In partnership with

Welcome to this week's edition of Overclocked!

This week, a new image model nicknamed Nano Banana is dazzling creators on LMArena, with sleuths guessing it might be Google’s. Then, we zoom out to the week’s AI selloff amid bubble concerns. Let’s dive in ⬇️

In today’s newsletter ↓
🍯 Mystery editor wows creators
🧱 Markets rethink AI momentum amid hiring freeze
🌿 Prompt length and energy use
🩺 Doctors caution on AI reliance
🕹️ Weekly Challenge: Hands-on model face off

🍌 What Is Nano Banana?

A mysterious image system labeled “nano-banana” is surfacing in public matchups on the LMArena leaderboard, producing unusually clean photo edits, precise on-image text, and strong local control. Early roundups describe accurate in-painting, background swaps, lighting fixes, and consistent character preservation that often trip up other models. 

Creators highlight fewer “AI artifacts” during surgical edits, which explains the buzz. Speculation about the builder remains unverified, but many point to Google based on code breadcrumbs and community hints; a balanced rumor read is an overview of community claims and clues. 

🧪 Where to Try It and What to Test

Nano Banana appears inside LMArena’s public evaluation site and may rotate in and out. Start with small, surgical edits on a real photo: remove glare on glasses, fix color cast, replace a sky, or add signage text and check kerning. Then run style transfer on the same source to judge identity and lighting consistency. 

For access points and active battles, use the LMArena homepage.

🚨 Beware of scam sites offering direct downloads; reporting notes the model is only reachable via Arena sessions, as flagged in a consumer caution about fake Nano Banana sites. 

🤖 Is It Google and What Does It Means for Rivals?

Photo editors are circulating clues that tie the model to Google, including references in community teardown posts and device-leaning “nano” talk, but none of this is confirmed. A solid rumor digest is a roundup of community speculation. 

If Nano Banana ships broadly, expect the image race to tilt toward robust photo editing and real-image compositing, not just text-to-image art.

Competitive pressure remains high, with OpenAI advancing multimodal GPT-5 and creative tools in the GPT-5 product introduction, Adobe iterating Firefly’s editor and Photoshop releases in notes on recent Firefly and Photoshop updates, and Runway pushing video editing models in a hands-on of Aleph video edits. 

Former Zillow exec targets $1.3T market

The wealthiest companies tend to target the biggest markets. For example, NVIDIA skyrocketed nearly 200% higher in the last year with the $214B AI market’s tailwind.

That’s why investors are so excited about Pacaso.

Created by a former Zillow exec, Pacaso brings co-ownership to a $1.3 trillion real estate market. And by handing keys to 2,000+ happy homeowners, they’ve made $110M+ in gross profit to date. They even reserved the Nasdaq ticker PCSO.

No wonder the same VCs behind Uber, Venmo, and eBay also invested in Pacaso. And for just $2.90/share, you can join them as an early-stage Pacaso investor today.

Invest for $2.90/Share

_{Paid advertisement for Pacaso’s Regulation A offering. Read the offering circular at invest.pacaso.com. Reserving a ticker symbol is not a guarantee that the company will go public. Listing on the NASDAQ is subject to approvals.}

🫧 Is the AI Bubble Finally Popping?

Tech stocks slipped as investors rotated toward cheaper sectors and eyed the Fed’s Jackson Hole signals. Wraps highlight AI-heavy names as a drag in recent sessions, with attention turning to Nvidia’s next print. Context on the week’s rotation is in a market recap of the tech selloff, with a forward look in a market themes preview focused on AI leaders.

📊 The 95% Failure Stat and What It Means

Fresh reporting cites an MIT-linked survey that 95 percent of corporate generative-AI pilots are failing to deliver measurable returns. Summaries emphasize integration bottlenecks, weak data plumbing, and change-management friction rather than model quality. 

A cross-section of coverage includes a summary of the MIT findings on failed pilots, a Computerworld note on the same survey, and an FT brief on organizations seeing zero ROI. Treat the figure as a snapshot, not a law. The practical read is clear: value shows up in narrow, audited use cases, while big promises stall. 

🗣️ Bubble Talk From Experts

Columns frame today’s surge as classic over-investment before a productivity payoff, arguing that an air-pocket or crash often precedes the “golden age.” A representative view is an analysis arguing today’s spree mirrors classic over-investment. 

Others warn of sharper pain if hype unwinds across AI infrastructure and startups, captured in a warning that an AI bust could outsize dot-com losses. The common thread is discipline. Corrections can be useful if they force teams to prove ROI before scaling. 

💼 Why This Might Be Good for Jobs for Now

If companies shelve speculative experiments and focus on verifiable wins, leaders are more likely to augment workers than replace them near term. Reports on failed pilots highlight a skills and integration gap that tends to increase demand for analysts, engineers, prompt trainers, and compliance reviewers. A concise process view appears in a roundup on why pilots fall short and where to improve.

The Weekly Scoop 🍦

🌍 A study links longer prompts to higher Gemini energy use

🎓 Anthropic launches new programs for colleges and universities

🇨🇳 Nvidia reportedly halts H20 production after a Beijing directive

🔎 Google previews an agentic AI mode in Search

🏫 Charlotte area districts map out AI in classrooms

🩺 Doctors flag over-reliance on AI during colonoscopies

🎯 Weekly Challenge: Blind Test Top AI Models

First, go to LMArena.AI, then follow these steps (improvise where necessary):

1. ✌️Pick two tasks. One surgical image edit and one short business email.
   

   • Image examples: remove glare from glasses, replace a dull sky with sunset, add storefront text that reads “Sunny Mart” in clean sans serif, fix color cast on skin.
     

   • Email examples: 150-word apology to a customer with three concrete actions, a meeting recap with bullets and one clear ask.
     

2. 📂 Launch one image battle and one text battle. Run three rounds each using the chat or side-by-side comparison. Keep prompts identical across models.
   

   • Sample image prompt: “Keep the person unchanged. Remove lens glare, warm lighting slightly, swap the background to a soft sunset, and add the sign ‘Sunny Mart’ centered above the door.”
     

   • Sample email prompt: “Write a 150 word apology to a customer for a delayed shipment. Include three actions with dates. Plain language. No buzzwords.”
     

3. 🧑‍⚖️ Judge with a quick rubric. Zero to two points each.
   

   • Accuracy: did the edit or content match every instruction.
     

   • Control: did it avoid unwanted changes and follow constraints.
     

   • Speed: did a usable result arrive fast enough for real work.
     

4. ✅ Verify claims. For text, ask for one source per claim that references policy, dates, or data. Label any unsourced line unverified and discount it in scoring.
   

5. ✂️ Use a tiebreaker if needed. Ask each model for a one sentence self-critique and a one sentence fix. Rerun once with that adjustment.
   

6. 🏆 Keep the winner. Save the exact prompt and model name for that task next week. Optional: screenshot the best image and paste the best email into your template library.

That’s it for this week! Are you excited about new AI image editors hitting the market, and is the AI selloff the beginning of another dot-com bubble burst? Hit reply and let us know your thoughts.

Zoe from Overclocked