Welcome, humans.
So, someone asked ChatGPT to “enhance” a blurry photo of their grandfather and it, well, accidentally turned him into Nelson Mandela.

The user genuinely asked “what did I do wrong?” as if they'd missed a checkbox somewhere. But here's the thing: AI doesn't actually “enhance” photos like CSI taught us, despite this trick working on city skylines. It just looks at your blurry pixels and thinks “hmm, elderly Black gentleman... I know! Nelson Mandela!” because that's who it's seen a million times in training data (not your actual grandpa). And no, before you ask, this is not the Mandella effect in action…
One commenter nailed it: “ChatGPT has no magical powers and can't know what your grandpa looks like without ever seeing him.” In fact, a lot of the comments are gold, with people sharing their own AI photo disasters, like someone trying to remove a relative from a wedding photo and getting “a small smiling Asian child” instead. Now, if you want to see an actually decent attempt at upscaling the photo, here ya go.
Here’s what you need to know about AI today:
- We break down Microsoft’s recent analysis of 200K Copilot chats.
- Disney rejected AI deepfakes for Dwayne Johnson in upcoming films.
- Google agreed to pause AI workloads during peak energy hours in the midwest.
- New wicked fast video model and ultra accurate text-making image model drop.

OpenAI’s AI Job Impact Predictions from 2023 Were Scary Accurate, According to Microsoft

Waaay back in the ancient year of 2023, OpenAI published a list of predictions of what jobs will be the most impacted from AI. Well, Microsoft just analyzed 200K real AI conversations to see what people actually use AI for at work.
Ethan Mollick pointed out this wild overlap between what experts thought would happen and what's actually happening. The 2023 paper predicted which jobs AI would affect most, and Microsoft's data confirms it—with correlation so high it's almost suspicious. The punchline: Those predictions were 90% accurate.
Here's what Microsoft discovered by creeping on our Copilot chats. Microsoft found there are ~40 jobs that are most applicable to AI, and 40 that are least applicable.
Top jobs targeted by AI:
- Interpreters and translators (basically doomed).
- Customer service reps.
- Writers and journalists (ouch).
- Sales representatives.
- Telemarketers.
Safest jobs (obvious, but still worth mentioning):
- Anything requiring physical touch (massage therapists, surgeons).
- Operating heavy machinery.
- Manual labor (roofers, dishwashers).
The most interesting finding? Jobs requiring communication and information-sharing are getting hit hardest (the lowest hanging fruit for AI to do). But AI still struggles with visual design, data analysis, and anything requiring real-world interaction (sorry agents).
So here’s what this means for your career (in our opinion): Over the weekend, we read a Hacker News thread debating predictions on whether AI will lead to more or less freelancing. The consensus? It's complicated, but there's opportunity on both sides.
See, AI produces a lot of work quickly, but much of it is mediocre. It's like having an intern who works 24/7 but needs constant supervision. Someone still has to check and fix everything.
Balaji recently wrote how AI does “middle-to-middle” work, not “end-to-end.” We still need us humans to “prompt, then verify” (← 9 other interesting takes in there, too).
This also means one skilled person (prompting + verifying) can now do what used to require a team. So, the winners will be specialists who:
- Master something AI can't easily copy.
- Work with AI instead of against it.
- Build strong reputations (cause AI can't network for you).
- Focus on solving real problems people pay for.
For Different Types of Workers
- If you're an employee: The competition is getting tougher because AI can do some of what you do. But there's also hope: companies will need more people to manage and fix AI's work.
- If you're a freelancer or consultant: You might actually do better, especially if you're really good at something specific. Companies are becoming less willing to hire big consulting firms, but will pay well for individual experts who can solve their particular problems quickly.
- If you want to start your own business: This could be your moment. AI tools mean one person can now do work that used to require a whole team. But so can everyone else, so the competition will be fierce.
The key is positioning yourself as someone who can deliver results… be like ChatGPT: focus on outcomes, not time spent working!

FROM OUR PARTNERS
Fine tuning your AI models? Use data you can trust.

Training AI? Let’s talk. If you're actively building or fine-tuning AI models, Shutterstock offers enterprise-grade training data across images, video, 3D, audio, and more—fully rights-cleared and enriched with 20+ years of human-reviewed metadata.
Our 600M+ asset library powers faster iteration, simplified procurement, and improved model accuracy.
We’re offering a $100 Amazon gift card to qualified AI decision-makers who join a 30-minute discovery call with our team. We’ll walk you through how our multimodal datasets, flexible licensing, and premium metadata can help de-risk development and accelerate time to deployment.
For complete terms and conditions, see the offer page.

Prompt Tip of the Day
It’s August. School is soon to be back in session. It’s probably got a lot of you students out there feeling like this:

Well, in honor of BTS (not the Korean band), here’s a prompt for helping you study with ChatGPT:
I'm going to upload a screenshot of a textbook page. Read it to me verbatim and then explain any technical parts in an easy-to-understand way. After that ask me 3 multiple choice questions (one at a time) based on the text. After I've answered the questions ask for the next upload.
Anyone can use this to help them understand any concept they’re trying to learn, btw
Of course, the top comment in response to this was “excuse me sir, but do you know NotebookLM?”
NotebookLM can make video overviews now, it can create study maps and mind maps for you, and of course, it can create a custom podcast on any topic that you can interrupt (via “Interactive Mode”) and ask questions to (you can even request the hosts to act “as close to a Harvard professor” as possible). Enjoy!
Check out all of our Prompt Tips of the Day from July here.

Treats To Try.
*Asterisk = from our partners. Advertise in The Neuron here.
- *Spinach records, transcribes & summarizes your meetings, then automatically updates your CRM and project tools.
- Harvey reviews your contracts, handles due diligence, and researches legal questions for you (and just hit $100M ARR + 500 customers in 54 countries).
- FastWan generates 5-second videos for you in only 5 seconds on a single GPU, making video creation 70X faster than traditional methods (read more).
- Qwen-Image creates and edits images in any style while rendering perfect text that looks naturally integrated—try it here.
- Galileo turns satellite and climate data into maps that track floods, wildfires, and glacier changes.
- Anime.js animates anything on your website with simple JavaScript code.
See our top 51 AI Tools for Business here!

Around the Horn.
- Google agreed to stop non-essential AI workloads during peak energy demand hours in the midwest.
- Disney apparently decided against using AI deepfakes for Dwayne Johnson in the live action Moana and new Tron movie due to the potential legal and public backlash.
- Google launched the Kaggle Game Arena, a new benchmark that pits AI models against each other in chess matches so you can see which ones actually think strategically.
- Meta’s latest salary offer for $1.5B over six years was rejected by AI researcher Andrew Tulloch, who now is at ex-OpenAI CTO Mira Murati’s new company, Thinking Machine Labs.
- Georgia Channing shared a thread on X of machine learning breakthroughs in the sciences that are worth checking out.
- Axios released a great recap on how AI is basically shouldering the American economy right now; paraphrasing Derek Thompson: “GDP growth = AI capex.”

Tuesday Tweets
A tale told through 10 tweets that capture the state of the industry atm.











A Cat's Commentary.

