AI Update

June 3, 2025

June 2025

• AI-Assisted Search-based Research is Now Working
  Deep Research plus the latest OpenAI o3 and o4-mini models are excelling at search and reasoning over results—this feels like a step-change from just a few months ago.
• Business Leaders’ Thoughts on AI Possibilities
  Google Cloud asked executives at nine companies how they’re using its AI tools; the article rounds up 600+ real-world use cases across a range of industries.
• Agent2Agent Protocol
  A new open standard (A2A) backed by 50+ tech partners—including Atlassian, Salesforce, MongoDB, SAP, and more—so AI agents can securely exchange data and coordinate actions across enterprise platforms.
• AI-Powered Therapy Shows Promise in First Clinical Trial
  “Therabot” delivered significant symptom improvements for participants with major depressive disorder, generalized anxiety, or an eating disorder—important results as I’m exploring solutions in this space.

April 2025

• ChatGPT 4o Image Generator
  Much better than DALL·E 3. I’ve used it to create technical diagrams, and a founder showed me some images for his investor deck that were pretty impressive.
• OpenAI’s Responses API, Browser-Use and Exa Get Contents
  These are making it easier for us to read and interact with web pages. This is going to be a big new push alongside MCP that enables Gen AI to integrate and operate broadly.
• How to Achieve High-Accuracy Results When Using LLMs
  I’m moderating this upcoming webinar. It should be a good one.

March 2025

• How to Achieve High-Accuracy Results When Using LLMs
• Google’s AI Co-Scientist
  We’ve used a similar approach across all kinds of business ideation problems. You should too.
• Bloomberg – Zuckerberg says Meta to spend $65B on AI in 2025
• Waymo doing better than human drivers
  From an insurance company: 88 percent reduction in property damage claims and a 92 percent reduction in bodily injury claims. Something I’ve thought, but I’ve been wrong so far, that once self-driving cars are known to be safer than human drivers, won’t the liability at some point shift where anyone operating a fleet of human-driven cars has additional liability since it’s known to be less safe?
• Microsoft Report – AI at Work is Here. Now Comes the Hard Part.
  “75% of global knowledge workers using it. And employees, struggling under the pace and volume of work, are bringing their own AI to work. While leaders agree AI is a business imperative, many believe their organization lacks a plan and vision to go from individual impact to applying AI to drive the bottom line. The pressure to show immediate ROI is making leaders inert, even in the face of AI inevitability.”
• Worker experience with AI Chatbots in their jobs (study)
  Usage is lower than you might expect.
• Customers Don’t Care About Your AI Feature
  “Labeling your product as ‘AI’ might not be the boost you think it is. It doesn’t necessarily build trust, justify a higher price, or convince your customers it’ll perform better. And if you misuse it, you may end up doing the one thing no marketer or product manager should do—turning your audience off.”
• Prompt Engineering Playbook – Open AI
• Prompting Guide 101 – Google Gemini
• Prompt Tuning Playbook
  Discusses how to test the tuning of prompts.
• Pre-training is Running Out of Steam and What’s Next
  OpenAI’s Ilya Sutskever says the way AI is built is about to change.
• Building in the Era of Autonomous Software Development
• US Copyright Office rules out copyright for AI created content without human input

January 2025

• LLMs May Have a Killer Enterprise App: ‘Digital Labor’ — at Least If Salesforce Agentforce Is Any Indicator
  While this is a bit hyped and reliability of agents is going to be a big question, it’s still indicative of what’s happening.
• Salesforce Closed 200 Deals for Agentforce in Just One Quarter
  The company now expects to bring in $37.8 to $38 billion in revenue this fiscal year, an 8-9% increase over the previous year, largely due to AI product sales.
• Benioff: Salesforce Customers Will Deploy One Billion AI Agents Next Year
  Benioff predicts AI agents will allow companies to have an unlimited workforce.
• A System of Agents Brings Service-as-Software to Life – $4.6 Trillion Opportunity as AI Transforms Software
• What Can Generative AI *Actually* Do Well?
  Starts a bit overly skeptical but ends with a deep dive into different use cases, including Generation, Transformation, Summarization, Decision, Comparison, Retrieval, Digital Navigation, Physical Navigation, and Behavior Simulation.
• Example of Generative AI ROI – Supply Chain Document Processing
  This sounds similar to what we’re doing at Affineon, but in our case, the workflow is a doctor’s inbox.
• Ben Evans – AI Eats the World
  Another amazing presentation by Ben Evans. Some key points: Is the only moat for LLMs capital? AI gives you infinite interns. And many more.
• Menlo VC – State of Generative AI in the Enterprise
• HBS – AI Companions Reduce Loneliness
• Recommendations on How to Use AI from an Executive
• Slack Survey – Reality Check on AI Usage at Work
  61% of desk workers have spent less than 5 hours learning AI. Nearly half of desk workers would be uncomfortable telling their managers they used AI.
• The AI Services Wave: Lessons from Palantir in the New Age of AI
• ByteDance Lays Off Hundreds of TikTok Employees in Shift to AI Content Moderation
• Why Aren’t More Workers Using ChatGPT?
  Great breakdown of bottom-up strategies to help adoption. Paradoxically, most people don’t have time to figure out how they can save time.
• Wharton – AI Adoption Report
  A survey of over 800 senior business leaders found that weekly Gen AI usage nearly doubled from 37% in 2023 to 72% in 2024.

October 2024

• HBR – AI Can (Mostly) Outperform Human CEOs
  A fascinating comparison of human vs. AI. “Rather than fully replacing human CEOs, AI is poised to augment leadership by enhancing data analysis and operational efficiency, leaving humans to focus on long-term vision, ethics, and adaptability in dynamic markets. The future of leadership will likely be a hybrid model where AI complements human decision-making.”
• AI-Powered Invention Machine
  “The software’s primary purpose is to scan the literature in both the company’s field and in far-off fields and then suggest new inventions made of old, previously disconnected ones.”
• Generative AI’s Act 01
  Two years into the Generative AI revolution, research is progressing the field from “thinking fast”—rapid-fire pre-trained responses—to “thinking slow”—reasoning at inference time. This evolution is unlocking a new cohort of agentic applications.
• AI in Organizations: Some Tactics
  People are using AI at work. They are seeing productivity gains. Yet, AI use that boosts individual performance does not always translate to boosting organizational performance for a variety of reasons. So how do you do R&D on ways of using AI?
• HBR – Gen AI Makes Legal Action Cheap — and Companies Need to Prepare
  I’ve seen a bunch of startups targeting either compliance or identification of litigation. Interesting times.
• AI Tool Cuts Unexpected Deaths in Hospital by 26%, Canadian Study Finds

September 2024

Business

• Evolution of SaaS Pricing with AI
  Seats purchased is going down as AI is added – what to do?
• AI at Work is Here – Now Comes the Hard Part
  Great study by Microsoft
• AI’s Impact in Hospitality and Travel
  Good ideas for every industry
• AI’s Impact on BPOs is Real
• Small Teams, Big Impact: How AI Is Reshuffling The Future Of Work?
• How Dwarkesh Patel Uses AI
• AI Summer – LLMs might also be a trap: they look like products and they look magic, but they aren’t.
• Customer Research Using AI

Technical

• GPT 5 – Everything You Need to Know
  A very long article, but worth the read if your browser can handle it
• Auto and Meta Evaluation
• LLM as Judge
• Building a Generative AI Platform
• Lynx Hallucination Detection
• OpenAI Structured Outputs
  finally reliable JSON output
• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
  General-purpose zero-shot prompt to rate responses from an LLM to a given question on a scale from 1-10. They find that GPT-4’s ratings agree as much with a human rater as a human annotator agrees with another one (>80%)

July 2024

Business

• State of GenAI in SaaS Startups
• AI Productivity Gains
• State of AI in Production
  AI adoption is still early stage. Investment continues to grow.
• AI Voice Agents – State of Market
• C3.AI Earnings Call
  C3.AI gave the example of one of their customers — DLA Piper, a top law firm —moving tasks as knowledge-intensive as due diligence of contracts to AI and gaining an 80% reduction in effort.
• Klarna AI assistant handles two-thirds of customer service chats in its first month
• The jobs being replaced by AI – an analysis of 5M freelancing jobs
  The number of writing jobs posted on Upwork have declined by 33% since the arrival of ChatGPT.

Technical

• What We Learned from a Year of Building with LLMs (Part I): Technical
  Great series of articles
• What We Learned from a Year of Building with LLMs (Part II): Product and Team
• What We Learned from a Year of Building with LLMs (Part III): Strategy
• LLM From the Trenches: 10 Lessons Learned Operationalizing Models at GoDaddy
• What is an Agent
• Levels of AI Agents
• Mixture of Agents
• Why we no longer use LangChain for building our AI agents
• Cost Of Self Hosting Llama-3 8B
  It’s more expensive unless you do it on your own hardware, but even then it’s a 5 year payback period.

May 2024

Business

• Economic Impact of Generative AI
• Impact of Generative AI on Work
• AI leads a service-as-software paradigm shift
• Unicorn Club – Enterprise SaaS Leads the Way
• ML, AI and Data (MAD) Landscape
• Every federal agency now needs to have a chief AI officer
• AI Startups Require New Strategies
• Way Enterprises Are Building and Buying Generative AI

Technical

• RAG and RAU: A Survey on Retrieval-Augmented Language Model in Natural Language Processing
• Agent strategies to improve results
• Agent design strategies
• Unusual Prompts
• Offering a tip to LLM improves results
• Prompting structure and setting up the LLM for success
• Generating and using rationales to improve response quality
• Multi Agent Debate
• Meta Prompting
• Meta-Reasoning over Multiple Chains of Thought
• Deceiving to Enlighten: Coaxing LLMs to Self-Reflection for Enhanced Bias Detection and Mitigation
• RAG Complexity and Evaluation
• Great overview of Evals
• Search-Augmented Factuality Evaluator (SAFE)
• Fine grain, atomic evaluation of factual precision in long form text
• Evaluating LLM answers
• Taxonomy of hallucinations
• Divide and conquer evaluations of larger outputs to look for hallucinations / consistency
• Information at beginning of context window is lost when larger context are passed in
• Lost recall in large context windows
• LLM as Optimizer of Prompts
• Automated optimization of prompts
• Lowering LLM costs