Welcome, humans.

Now that it has a much stronger memory, ChatGPT is apparently judging all of us based on our chat history—and the results are hilariously savage.

The prompt? “Generate an image of what you think I look like based on our entire chat history.” The results? Let's just say ChatGPT has some... opinions.

Example: this woman who got “done dirty” when GPT transformed her into a bearded dude with glasses—though it did spark an impromptu family reunion as dozens of users with identical AI portraits realized GPT thinks they’re each other's long-lost relatives.

ChatGPT isn't just roasting our physical appearances—it's psychoanalyzing our inner minds too. When asked to “visualize the inside of my mind”, the results ranged from cotton-candy dreamscapes to trauma-core warning labels.

Here’s what you need to know about AI today:

• 21 robots raced thousands of humans in China.

• OpenAI and SoftBank explored taking Stargate international.

• The UEA launched the world’s first AI to modernize gov regulation.

• Data shows AI Overviews lead to 34.5% lower CTRs.

Robots hit the track (and sometimes the fence) in China's first half-marathon race for humanoids

Remember when you had to run a mile in gym class and felt like your legs might fall off? Now imagine being a robot trying to complete a half-marathon—and your legs actually could fall off.

This weekend, 21 humanoid robots lined up alongside thousands of human runners in Beijing for what's being called the world's first robot half-marathon.

Here’s the deets:

• The 13-mile race served as a real-world stress test for China’s robotics industry.

• Race had 3.5-hour cutoff, requiring robots to maintain at least 3.7 mph.

• Beijing aims to be a world leader in humanoid robots by 2027, and not only that…

• The government is offering subsidies, talent bonuses, and tax breaks to robot companies—so the race was super high stakes.

So how did the robots do? Well, let's just say they won't be qualifying for the Olympics anytime soon.

• Only two robots (Tien Kung Ultra and Little Rascal N2) completed the course on time.

• The race's robot champion, Tien Kung Ultra (a 5'9”, 115-pound humanoid), crossed the finish line after 2 hours, 40 minutes and 42 seconds.

• The top human champion earned a blistering time of 1 hour, 2 minutes, and 36 seconds (about 1 hour and 38 minutes faster).

• To put both times in perspective, the average human half-marathon time is ~2 hours.

And not all robots even made it to the finish line…

• Huanhuan (who’s normally very chill) wobbled unpredictably before giving up entirely.

• Shennong crashed spectacularly into a fence right after the start, breaking into pieces.

• Race staff had to rush to clear robot debris from the track after several dramatic wipeouts.

TBH, our favorite was definitely this little guy—or should we say, “Little Giant.” But even our short king had technical difficulties, at one point having smoke shoot out of his head. Because he’s just so smokin’ cute, that’s why!

Organizers eventually did extend the deadline to 4 hours and 10 minutes, so a total of 4 robots (two that were basically the same make) completed the race out of 21.

Even the fastest robots had some technical difficulties tho. TK Ultra face-planted one time after its battery failed, and needed three battery changes during the race. Most humanoid bots can operate for ~2 hours on a single charge, and faster speeds drains power even faster.

This raises an interesting question for future robot athletics: is a battery swap equivalent to a water break, or more like doping?

Why this matters: Developing basic abilities like running is critical for eventually deploying these bots in factories and industrial settings, where China aims to increase automation.

Our take: This race definitely showed how we’ve come in five years—and how far we’ve still left to go. As Michael Parekh pointed out, there was actually a very similar self driving car grand prix way back in 2004—cut to 2025, and we are still at the early stages of self driving car adoption. Mass humanoid robot adoption could either go faster thanks to self driving tech gains, or take just as long (if not longer).

As much as this race evokes the man vs machine imagery of John Henry, the real race here wasn’t between humans and robots, but between competing tech companies and nations. We’ll see if this inspires more US robot mega-rounds in competition with China, or somewhat cools the humanoid hype.

FROM OUR PARTNERS

🧰 The Tools, Templates & Playbook for Your AI Consultancy

Remember when AI experts were just “that person who knows ChatGPT”? 

Well, companies are now dropping serious cash—the AI consulting market is exploding 8x from $6.9B to $54.7B by 2032.

The problem? Everyone wants to be an AI consultant, but most are just winging it. No system, no frameworks. Translation? They're leaving money on the table while working twice as hard.

Our friends at Innovating with AI have welcomed 700 students into The AI Consultancy Project. 

Here’s why they like it: 

• Tools to find clients and deliver top-notch services.

• A 6-month plan for a 6-figure consulting business.

• First clients in as little as 3 days.

Click here to request access

Prompt Tip of the Day

The most innovative part of OpenAI's new o3 and o4-mini models is that they can literally “think with images” in their reasoning process to solve complex problems step-by-step.

In the same way that you can talk to ChatGPT via voice mode, you can now use screenshots or pictures to get ChatGPT to reason through your problems.

Here’s a few tips on how to use it:

• Be Clear About Your Goal: Specify what you want (e.g., “Summarize this chart,” “Find errors in this code,” “Translate this handwritten note”) helps produce more targeted responses.

• Use Images for Ambiguous or Visual-Heavy Tasks: If your question relies on visual context (e.g., layout, handwriting, diagrams), uploading an image will yield better results than text.

• Don’t Worry About Image Quality: The models are designed to handle low-quality, rotated, or cluttered images. They’ll process and manipulate the image to optimize their reasoning.

Treats To Try.

1. *Your AI data analyst is only as good as its trust score. Learn how to verify AI insights with Basejump and calibrate for accuracy before your first hire. Join our free webinar!

2. Gemini Flash 2.5 is the faster, cheaper version of Gemini 2.5 Pro that lets you pay only for the reasoning you need (optimized for cost vs performance).

3. Ad Clone scrapes your website, remixes top-performing LinkedIn ads with your branding, and generates customized creatives in 30 seconds—50 credits for $25 ($0.50/each) or 100 credits for $40 ($0.40/each).

4. Microagents connects your tools (Slack, G Suite, Notion, etc), groups them into teams, and completes tasks across platforms—free to try.

5. Google Whisk brings your images to life as 8-second animated videos—free for Google One AI Premium subscribers.

6. Checklist helps you generate and manage checklists for repeatable tasks—free to try.

7. Stupid Notes is a minimalist note-taking app with AI-powered search—free to try (app store only rn).

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

• OpenAI and SoftBank explored a possible international expansion of their Stargate data center project into the UK, France, and Germany (lets see if they raise that $100B first).

• Ahrefs (internet traffic and keyword checker) says keywords with AI overviews in the search result correlated with a 34.5% lower click thru rate for the top ranking page—that’s according to the 300K keywords they checked.

• The UAE launched the world's first AI-powered regulatory intelligence ecosystem to modernize government legislation, potentially reducing drafting time by 70%.

• Youtube now offers stable volume, an AI feature to automatically level audio between tracks—and within five years, plans to auto-dub all videos into every language.

• Here's when to use each new OpenAI model:

  • o3 for complex reasoning tasks (capable of 600 integrated tool calls and contextual image analysis).

  • GPT-4.1 for developers requiring precise instructions (API-only with 1M token memory).

  • o4-mini for cost-efficient performance (matches o3's math and visual capabilities at 9x lower cost with higher usage limits).

Under the Hood.

• Anthropic published their best practices for “agentic coding” with Claude—Simon Willison pointed out there’s a term called “ultrathink” that’s like a cheat code to get Claude to spend its 32K context window thinking before answering.

• Orpheus turns your text into emotional speech that can instantly clone any voice, completely open-source and free.

• Gemma 3 QAT shrinks Google’s high-performance Gemma AI models small enough to run on your home GPU (or on your phone w/ Google AI edge)—try it on ollama, Hugging Face, Kaggle and if you’re technical, read this + this.

• This is super important: while LLMs are powerful tools for engineers, they risk atrophying problem-solving skills—so developing focus, not just AI proficiency, will be crucial for tackling novel challenges in the future.

Monday Meme

A Cat's Commentary.