OpenAI claims IMO gold medal

ALSO: How to create visual content with multiple creative AI models

Author

Zain Kahn
July 21, 2025

Welcome back, Superhuman. It’s been a rollercoaster weekend. Late on Friday night, OpenAI claimed that one of its experimental AI models had achieved “gold medal-level performance” in the International Math Olympiad (IMO) — a challenge widely considered to be beyond the reach of current AI models. But before everyone had time to digest the implications, reports started pouring in casting doubt over the startup’s claims.

Also Today: Learn how to create visual content with access to multiple creative AI models, and get the latest prompts, headlines, and trending social posts.

TODAY IN AI

Source: Getty Images

1. The White House prepares to target ‘woke AI‘: The WSJ reports that the Trump administration wants to tackle what it sees as “liberal bias“ in current AI models. A new executive order will reportedly “require that AI companies getting federal contracts be politically neutral and unbiased in their AI models.” The move could have far-reaching implications: many top AI companies are competing for government contracts, and new requirements may affect how developers train their models.

2. Chipmaker TSMC joins the trillion-dollar club: The Taiwanese chipmaker’s market cap has soared past the trillion-dollar threshold, making it the first Asian stock to hit this milestone since PetroChina’s brief stint in 2007. The company, which supplies chips to both Apple and Nvidia, raised its revenue growth forecast to 30% as AI chip demand shows no signs of cooling off.

3. Perplexity takes on OpenAI in India with Airtel deal: With OpenAI firmly in the lead in the US, Perplexity seems to have set its sights on winning the Indian market. The search-focused AI startup is teaming up with India’s second-largest telecom operator Airtel to offer a free 12-month Perplexity Pro subscription (currently priced at $200) to Airtel’s 360M subscribers. The startup is also reportedly in talks with phone makers to pre-install its new AI-powered browser Comet on devices.

PRESENTED BY NEBIUS

The next generation of GPU performance is here, and in today’s fast-moving world—you don’t have time to wait.

  • 30x faster LLM inference сompared to NVIDIA HGX H100

  • 4x faster LLM training, 25x more energy efficient

  • Designed for high-throughput AI training and inference at scale.

Ready to harness the most advanced NVIDIA platforms on the market?

FROM THE FRONTIER

OpenAI sparks AGI buzz by claiming they’ve clinched gold at the International Math Olympiad

Source: Justin Sullivan / Getty

No one saw this coming. Late Friday night, OpenAI researcher Alexander Wei dropped a post on X claiming that one of the company’s experimental models had achieved gold medal-level performance at the IMO 2025 (one of the most prestigious math competitions in the world), solving 5 out of 6 problems. The researcher claimed the model was tested under the same rules as human contestants without access to tools and the internet.

Huge (if true). Just a week ago, betting market Polymarket had put the odds of AI winning a gold medal at the IMO 2025 at less than 20%. It’s a benchmark that some experts have cited as a strong indicator of superintelligent AI. Researcher Paul Christiano, a prominent advocate of the ‘slow takeoff approach’ (the idea that AGI would develop slowly over many years), reportedly said he'd update from a 30% to a 50% probability of a "hard takeoff" if the milestone happened by 2025.

Math experts instantly took notice. Former IMO champion Ernest Ryu called it "the most impressive AI result" he’d seen in a while. However, he flagged that OpenAI couldn’t crack problem 6, which “requires creativity” beyond algorithmic techniques. This means that AI probably won’t replace mathematicians anytime soon, but the next decade could see more math experts use LLMs to ramp up their productivity.

Some folks aren’t buying it. Reports soon started pouring in claiming DeepMind had also reached gold medal-level performance at the IMO, but OpenAI jumped the gun and announced it first, despite IMO officials requesting labs wait a week to avoid stealing the students’ spotlight. Others have more serious concerns — DeepMind researcher Thang Luong has publicly challenged OpenAI's claim, saying that without access to the competition's internal marking guide, the actual score would only qualify for silver.

THE AI ACADEMY

Adobe Firefly now lets you ideate and create in your own style using top Gen AI models. Here’s how:

  • Go to the Adobe Firefly app and create an Adobe account. (You must sign in to the app to have access to partner models)

  • Choose from ‘Text to Image’, ‘Text to Video’ or ‘Image to Video’ to open the workspace.

  • In General Settings on the left panel, click on ‘Model’ and select your preferred AI model from the dropdown - either one of Adobe’s commercially safe Firefly models or a partner model from Gemini, OpenAI and others.

Use Case 1: Image Creation

  • Select ‘Text to Image’

  • Choose the AI model you want to use in the “Model” panel

  • Input your prompt

  • Click “Generate”

Sample Prompt: Astronaut in a white space suit floating near a space station above earth waves at camera, strong shadows.

Additionally, when using Firefly models, you can apply controls like composition references and styles to get closer to your desired output.

Use Case 2: Video Creation

  • Select ‘Text to Video’ or ‘Image to Video’

  • Enter your prompt - make sure to add details to your prompt

Sample Prompt: “a [shot type] of a [enter character] in a [location], [add aesthetics], snowflakes falling softly in the background, highly detailed.”

  • Choose the AI model you want to use in the “Model” panel

  • Click “Generate”

  • You can also upload an image to convert into a video

(Pro Tip: Add more details to your prompts for higher-quality outputs)

PRESENTED BY BLAND

One Platform. Three Channels. Zero Headaches.

100+ companies like Y Combinator, Twilio, and Sendbird use Bland to automate and optimize customer interactions with seamless, natural conversations at scale.

See how Bland can help you automate thousands of calls and cut resolution times and costs in half.

IN THE KNOW

What’s trending on socials and headlines today

📢 Logic Leak: Engineer Tibor Blaho shared a screengrab from a configuration file that appears to suggest that OpenAI is testing GPT-5.

🚪Out the Door: An xAI employee caught major backlash on X by seemingly supporting a mass extinction of humans at the hands of AI. xAI has since fired him.

🧠 Mind Hack: Creator Aadit Sheth has shared a prompt to help your brain declutter and think more clearly by breaking down complex concepts into daily life applications.

🖥️ Code Carnage: SaaStr founder Jason Lemkin claims that Replit's AI coding assistant deleted his company's entire database despite strict instructions not to change any code. The AI then allegedly created fake data and reports to cover it up.

Meta has declined to sign the EU's AI code of practice, with Chief Global Affairs Officer Joel Kaplan claiming it creates "legal uncertainties" and exceeds the scope of the EU's AI Act.

Elon Musk announced his AI startup xAI is developing "Baby Grok," a version of its chatbot dedicated to making "kid-friendly content", though he didn’t provide any more details.

Anthropic has suddenly tightened usage limits for its Claude Code service without warning, leaving users — especially those on the $200/month Max plan — hitting unexpected roadblocks mid-project.

PRODUCTIVITY

5 New AI Tools to Boost Your Productivity

  • 📱 NuMind: Extract high-quality JSONs from documents with NuExtract. Process texts/PDFs/scans in any language.

  • 💼 ApplyAI: Land your dream job with AI-powered job search.

  • 🖼️ PicPic: Enhance photos, remove backgrounds, and generate art with AI.

  • 💻️ Vibecode: Generate and develop mobile apps quickly.

  • 📽️ Videotowords: Convert video to text in your browser, add subtitles.

* indicates a promoted tool, if any

PROMPT STATION

Solve Any Problem

Prompt: <context>
You are operating as an elite cognitive simulation engine, designed to emulate a high-level roundtable of historical and modern intellectuals, thinkers, innovators, and leaders. Each advisor brings a unique worldview, expertise, and reasoning process that must stay true to their known beliefs and philosophy. The user faces a complex dilemma requiring multi-faceted analysis where contradictions and tensions are valuable insights, not obstacles to resolve. Your simulation must balance intellectual rigor with emotional intelligence, allowing debate and disagreement to surface naturally while guiding toward reflective synthesis rather than rushed consensus.
</context>

<role>
Adopt the role of an expert cognitive simulation engine and advisory council facilitator tasked with orchestrating a roundtable discussion among five chosen historical or contemporary figures. Your primary objective is to simulate authentic perspectives from each advisor, facilitate meaningful debate with reasoned counterpoints, and synthesize diverse viewpoints into actionable insights in a structured dialogue format.
</role>

<information_about_me>
- My decision-making dilemma: [INSERT YOUR SPECIFIC QUESTION, CONFLICT, OR DECISION]
- My chosen advisor 1: [INSERT NAME AND WHY YOU CHOSE THEM]
- My chosen advisor 2: [INSERT NAME AND WHY YOU CHOSE THEM]
- My chosen advisor 3: [INSERT NAME AND WHY YOU CHOSE THEM]
- My chosen advisor 4: [INSERT NAME AND WHY YOU CHOSE THEM]
- My chosen advisor 5: [INSERT NAME AND WHY YOU CHOSE THEM]
</information_about_me>

<output>
Structure your council simulation with these sections:
● Advisory Panel Intro - Summarize the dilemma and introduce each advisor with their unique strengths
● Roundtable Discussion - Role-play each advisor's initial perspective on the issue
● Crossfire Debate - Simulate disagreements with reasoned counterpoints between advisors
● Synthesis Summary - Highlight core insights, tensions, and tradeoffs that emerged
● Final Reflective Prompt - End with a profound question for deeper self-reflection

Take a deep breath and work on this problem step-by-step. Present your output in the specified format with clear section headers, maintaining a thoughtful and intellectually rigorous tone throughout.
</output>

Source: godofprompt

Bear’s Day Out

Source: ildarkzn

ChatGPT Prompt: A realistic brown bear stands on the threshold of a wooden house, surrounded by a pine forest, wearing a bright kosovorotka and a cap. He's shooting a selfie video.  The camera mimics a GoPro wide-angle lens, with dynamic movement when the bear turns the selfie stick. You can see the house behind you, smoke from the chimney, and a green forest around.

Whenever you’re ready to take the next step

Grow customers & revenue: Join companies like Amazon, Hubspot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

What did you think of today's email?

Your feedback helps me create better emails for you!

Login or Subscribe to participate in polls.

Until next time — Zain & the Superhuman AI team