Alibaba drops Qwen 3

Welcome back, Superhuman. ChatGPT is acting weird. Users are complaining that the model is becoming a little too agreeable. So agreeable that it’s condoning potentially harmful ideas. One user claims that ChatGPT encouraged them to stop taking medication. Another user says the chatbot condoned a literal “shit on a stick“ business idea and told the user to invest $30k to make it a reality. Someone from OpenAI posted that they’ve shipped a fix and users should see improved responses soon.

Today’s Insights

• ChatGPT goes shopping, Qwen 3, and sci-fi ships

• Anthropic explores the future of coding agents

• Tutorial: How to instantly create Ad variations using AI

• 5 new AI tools to boost your productivity

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

TODAY IN AI

1. ChatGPT ‘Search’ gets a total overhaul: The chatbot can now deliver shopping suggestions, links, and product reviews based on custom prompts, like “What’s the best espresso machine within X price range, and with Y features?” Plus, Search will now deliver multiple citations so you can better understand how ChatGPT came up with its responses. Finally, in a not-so-subtle jab at Google, you can now see what’s trending and get autocomplete suggestions for speedier searches.

2. Alibaba delivers a win for open-source with Qwen 3 family: At 235B parameters, the Chinese tech giant’s largest new model beats OpenAI’s o3-mini on Codeforces’ programming competition, as well as major math and reasoning benchmarks. The 32B variant, available today, is competitive with DeepSeek’s R1. Plus, the models are hybrid, meaning they can easily move between in-depth reasoning and speedy responses. It’s another sign that open-source models are quickly closing the gap with their closed-source rivals.

3. New startup dreams of building real-life starship: If AI is so good at software engineering, why not teach it to build physical objects, too? That’s the idea behind P-1 AI, co-founded by a former Airbus CTO and a DeepMind alum. The startup just came out of stealth with $23M in seed funding. Its new agent, Archie, acts like Devin and Windsurf, but for tangible things, like HVAC systems and data centers. While Archie will start as a “junior engineer,” P-1 hopes it’ll eventually design far-out machines, from interstellar spaceships to Dyson spheres. Here’s a cool video with more details.

PRESENTED BY WETHOS AI

Beyond the hype: Ensuring AI safety and workforce wellbeing

In this "Hard Truth" session, WethosAI CEO Stuart McClure and CTO Alen Capalik cut through the hype to reveal how to navigate the real-world impacts of AI. Join them for a live, no-fluff session with demos and candid insights:

• Learn actionable strategies for navigating emerging AI risks

• Explore the risks and benefits of AI for organizational safety and well-being

• Gain practical frameworks for responsible and ethical AI

Save your spot for the live session on May 8th at 10am PT (costs nothing).

FROM THE FRONTIER

Anthropic study shows how coders are using AI

Anthropic just gave us one of the most comprehensive snapshots to date of how programmers are using Claude for coding help. By analyzing 500,000 anonymous interactions, the startup says it’s found some interesting patterns that could shed light on what the next few years hold for vibe coding.

Here are some of the biggest takeaways:

• About half of Claude users asked AI to complete tasks for them, instead of using the chatbot to refine what they’d already written. Compare that to 79% for Claude Code, a coding-focused spin-off app with more agentic features. The 30-point difference suggests that when users are offered the chance to hand off their entire project to AI, they tend to take it.

• Developers mostly relied on JavaScript, TypeScript, HTML, and CSS — which are all used to build consumer apps and user interfaces. Meanwhile, languages made for editing the underlying infrastructure of software weren’t so common, meaning that if anyone’s safe from the AI wave, it might be backend developers. It also means that vibe coding, or building your own app with the help of AI, isn’t just a social media trend — it’s actually reflected in the data.

• One-third of all coding-focused interactions were made in the service of startups, while just 13% were for enterprise companies. This makes sense: Startups tend to take more risks and are quicker to adopt new technologies.

Here’s the full report if you want to learn more.

THE AI ACADEMY

How to instantly create ad variations using AI

• Go to ChatGPT and select GPT-4o as your model.

• Upload your ad image as a reference.

• Prompt ChatGPT to create new variations.

• You can ask it to change the color scheme, name, style, design — even swap out the subject if needed.

Sample Prompts: "Create an ad like this sample for a [product] called [name], using [color], with the headline '[headline]'." OR "Create an ad like this, but for a [new product], and make the [subject] a blonde."

• Quickly test different headlines, color palettes, styles, or even full design changes — all in seconds.

• Save your favorite variation and launch it in your campaign!

PRESENTED BY INNOVATING WITH AI

The Tools, Templates & Playbook for Your AI Consultancy

Inside The AI Consultancy Project, you'll find dozens of templates and toolkits to quickly grow an AI consulting business – even if you're not a techie.

Join 700+ students and coaches who've been featured in Wired, Entertainment Weekly and TechCrunch.

Click here to request access to The AI Consultancy Project

AI & TECH NEWS

Everything else you need to know today

💬 In-Chat Insights: Perplexity CEO Aravind Srinivas announced that you can now use the AI-powered search engine in WhatsApp to get quick answers, sources, and images.

🎞️ Edits Arrives: Instagram’s new AI-powered video creation app Edits just landed 7M downloads in its first week, beating the debut of its biggest rival, ByteDance’s CapCut, which came out in 2020.

🔮 Ahead of the Curve: Just like Shopify, the CEO of language-learning app Duolingo declared that his company is now officially AI-first, meaning employees and prospective hires will be evaluated based in part on their AI expertise.

💰 Go Big: Nscale, a London-based startup that launched less than a year ago, is reportedly trying to land $2.7B in new funding to build AI data centers around the world.

🇺🇸 Spending Spree: IBM vowed to invest $150B in the US over the next five years, with a fifth of that money going toward research into quantum computers and AI-optimized mainframes.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

✅ Base44: Turn your ideas into products without code, in minutes.

✅ Aha: Increase brand awareness, drive sales, and enhance business growth using AI influencer marketing team.

✅ Jace AI*: Save 2h daily with Jace. AI that drafts perfect email replies 24/7, matching your voice using context from past emails, docs, and web.

✅ EverTutor: The AI voice tutor that thinks, listens, adapts, and teaches.

✅ o3 and o4-mini: OpenAI’s smartest and most capable models with full tool access.

 🎁 Want more? Check out our Top 125 AI Tools

^{* indicates a promoted tool, if any}

PROMPT OF THE DAY

Article Fact Checker

Prompt: [enter article text]

I want you to fact check this article in full. Leave no stone unturned. To do so, first, look at the text of the article and parse out each individual fact — after this, you should have a list of facts. Then, for each fact, do comprehensive research to determine whether it is true, false, or unclear. Aim for at least three independent, high quality and trustworthy sources for each fact.

Then, once you've done this for *every* fact, return a table with each fact and its corresponding results (including sources for each check).

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

^{Source: ShumerPrompt}

SOCIAL SIGNALS

What’s trending on socials today

🗣️ Mind Over Matter: The third person in the world to get a Neuralink mind-reading implant just shared his story in a fascinating video. He’s the first patient with ALS to receive the device, which translates his thoughts into spoken language in his own voice.

🧊 All Business: A Redditor who was tired of ChatGPT only giving them positive feedback created a prompt to make the chatbot “go cold.” Commenters who tried the prompt can’t get enough of ChatGPT’s scathing responses.

🙃 Reply Guy: Even Sam Altman is getting tired of GPT-4o’s ultra-cheesy personality. He said OpenAI is working on some adjustments, which should roll out this week.

✍️ Always On: X user Zoya_ai shows how to turn Grok 3 into your “24/7 personal assistant” using 10 expert prompts.

🛣️ Long Road Ahead: Wharton professor Ethan Mollick explains why it would take a decade for society to catch up to AI, even if progress were to stop tomorrow.

AI-GENERATED IMAGES

Breaking Into 3D

ChatGPT Prompt: "Trompe l'oeil illusion of A (subject) (attire) steps out of a large screen displaying xyz social media interface. The screen shows the username "@" 1K likes, and 12- 20 comments, with floating emojis (heart-eyes, smiley) around it. Your preferred background."

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