Credit: NASA
In this edition:
My thoughts
Notable headlines
Learning, Tools, and Experiments
AI MISTAKES
My Thoughts
On May 25, 1961, President John F. Kennedy stood before Congress and proposed that the US "should commit itself to achieving the goal, before this decade is out, of landing a man on the Moon and returning him safely to the Earth."
This was an incredible challenge, met with an even more astonishing effort to bring it to reality.
In less than 10 years, on July 24, 1969, the Apollo 11 mission splashed down in the Pacific Ocean, answering the President's challenge. Itās the story of engineering prowess, scientific exploration, and man's triumph over incredible challenges.
We worked like mad to make it happen. We put 12 men on the moon.
And then we stopped.
We never went back.
Itās been over 50 years.
We lost the vision.
I recently watched a documentary about the Apollo missions, released in 2019, When We Were Apollo.
Near the end of the documentary, one of the people interviewed explained the problem:
We had āno good vision of what to do next.ā
Ok, I hear you saying, āWhat does all this have to do with AI?ā
Vision is everything.
The will to act is what drives us, and AI is becoming increasingly capable of the "actionā part.
As humans, we have the distinct ability to dream. In fact, I would argue it is not only an ability, but an obligation.
While we have given computers the ability to speak and do things, the dream still has to come first.
Just as the Moon landing inspired a generation to look upward andā¦
āā¦do the other things, not because they are easy, but because they are hard, because that goal will serve to organize and measure the best of our energies and skills, because that challenge is one that we are willing to acceptā¦ā āJFK
AI has the potential to captivate and enable minds. To inspire the next wave of thinkers and creators.
Itās making things that were only dreams of the past not only possible but real.
So, the question I have for you is:
āWhat is your DREAM?ā
This question is more complicated than you think. It will likely take time to answer this question, but you must act once you know the answer.
Once you read this, my challenge for you is to act.
What action must you take? Speak your dream into existence.
What would happen if you spoke that dream into a session with ChatGPT (or Claude, or Grok, or Meta, because they all talk back now, too)?
I know, it still seems a little crazy to just openly talk to a computer. It still feels a bit strange even to write this idea here.
Talk to an AI. Share your dream with advanced AI like Anthropicās Claude 4 or OpenAIās o3.
Speaking about your dream changes your mind, but speaking about it, even to an AI, could help you realize your dream.
The sooner you speak it, the sooner you can refine it and iterate with rapid feedback.
But donāt stop at speaking. With the latest AI capabilities from the likes of ElevenLabs, Midjourney, RunwayML, and the new Veo3, you can not only see your dream in text, but also hear it, see it, and watch it.
This is the superpower that AI gives us.
If youāre worried someone will see your mistakes, worry no more.
Let go of the fear of MISTAKES and start making progress today.
Use the MISTAKES framework as well, for help.
And if at any stage, after you recognize and get clarity on your dream, you need someone to listen, I would love to hear from you and cheer you on.
Until then, hereās more on whatās notable in AI that might help kindle your dreams.
Notable Headlines
Google Gemini 2.5 Pro
Google's latest Gemini model upgrades interactivity, code generation, and workflow automation.Veo3 from Google
Googleās new Veo3 elevates text-to-video generation with cinematic fidelity and fast rendering for creators and studios.Claude 4 by Anthropic
Claude 4 brings long-context reasoning and strong tool use to the forefront, raising the bar for enterprise-grade AI performance.OpenAI Codex
OpenAI announces a major overhaul to Codex, integrating newer models with tighter IDE support and team productivity features.OpenAI Retiring GPT-4.5 in āmonthsā
OpenAI has shared they intend to deprecate GPT-4.5, consolidating its offerings around GPT-4o and the forthcoming GPT-5 lineup.Meta AI Hits 1B Users
Meta's AI assistant sees massive adoption across its ecosystem, including WhatsApp, Instagram, and Messenger.Opera Neon: The AI Browser
Opera launches an "agentic browser" that can research, automate tasks, and even code.Anthropic's AI for Science Program
Claude is being used to accelerate discovery in biology and life sciences with a new research partnership program.DeepSeek R1 Model Updates
DeepSeek releases an update to R1. No news on a ācleaned-upā western version yet.Perplexity Labs Debuts
A new workspace from Perplexity turns simple queries into rich reports, dashboards, and apps.Godela for Engineering Design
This physics-trained AI model solves real-world engineering problems with unmatched speed and accuracy. You really can ābuildā with AI now.
Learning, Tools, and Experiments
If you havenāt yet, speak your idea into one of these tools and build something, even small.
Auto-generates PDF research reports with citationsāperfect for business briefings. This is great for getting a quick document, and I also use it to accelerate working across multiple AI tools like Anthropic, Grok, and Meta AI.
Veo3 ā Googleās Cinematic Video Model
Generates high-quality videos from text prompts with cinematic style and advanced control. Iām collaborating with an industry insider for an upcoming YouTube video. (Donāt forget to subscribe!)
Anthropic dropped new models and added voice mode. Go try it out!
A Few Tools I Regularly Use
Some of you have asked me about the tools I regularly use to share content and manage my work:
Descript: AI Video Editor
Perplexity Pro: My default go to for any question and some serious research
Gamma.app: Quickly make a presentation on any topic with ease.
Typeset: Build presentations, documents, social media content.
StreamYard: Live Video Streaming to multiple platforms.
Keep Learning
A comment on X (formerly Twitter) I saw earlier this month reminded me of this:
Compared to the Google days when coming up with a good search keyword was an important skill, coming up with a good prompt is a 100 times more difficult and 100 times more powerful. Which means AI won't get rid of education but will reward you even more for educating yourself. ā @cocktailpeanut
There are many prompt courses available, and I even created aĀ video on āIs Prompt Engineering Dead?ā last year; however, there is still value to be gained from learning this core skill.
If you arenāt using DSPy, and you havenāt already adopted one of the methods I shared in my prompt engineering presentation from 2023, you can pick up one of the courses offered by Microsoft, Anthropic, or OpenAI.
If you spend a significant amount of time using AI tools, it never hurts to step up your game to ensure you are using good-quality input.
AI MISTAKES
Any time we push the boundaries of whatās possible, we havenāt truly done the work well until we reach a point of failure. There were some real AI breakthroughs over the last month, but there were also some real issues.
At the end of April, OpenAI updated their latest GPT model, and suddenly, it seemed everyone was complaining about how it seemed to be choosing agreement over accuracy.
A few days later, OpenAI admitted that their latest model updates included some AI MISTAKES.
In a May 2 post, Expanding on what we missed with sycophancy, they shared a bit more detail on what happened and what was learned.
This launch taught us a number of lessons. Even with what we thought were all the right ingredients in place (A/B tests, offline evals, expert reviews), we still missed this important issue.
I applaud the OpenAI team for recognizing the problem, taking steps to reverse the change, and then publicly sharing what theyāve learned.
Remember the MISTAKES framework:
Map, Investigate, Sketch, Test, Adapt, Kickstart, Evaluate, Sustain
This was a real-world example of what the āEā and āSā in the MISTAKES framework are all about: Evaluate what went wrong, and Sustain the learning publicly.
This situation has elevated my respect for how OpenAI has been operating after the Sam Altman leadership challenges last year, and I was pleased to see how they handled it.
From my perspective, all the frontier model builders have improved in how they handle model behavior problems over the last year. However, they must also continue to share publicly what they are learning. This is both the principle of sound engineering and the value of AI MISTAKES.
Letās make June the month of bold vision, shared dreams, and forward action.
Go speak your dream. Build with AI. And make some meaningful AI MISTAKES.
If youāre enjoying my newsletter, do me a great favor and share it with someone that could use this valuable information for their own benefit.