o3 sparks AGI debate

Welcome back, Superhuman. Choosing an AI model is always a compromise. You usually have to pick between one that’s fast but limited, or one that’s smart but slow. But Google’s new Gemini 2.5 Flash has surprised the industry by unveiling a model that provides close to frontier performance for a fraction of the cost.

Today’s Insights

• Gemini 2.5 Flash, DeepSeek’s mounting challenges, and Grok’s memories

• Did o3 get us to AGI? There’s no easy answer.

• Tutorial: How to convert YouTube videos into blogs

• 5 new AI tools to boost your productivity

• News, memes, what’s trending on socials, and more

TODAY IN AI

1. Gemini 2.5 Flash is here, and it’s fast: Developers can now use a preview version of the ultra-efficient LLM in AI Studio and Vertex AI. It offers speeds that match 2.0 Flash, but with much more power under the hood that comes close to, or exceeds, many top models from rival AI companies. It’s also the company’s first hybrid reasoning model, meaning you can turn thinking on and off depending on your needs. If that wasn’t enough, the search giant also added more starter apps — programs you can experiment with and build off of — to AI Studio.

2. DeepSeek has a bumpy road ahead: The Chinese startup’s last major release, R1, sent the industry into a tailspin earlier this year. With R2 only weeks away, competitors are racing to prevent a repeat. OpenAI will now ask developers for an ID before they can access its tools — an effort to stop DeepSeek from swiping its secrets. And new reports suggest that even the US government is moving to cut off DeepSeek’s access to Nvidia chips and potentially even ban the platform altogether.

3. Grok finally gets long-requested feature: xAI’s chatbot can now remember your past conversations, helping you get “personalized responses” when you ask for recommendations or advice. For example, if you previously mentioned that you go to the gym every day, Grok will give you more expert-level suggestions for your next workout plan. Of course, you can always choose what Grok remembers and what it forgets.

PRESENTED BY PRODUCT FACULTY

Future-proof your career with the #1 Rated AI Certification for Product Managers

Become indispensable with the #1 Rated AI Product Management course: an 8-week, high-impact program led by Miqdad Jaffer, Product Leader at OpenAI.

Designed for ambitious Product Leaders or talented non-PMs, this hands-on course gives you skills to confidently build and launch real AI products.

• April’s cohort is nearly full (price rising soon)*

• No coding experience required

• Live online class sessions

• Rated 4.9/5 stars by past students

“Isn’t just theory... It emphasizes how to apply AI in real product environments—from ideation to deployment—ideal for PMs who want to go beyond buzzwords and actually deliver working LLM features.” – Riti, Senior PM

Register now and save $500 (before it sells out).

FROM THE FRONTIER

Have we reached AGI with o3? Experts are divided.

OpenAI’s o3 just gained the ability to use tools, sending it straight to the top of most major intelligence benchmarks. Those impressive results are fueling a debate around whether we’ve just reached AGI — AI that matches or exceeds human capabilities. But coming up with a definitive answer is a lot harder than you’d think.

• Economist Tyler Cowen is convinced. After testing out o3, he concluded that “on a vast array of topics and methods, it wipes the floor with the humans.”

• Model trainer John Hallman agrees. He can see “the light at the end of the tunnel” and has “felt for the first time tempted to call a model AGI.”

• Count immunologist Derya Unutmaz as a believer, too. Based on his experiences, o3 is already “at or near genius level.”

• But not everyone’s onboard. o3 still whiffs some pretty easy prompts, like matching names and colors or solving basic logic puzzles. Many experts say that’s reason enough to doubt any AGI claims.

These conversations are going to get murkier in the months and years to come. AGI might not arrive in one eureka moment. It could be a gradual process, with experts slowly coming around to the idea as the evidence mounts. One clear sign that we might be close is that GPT-4.5 recently passed the Turing Test, convincing participants it was a real human 73% of the time.

THE AI ACADEMY

How to convert YouTube videos into blogs

• Go to NotebookLM and click "Create new".

• Click "Add source" and upload your YouTube video link or transcript.

• Identify primary/secondary keywords and questions people ask on the topic.

• Prompt NotebookLM to create a blog post for you based on the video.

Sample Prompt: "Write a [enter word limit] word SEO-optimized blog post titled [enter title]. Start with a 100-word intro explaining why the concept confuses even experts. Use clear headings to organize sections into [add sections]. End with simple analogies or tips to make it easier to understand. Naturally include keywords like [enter keywords]. Keep the tone [enter tone].”

• Click "Generate" and review the draft for keyword use, structure, and accuracy.

• Once done, copy text, add images, break up long paragraphs, add internal/external links, insert a table of contents if needed, add a CTA, and include graphics or charts.

• Review everything one final time and publish your blog.

PRESENTED BY AISERA

Let AI Agents Solve Your Complex Employee Requests

Enterprises are facing an increasing volume of complex requests that require unified access to multiple systems.

With siloed infrastructures and workflows, users are frustrated. An enterprise-ready AI agent platform is critical—one that can understand, reason, and proactively fulfill requests.

Discover the strategic value of Agentic AI in the enterprise.

AI & TECH NEWS

Everything else you need to know today

👀 Seeing Double: Anyone using Microsoft’s Edge browser in the US can now try Copilot Vision, the tool that lets AI see everything on your screen, although it’s only available across nine websites to start out.

💰 Branching Out: Anthropic just made its first-ever investment in another startup. It gave $1M to Goodfire, a year-old company that is trying to figure out how LLMs make decisions.

📱 Query Companion: Perplexity is reportedly bringing its search-centric model to Motorola phones as a built-in virtual assistant. It’s also trying to beat out OpenAI and Google for an exclusive partnership with Samsung.

🏅 Keeping Score: LMArena, which hosts one of the most-cited AI leaderboards in the industry, announced it’s forming a company called Arena Intelligence. Its benchmark is unique in that it uses crowd-sourced ratings to rank models’ performance.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ Nily: An AI-assistant with 20+ AI models, LLM response comparison, and mixture AI for delivering the most optimal answer.

✅ DocsHound: Turns your product demo into instant documentation, chatbots, and onboarding content—auto-generated from a single UI capture.

✅ Docwelo: An AI tool that helps freelancers, founders, and small businesses get standard contracts like NDAs, employment, and service agreements done in minutes.

✅ Tool Finder: Find reviews, filter tools, search instantly, and discover the best tools for smarter work.

✅ Webdraw: Explore, remix, and build AI apps—no coding needed.

 🎁 Want more? Check out our Top 125 AI Tools

^{* indicates a promoted tool, if any}

PROMPT OF THE DAY

Prepare for interviews

Prompt: Act as a senior hiring manager in [insert industry or role — e.g., software engineering, marketing, finance] and help me prepare for an upcoming interview for a [insert job title] position. Ask me 8–10 realistic and challenging interview questions based on the job title, including both technical and behavioral ones. After each question, evaluate my answer for clarity, structure, and impact — and suggest improvements if needed. Also, give me tips on how to improve my delivery, body language, and confidence. Based on the job role, share common red flags candidates often trigger and how I can avoid them. Finally, provide one polished, high-impact version of my answer to the “Tell me about yourself” question that ties my background directly to the role I’m applying for.

🤓 Want more? Check out our Top 1,000 Prompts

SOCIAL SIGNALS

What’s trending on socials today

🧠 Same Page: Nineteen-year-old builder Dhravya Shah created a “universal memory” MCP that lets you carry all of your past chats across different LLMs — so each of them knows the same things about you.

🔮 Crystal Ball: OpenAI’s Aidan McLaughlin shows how to do forecasting with o3 — for example, estimating the likelihood that Stanford refuses federal compliance in line with Harvard.

🌎 Geography Nerd: The latest viral trend sweeping ChatGPT: Using it to do a reverse image search (i.e. asking it to find the exact location where a photo was taken).

🎾 Brand Ace: Designer Victor Yuyang Luo shows how he used Runway’s Gen-4 to create a sleek brand story for an imaginary tennis company called Dropshot.

AI-GENERATED IMAGES

Prompt or Paintbrush?

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 1M+ readers and 2M+ followers on socials working at the world’s leading startups and enterprises. Companies like Amazon, Hubspot, and Salesforce feature their products in Superhuman. You can learn more about partnering with us here.

🧞 Your wish is my command

Got more feedback or just want to get in touch? Reply to this email and we’ll get back to you.

Thanks for reading.

Until next time!

Zain & the Superhuman AI team