Mar 23 · OPEN Tech Talks: Technol… 43:20   ​

Courses to attend:

​10 Lessons teaching everything you need to know to start building AI Agents

​OpenAI has launched academy with several contents and courses to educate.

Events:

• ​Dubai AI Festival 2025, April 23-24, 2025, Dubai, UAE
• ​Machines Can See conference, April 23-24, 2025, Dubai, UAE

Tech and Tools…

• ​Krillin AI is a solution for video translation, dubbing to voice cloning, formatting, seamlessly converting videos between landscape and portrait modes for optimal display across all content platforms(YouTube, TikTok, Bilibili, Douyin, WeChat Channel, RedNote, Kuaishou)
• ​Chat SDK is a free, open-source template built with Next.js and the AI SDK that helps you quickly build powerful chatbot applications
• ​Dify is an open-source LLM app development platform. Its intuitive interface combines agentic AI workflow, RAG pipeline, agent capabilities, model management, observability features.

The Investment in AI…

• ​Unravel has raised $7 million in Series A funding for its artificial intelligence (AI)-powered travel video content which helps user discover and book trips.
• Google’s Isomorphic Labs raised $600 million in its first external funding round to apply artificial intelligence to the drug development process.

And that’s a wrap!

Thank you, as always, for taking the time to read.

I would love to hear your thoughts. Please reply and share what you found most valuable this week. Your feedback means a lot to me.

Until next week,

Kashif Manzoor