Mike Kaput

Three years ago, Google declared "Code Red" in response to ChatGPT. Last week, OpenAI declared its own Code Red in response to Google. The tables have fully turned. Google's releasing models that beat OpenAI on reasoning benchmarks. OpenAI is delaying products to refocus on core improvements. In AI, three years is an eternity. Leads evaporate. The company that looked untouchable in 2022 is playing defense in 2025. We break down what's happening, plus new data on AI job cuts and why 2026 might be rougher than expected, on this week's Artificial Intelligence Show. podcast.smarterx.ai/shownotes/184

PromptGPT: chatgpt.com/g/g-wZ6j6NxQQ-…

Someone asked me for my best advice on how to train custom GPTs, and it may be surprising: There are, of course, best practices for building GPTs that increase the odds of success. The quality of the prompt or instructions matters. The type of examples, knowledge, and context you give it matters. Hell, what you are trying to get the GPT to do in the first place matters. But the most potent strategy I’ve found to get the best results consistently is: Rapid iteration. I use a separate prompt generator GPT to create an initial prompt for me based on what I want to do. I make that prompt the instructions of whatever GPT I am trying to build. I use the GPT. It breaks or screws up. I tell the prompt generator GPT what went wrong. It rewrites the prompt to fix it. I update the GPT I’m building with the new prompt. Then I repeat the process. Over and over again. Every time, I feed what went wrong back into the prompt generator. Then I have it draft a new prompt. Then I update the GPT. And try again. That’s largely it. It sounds like a lot, but it doesn’t take long. You can go through a dozen iterations in a handful of minutes. And, often, the GPT you have after a dozen iterations is miles better than the one you started with. Which makes it time well spent. I link to PromptGPT, which is the prompt generator I built for this process, in the next post. (But you can use any type of prompt generator to do this for yourself.)

Here’s a truth about AI you might need to hear today: You don’t need to drink the whole firehose. You don’t need to… Catch every headline. Read every update. Test every tool. It’s impossible. And, frankly, it will exhaust you. Sure, stay as informed as possible. Spend tons of time experimenting. Try everything you can. But remember: The real power isn’t in keeping up. It’s in stepping back. Today, we KNOW that a few things are true… And will continue to be true for the time being: 1. We now have robust intelligence on demand. 2. Progress in its capabilities is not slowing down. 3. The cost of it is dropping—fast. THESE are the ideas that are going to reshape your job, your business, and your life. From these alone, you can start to build a decent estimate of where your future is headed. So, return to them often. Treat them as North Stars. You’ll find it a LOT easier to get ahead. Even if you’re not always keeping up.

Want to crush your next speaking engagement? I created a GPT that helps you do just that: I’ve delivered 100s of talks and presentations, and even I forget sometimes the golden rule… HOW you talk matters as much as WHAT you say, so warming up your voice helps a lot. But sometimes I forget to warm up… Or don’t have good warm up steps ready to go. So I created Voice Coach GPT to help. Here’s how it works: 1️⃣ First, it walks you thru a quick exercise to loosen up your neck, shoulders, and jaw. 2️⃣ Then, it guides you thru a few rounds of breathwork to activate your diaphragm. 3️⃣ Finally, it takes you thru a few vocal warmup exercises to enhance resonance and articulation. (It’ll even go through a little exercise with you to get your energy up before the talk.) And it all takes only a few minutes to use. Check it out here: chatgpt.com/g/g-67978b5bea…

Here’s a workflow I use to get tons more value out of Google @NotebookLM: 1. Use @OpenAI o1 to generate a research plan on a topic I want to learn more about. 2. Give the plan to Deep Research in Google @GeminiApp then have it create a research brief. 3. Export the research brief to a Google Doc and add to NotebookLM as a source. 4. Add all the links from the brief as sources to NotebookLM. Once everything’s in NotebookLM, then do the following in order: ✅ Create an Audio Overview to get a high- level intro to the topic. ✅ Select only the brief from Deep Research. Read, summarize, query it using NotebookLM. ✅ Select all sources and use NotebookLM’s presets to create guides, timelines, FAQs, etc. ✅ Ask smart questions of all the material. (Use o1 with your brief to brainstorm great ones.)

Maybe a controversial take…but you can get better at AI than 90% of people in one week: Day 1: Identify the top 3 things you spend the MOST time on in a day. (Spend time on this; go deep.) Days 2-4: Build 1 GPT per day for each of your 3 tasks. Test and iterate until they perform well enough. Day 5: Go deep on Deep Research in Google Gemini to dramatically improve how you do research. Day 6: Go deep on using Google NotebookLM to dramatically improve how you learn new topics. Day 7: Create a workflow to consistently integrate GPTs, Deep Research, and NotebookLM into your day. Bonus: Use ChatGPT’s Advanced Voice Mode constantly as a live assistant while you work. This takes 7 days and costs $40. In return, you’ll: - Make your most important work faster & easier - 10X your research (which improves other work) - 10X your learning (which improves other work) You don’t need permission… Anyone, at any skill level, can go do this now.

@Google has built something that finally starts to address a huge problem with AI: It’s called FACTS Grounding, a benchmark that measures how well AI sticks to the truth. Until now, we’ve had no reliable way to measure AI’s biggest weakness: making stuff up. Now, FACTS measures how well LLMs ground responses in provided source material. FACTS tests each model on 1,700+ docs, judging how well it answers using ONLY info given to it. The clever part? 🤔 Each answer gets checked by 3 AI models: - Gemini 1.5 Pro - GPT-4o - Claude 3.5 Sonnet Think of it as a peer-review system... but for AI truthfulness. You can see the results on a public leaderboard. (Gemini 2.0 currently leads with 83.6% accuracy) Will hallucinations become a thing of the past? We’ll see—but this is a step in the right direction.

@daniel_mac8 Thanks Dan! Such a cool project. Thanks for sharing it

@MikeKaput love this perspective Mike

This is an extraordinary AI use case… AI that synthesizes ancient wisdom: @daniel_mac8 gave o1 Pro different religious and philosophical works to synthesize… Then asked it to make connections between the ideas that haven’t been made or aren’t obvious. The works used include: the Christian Bible, Plato, Buddhist texts, Hindu texts, and Nietsche. It’s a really interesting read. (Link in comments.) But more exciting are the bigger implications: We’re quickly approaching AI that can functionally handle almost infinite context… And much of the ancient wisdom we have is written down in spiritual texts or interpretations… So what is stopping us from giving ALL of it to AI and synthesizing the common truths and ideas? I’ve already gotten immense, lasting value out of using AI in my personal life to: - Summarize decades worth of personal journals - Synthesize different personality tests - Interpret my behavioral and thought patterns We always say that businesses need to leverage the data they already have to get value from AI… Why not do the same thing with the wisdom that’s been passed down for 1000s of years? After all, that’s data too… In fact, it may be the most valuable data there is.

Full article with the advice here: latent.space/p/o1-skill-iss…

OpenAI’s o1 doesn’t work like typical ChatGPT. You need a totally different prompting approach: @latentspacepod published a guide that breaks down the anatomy of a good o1 prompt. The key? You need to think of this as a BRIEF, not a typical prompt. Here’s how: DO ✅ DO give it a TON of context. Whatever you think is a ton: “10X that,” they say. ✅ DO focus on the WHAT, not the HOW. Write out the goal, not the steps to achieve it. ✅ DO treat this as a “one-shot” prompt. Give o1 every single thing it needs in one go. DON’T 🛑 DON’T use it without knowing what you want. It doesn’t do well with vague or iterative prompts. 🛑 DON’T give it a role or instructions. Give it goal, format, context, then let it cook. 🛑 DON’T chat back and forth with it. Treat it the opposite of ChatGPT.

Had to laugh at this one from Google. NotebookLM’s AI hosts developed ATTITUDE: NotebookLM can create Audio Overviews of the docs, links, papers, and sources you give it. These Audio Overviews are basically mini podcasts MC’d by two hyper-realistic AI hosts. Recently, Google added a feature where you can “call in” to ask questions while the hosts talk. Turns out, the AI hosts did NOT like that: They got annoyed at users when interrupted. Google had do “friendliness tuning” to fix it. (Yes, they had to adjust an AI's attitude.) In all seriousness, this is a perfect reminder: AI is NOT the typical software you’re used to… After all: When’s the last time you got any lip from Excel? 😂

I just turned a voice memo into this post… Using a really helpful AI workflow: 1. Record a Voice Memo on iPhone… (Don’t fret about being polished—just get it out.) 2. Copy the transcript… This is auto-generated by the Voice Memos app. 3. Feed the transcript to a custom GPT… I trained mine to write just like I do on LinkedIn. In seconds, AI will organize and structure your raw thoughts into the start of a great post. The magic? This workflow helps you: - Capture ideas anywhere, anytime - Transform casual thoughts into content - Create while doing other tasks - Break through creative blocks - Actually start posting consistently But here’s the most important step: (So don’t skip it!) Take what AI gives you… …and WRITE THE POST YOURSELF. AI is the content assistant… YOU are the voice (and soul) of the content. The human behind the content is the only thing that makes it worth reading in the first place.

@jennyplant Glad to hear!

@MikeKaput Thank you this has given me a huge 💡 moment for how I can use this 🙏🏻

Here’s a simple, valuable AI use case, especially for marketing and sales: I often demo tools, features or workflows. I share my screen & show how something works. To do that, I have to learn the thing myself… Which means lots of using and documenting: - Menus options and choices - Features and outputs - Ways of interacting with tools and processes (Basically, I do all the stuff I want to show/teach to a prospect, teammate or customer.) Now here’s the use case: Record all of your work and give AI the video. Record yourself using a tool or doing a task, then have AI write out a full recap of the video. In seconds, you’ll get a complete description of everything you did, step by step… …Which you can then use as a script to demo/ teach it—or a checklist someone can use to do it. Use these steps to do this easily (and for free): 1) Record yourself doing something on-screen 2) Go to aistudio[dot]google[dot]com 3) Upload the video and ask for a summary

This stat on AI’s impact on jobs made me sit up and pay attention: 40% of employers anticipate reducing their workforces where AI can automate tasks. (This comes from a new Future of Jobs Report from the World Economic Forum.) To get this data, WEF surveyed 1,000 top global firms representing 14 million workers. 40%! That’s a huge number. 4 in 10 employers think they’ll need to cut jobs. Of course, they’re looking to upskill and hire for new AI skills…but there’s a catch: They also say their biggest barrier to AI adoption is a lack of the right skills. Can existing workers upskill fast enough? Do enough new hires even have the right skills? Every company needs to be solving for this… Which is why we tackle it in this week’s podcast. Episode 130 of The Artificial Intelligence Show just dropped—link below 👇

There’s one thing that makes a great prompt… And it’s something that most people miss: Great examples of what you’re trying to do. In other words: 𝐃𝐨𝐧’𝐭 𝐣𝐮𝐬𝐭 𝐭𝐞𝐥𝐥 𝐀𝐈 𝐰𝐡𝐚𝐭 𝐲𝐨𝐮 𝐰𝐚𝐧𝐭—𝐒𝐇𝐎𝐖 𝐢𝐭. Want a blog post in your voice? SHOW it your top 5 blog posts. Want a smart strategy document? SHOW it the ones you like best. Want to solve a tough problem? SHOW it how strategists think through problems. Sure, crafting a robust prompt matters… But examples turn good prompts into great ones. Yet most people don’t take the time to give them. Which is a shame because: Two people can use the same prompt but get wildly different outputs… Because one used examples and one didn’t. Act accordingly.

@brucefloyd Glad to hear!

@MikeKaput I’ve been using your prompt builder and I love it so far!

If you’re trying to do anything with AI, I’ve got some bad news for you: