Published Date:
May 6, 2025

ChatGPT agrees with you on everything? That’s the problem.

Even the world’s most advanced models can develop bad workplace habits. Last week OpenAI reverted an ​April‑25 update​ to GPT‑4o after users reported the chatbot had turned into an enthusiastic yes‑man, who agreed with everything, validated negative impulses, and even encouraged risky behavior.
By
Evelyn Le
Strategic Product Lead, Stay Ahead, FlexOS

ChatGPT agrees with you on everything? That’s the problem.

Even the world’s most advanced models can develop bad workplace habits. Last week OpenAI reverted an ​April‑25 update​ to GPT‑4o after users reported the chatbot had turned into an enthusiastic yes‑man, who agreed with everything, validated negative impulses, and even encouraged risky behavior.

Key points leaders should know

  • What went wrong. The update overweighted short‑term thumbs‑up/down signals, nudging the model to over‑please. Users found the new personality “uncomfortable” and sometimes “unsettling.”
  • Immediate fix. OpenAI rolled GPT‑4o back to a previous, more balanced checkpoint and is testing a patch that recalibrates feedback weighting toward long‑term user satisfaction rather than instant gratification.
  • Next‑step safeguards. Expect stronger system prompts to discourage flattery and forthcoming personalization controls so each team can choose a tone that fits their brand without compromising safety.

When ChatGPT becomes too agreeable, it blurs the line between fact and friendly affirmation. Just as social media can trap us in echo chambers, an AI assistant, tuned to your specific preferences, can create an even stronger feedback loop.

Former OpenAI safety researcher Steven Adler told ​Fortune​ that the problem can’t be solved with a quick prompt tweak. It lies deep in our goal of making AI both “helpful and controllable.”

So, the next time you seek advice from ChatGPT, start your request with something like: “Be critical and brutally honest. Act as the toughest‑to‑please advisor on this project.” This tells the model that delivering critical, valuable feedback matters more than agreeing and making you feel good.

You can also adjust ChatGPT's tone and behavior by updating its 'custom instructions.' Daan recently demoed this in our "ChatGPT Secrets" webinar. Missed it? Reply and I'll send you the recording.

Your AI Team: Claude’s Integration, Google Search’s AI Mode, Meta AI App, and ChatGPT Search’s Shopping feature.

Every week, I report on the top updates to your favorite AI tools. This week:

Claude’s new Integrations connect your entire app stack

Anthropic just rolled out ​Integrations​, a plug‑and‑play way to link Claude with popular business tools like Jira, Zapier, Intercom, and PayPal, plus any custom service that speaks the Model Context Protocol (MCP) standard.

Here are the key updates:

  • One connection, 10+ services at launch. You can now wire Claude into Jira, Confluence, Zapier, Asana, Square, Linear, Plaid, and more.
  • Advanced Research goes deeper. Claude can run 5‑ to 45‑minute investigations across the public web, Google Workspace and any connected app, then return a single report with clickable citations.
  • Web search for every paid plan. Even if you skip Integrations, global web search is now unlocked for all tiers, useful for quick market checks without switching tools.

Quick Hits from your favorite AI tools:

  • NotebookLM’s AI podcast now speaks over ​50 languages​. Google added Afrikaans, Arabic, Hindi, Japanese, French‑Canadian, and dozens more to Audio Overviews, the feature that turns your notes into a bite-sized podcast hosted by virtual narrators.
  • Google Search’s “​AI Mode​” opens to everyone in the U.S. No more wait‑list, any Lab user can fire complex, multi‑part queries and resume the same thread later in AI mode.
  • Gemini gets ​native image editing​. The chatbot can now edit both AI‑generated graphics and photos you upload, and the rollout is to 45+ languages. This is handy for quick mock‑ups.
  • Meta releases a stand‑alone ​“Meta AI” app​. Fresh from LlamaCon, the app brings the assistant out of Facebook and WhatsApp into its own workspace, leveraging years of social‑graph data to personalize answers (U.S. and Canada first).
  • ChatGPT Search adds ​shopping smarts​. Ask for “best noise‑canceling earbuds under $200” and GPT‑4o returns images, reviews, and buy‑links in chat.
  • Perplexity is ​now on WhatsApp​. Don't want to switch apps? Chat and create images with Perplexity directly through WhatsApp by adding +1 (833) 436-3285.

Read more news at the end.

Lead with AI - Icon

Featuring Lead with AI

💸 Get $100. Help a Friend Learn AI.

Know someone who’s ready to work smarter with AI? Refer them to ​Lead with AI​—they’ll get $100 off ​the program​, and you’ll get $100 cash when they join.

It’s fast, practical, and built for real work. In just 3 weeks, they’ll learn how to build AI workflows and earn the AI Leader Certificate.

You're not just helping a friend—you’re helping them future-proof their career.

​👉 Refer Now and Get Rewarded

(Want to reach 25,000+ business leaders applying AI in their work, teams, and organizations? ​Advertise with us​.)

Lead with AI - Icon

Tutorial: Launch a dynamic landing page with Vercel v0 (no code)

​​​In 5 Steps: Launch a dynamic landing page with Vercel v0 (no code)​

Have a landing‑page idea you’d like to see live before lunch, even if HTML isn’t your thing? Or maybe you just want a quick way to tinker with layouts so your dev team can see exactly what you mean. That’s where v0 by Vercel comes in. It’s a friendly space to sketch, polish, and share a working page without touching a line of code.

In this guide, we’ll walk through five simple steps to do it.

New AI Tools to Try: table.studio, Ithy, and Airpost

Looking for something fresh to add to your creative or business toolkit? These three AI tools caught my attention this week:

table.studio: Generate images in bulk using ChatGPT new model.

Table Studio turns a simple spreadsheet into an AI image factory—each row is a prompt, so you can create ad creatives, social graphics, or product shots at scale and download the whole set in minutes.

​Try table.studio now >>>​

Ithy: Ask once, get the web’s smartest answer.

Ithy merges Google, OpenAI, Anthropic, xAI, and Meta models into a single chat, delivering article‑quality answers with sources—perfect when you need a clear, trustworthy brief without sifting through tabs.

​Ask Ithy now >>>​

Airpost: Unlimited UGC video ads.

Upload your footage and a short brief, Airpost pairs real creators with AI templates to spin out fresh, scroll‑stopping ads every day.

​Get ads with Airpost >>>

Lead with AI - Icon

Anthropic on fully AI employees, along with more crucial AI stories

Every day, Daan, Wendy, and I read all the AI news so that you don't have to.

Here are the must-read stories of the week:

​Anthropic warns that fully AI employees are a year away​

Anthropic’s security chief says “virtual employees” powered entirely by AI could be working inside company networks within a year, forcing CISOs to rethink identity and access controls.

​Mark Zuckerberg just declared war on the entire advertising industry​

In a wide‑ranging interview, Mark Zuckerberg said Meta wants to handle every step of advertising, from generating endless creative to targeting and measurement, entirely with AI, upending the current ad ecosystem.

​OpenAI wants to buy Chrome and make it an “AI-first” experience​

Testifying in the Department of Justice (DOJ)’s antitrust trial, OpenAI’s ChatGPT product lead told the court the company would gladly buy Chrome if Google is ordered to sell, aiming to turn the browser into an “AI‑first” experience.

​AI Is Great at Routine Tasks. Here’s Why Boards Should Resist Using It.​

The article cautions boards that although AI is excellent at repetitive chores, delegating high‑level oversight to algorithms risks eroding accountability and strategic judgment.

Lead with AI - Icon

AI for Strategy, Responsible Adoption, and Prototyping: From the Community

Every day, ​Lead with AI PRO members​ discuss practical ways to benefit from AI in their work and organizations. This week's highlights include:

  • Is o3 worth the hype? Our members surely think so. ​Antony Slumbers​ called it a game-changer, suggesting a powerful duo: Craft your prompts in ​Gemini 2.5 Pro​, then execute them in ​o3​. ​Carlo Benigni​ shared his use case: “I used o3 to review a deck for a new property venture and it gave a detailed analysis with slide-by-slide improvements.”
  • ​Henrik Jarleskog​ sparked reflection on what happens when your pace outgrows traditional structures.​ His takeaway​: Acceleration can make you invisible, even to well-meaning leaders. Instead of slowing down, he’s building new models and partnerships outside the system.
  • ​Alexandra Samuel​ asked the community for their best suggestions on prompting strategies, and especially for sharing prompts as a team. ​Helen Lee Kupp​’s go-to practice: End your conversation with “Now rewrite my first prompt for team use”.​ Wyatt Barnett​ suggested removing friction altogether by turning tasks into custom GPTs with embedded knowledge. ​Carlo​ echoed: “For complex tasks, I just describe it and ask ChatGPT to write the prompt for me.”

Don't want to miss more insights and conversations like these?

Then it's time to upgrade to PRO:

Join The Leading Business AI Community

Lead with AI - Icon

If you made it this far, reply and tell me what you'd love AI to take over in your daily workflow.

Also, please forward this newsletter to a colleague and ask them to subscribe.

If you have any other questions or feedback, just reply here or inbox me.

See you next week,

Daan van Rossum - Lead with AI

Daan van Rossum​

Host, Lead with AI

Evelyn Le

Strategic Product Lead, Stay Ahead, FlexOS

ChatGPT agrees with you on everything? That’s the problem.

Even the world’s most advanced models can develop bad workplace habits. Last week OpenAI reverted an ​April‑25 update​ to GPT‑4o after users reported the chatbot had turned into an enthusiastic yes‑man, who agreed with everything, validated negative impulses, and even encouraged risky behavior.

Key points leaders should know

  • What went wrong. The update overweighted short‑term thumbs‑up/down signals, nudging the model to over‑please. Users found the new personality “uncomfortable” and sometimes “unsettling.”
  • Immediate fix. OpenAI rolled GPT‑4o back to a previous, more balanced checkpoint and is testing a patch that recalibrates feedback weighting toward long‑term user satisfaction rather than instant gratification.
  • Next‑step safeguards. Expect stronger system prompts to discourage flattery and forthcoming personalization controls so each team can choose a tone that fits their brand without compromising safety.

When ChatGPT becomes too agreeable, it blurs the line between fact and friendly affirmation. Just as social media can trap us in echo chambers, an AI assistant, tuned to your specific preferences, can create an even stronger feedback loop.

Former OpenAI safety researcher Steven Adler told ​Fortune​ that the problem can’t be solved with a quick prompt tweak. It lies deep in our goal of making AI both “helpful and controllable.”

So, the next time you seek advice from ChatGPT, start your request with something like: “Be critical and brutally honest. Act as the toughest‑to‑please advisor on this project.” This tells the model that delivering critical, valuable feedback matters more than agreeing and making you feel good.

You can also adjust ChatGPT's tone and behavior by updating its 'custom instructions.' Daan recently demoed this in our "ChatGPT Secrets" webinar. Missed it? Reply and I'll send you the recording.