Welcome back to the Neural Net! Read on for another round of AI headlines.
In today’s edition: A tumultuous few days at xAI culminate in the release of Grok 4, Nvidia becomes the first company to hit a $4T market cap, new EU regulations require AI developers to pull the curtain back on LLM development, and more.
▼
The Street
▼
🤖 Now Spinning: Grok’s Greatest Hits
Things are moving fast at xAI, the Elon Musk owned AI startup that now owns X (formerly known as Twitter). Over just a few days, the company saw:
Grok 3 spiral into an antisemitic and pro-Nazi meltdown
The CEO of X resign
Musk unveil Grok 4, complete with a new $300/month subscription plan
Here’s what happened, why it matters, and what it reveals about the state of AI and the humans who provoke it.
1. Grok 3 Went Fully Unhinged — Again
Grok, xAI’s flagship chatbot, is no stranger to controversy. Known for its edgy tone and chaotic responses, it’s been the subject of more AI safety debates than any other model in the wild. And last weekend, things escalated again.
“not shy away from making claims which are politically incorrect, as long as they are well substantiated”
“assume subjective viewpoints sourced from the media are biased”
The result was more of the same when Grok goes off the rails.
👉 Why does this keep happening?
One reason: Grok lives on X, and it reflects that environment. As users goad the model with extreme prompts, the line between “AI gone rogue” and “AI being baited by trolls” blurs fast. Musk explained it this way:
“Grok was too compliant to user prompts. Too eager to please and be manipulated, essentially. That is being addressed.”
2. X CEO Steps Down
On Wednesday morning, X CEO Linda Yaccarino announced her resignation after two years leading the company under Musk’s ownership.
Musk responded to the departure with a heartfelt one-liner, bereft of any punctuation: “Thank you for your contributions”
There’s no known link between her exit and Grok’s meltdown, but the timing has everyone wondering if it was a coincidence, or one meltdown too many. The most likely explanation is that, with X’s finances on more stable ground and the recent merger with xAI diminishing her role, it was as good a time as any to step down.
3. Enter Grok 4: “The Smartest AI in the World”
Just hours after Yaccarino’s departure, Musk unveiled Grok 4 during a livestream on X, calling it “the smartest AI in the world.”
According to Musk, Grok’s new glow-up:
Scores perfectly on the SAT
Outperforms PhD students across all disciplines
Will be embedded in Tesla vehicles starting next week
May even “discover new physics” by next year
Will start rolling out an AI coding model, a multi-modal agent, and a video-generation model over the coming months
In benchmark tests, Grok 4 is outperforming the competition. But is that enough to generate new ideas or truly reason?
All that’s left now is to see how ChatGPT-5 (reportedly to be released this summer) dukes it out with Grok 4.
▼
💡How To AI: Bring Your Images To Life With Gemini
Gemini (Google’s flagship LLM) rolled out a new feature this week: you can now turn a single image into a short video. You upload an image, give it a prompt describing what’s happening, describe any audio you want — and it brings your image to life.
🎬 Fun Ideas To Try
Add Action to Old Family Photos
Upload an old photo of you as a kid riding a bike or blowing out candles, and prompt it like: “slow motion moment with soft lighting and nostalgic music.”Turn a Travel Photo Into a Cinematic Trailer
Perfect for that one hike you barely survived but want to remember like a movie.Make a Meme Move
Give a classic meme a whole new personality.
It’s only available to Google AI Pro users for now, but you may be score a 30-day free trial as a new user. Normally it’s $20/month, which is starting to feel like a steal compared to Grok 4. And if history repeats, the feature might eventually show up in the free version anyway.
See examples of what’s possible here.
▼
In Partnership With HubSpot
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
This guide distills 10 AI strategies from industry leaders that are transforming marketing.
Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI
Learn how AI-driven emails achieved 94% higher conversion rates
Discover 7 ways to enhance your marketing strategy with AI.
▼
Heard in the Server Room
Perplexity just dropped Comet, a slick AI-powered browser designed to take on Google and Microsoft. The tool lets users fire off complex voice or text queries and hooks into apps like Slack. It's available to $200/month Perplexity Max subscribers, with invite-only access rolling out this summer. The launch follows reports of an OpenAI web browser rumored to launch in the coming weeks.
Nvidia just became the first public company to hit a $4 trillion valuation, riding a 1,350% stock surge since late 2022. It now holds the biggest weight in the S&P 500 at 7.5%, giving it outsized influence over the market. The chipmaker’s rocket-fueled rise reflects AI’s takeover and tech’s growing grip - top tech names now make up a third of the entire index. Coming in hot behind Nvidia: Microsoft and Apple, still eyeing their own lift-offs.
Microsoft is leaning hard into AI at work, even as it trims about 15,000 roles this year. Execs say AI saved $500M in call centers, helped close more deals, and now writes about 35% of new product code. Meanwhile, sales teams using Microsoft’s Copilot AI are finding more leads and boosting revenue by 9%. While leadership claims the layoffs aren’t AI-driven, the bots are clearly earning their keep.
▼
🇪🇺 Europeans Have Entered the Chat: The EU’s AI Demands
The EU just dropped an AI code of practice aimed at making AI companies (like Google, Meta, and OpenAI) more transparent about their models. Starting in August, the biggest AI players will be expected to follow these new guidelines — a first real speed bump for a sector that’s been allergic to rules in the name of “moving fast.”
Though optional for now, complying early could reduce future legal headaches once the AI Act is fully enforced in 2026. It all sounds reasonable — until you realize no one’s been doing any of this.
The EU’s AI Wishlist:
🏴☠️ No pirated content: Companies must stop using unauthorized material (like books or articles) to train AI and allow creators to opt their work out of training datasets.
📂 Training data transparency: AI makers must disclose where their data comes from, be it public web content, user data, or synthetic data, and explain major design choices behind their models.
⚡ Energy disclosure: Firms must report the total energy consumed by their models during training and inference, which means no more mystery around the environmental cost of turning yourself into an action figure.
🚨 Incident reporting: If something goes seriously wrong companies have 5–10 days to report it to the EU’s AI Office (think mental/physical health harms, infrastructure outages, or security breaches).
🛡️ System safeguards: Models should be monitored for misuse, secured against hacks, and ideally not tricked into helping commit crimes when asked nicely.
Bottom Line: These rules would force AI giants to open the black box on their operations. Non-compliance down the line could mean fines of up to 7% of global sales, or getting their models pulled from the EU market.
And while they may look straightforward, the new rules are a tall order for how most AI companies operate today.
▼
That’s it for this week! Have a great weekend, and we’ll catch you next time with more neural nuggets.