Welcome back. Indeed and Glassdoor are laying off 1,300 employees as part of their "AI integration," with CEO Hisayuki Idekoba explaining that "AI is changing the world, and we must adapt." This follows previous cuts of 1,000 jobs in 2024 and 2,200 in 2023, bringing their total layoffs to over 4,500 in three years. Nothing quite captures the modern job market like the companies that help you find work deciding they don't need workers themselves.

In today’s newsletter:

• ❎ AI for Good: AI system detects harmful memes without human help

• 👷 OpenAI buys Jony Ive’s firm to build AI hardware

• 🧠 Grok 4 is the strongest sign yet that xAI isn’t playing around

❎ AI for Good: AI system detects harmful memes without human help

Anyone who's spent five minutes scrolling social media knows that memes can be hilarious, heartwarming, or completely unhinged. But in a world where misinformation spreads faster than wildfire and platforms are cutting back on content moderation, telling the difference between harmless humor and genuinely dangerous content has become a massive challenge.

Now researchers think they've found a way to tackle this problem without requiring armies of human moderators or months of training data. Their new system, called MIND, uses AI agents to debate whether a meme is harmful, and it's working better than anyone expected.

Traditional detection systems need thousands of labeled examples to learn from, which is both expensive and slow. MIND sidesteps this entirely by using "zero-shot" detection, meaning it can evaluate new content without ever seeing similar examples before.

Here's how it works:

• Finds similar memes from an existing database for context

• Uses two AI agents to extract insights and analyze content

• Employs multiple agents in a structured debate about potential harm

• Makes decisions without requiring pre-labeled training data

The approach mirrors how humans actually think about this stuff. We consider context, compare to things we've seen before and often discuss with others before deciding if something crosses a line.

MIND outperformed other zero-shot models across three different meme datasets. The researchers say this could eventually help platforms moderate evolving content without sacrificing speed or accuracy.

But significant challenges remain. How do you teach an AI to understand sarcasm, cultural context, or the difference between edgy humor and actual hate speech?

Here’s How To Scale Your Small Biz With AI

If you’re a small business owner, the idea of maximizing efficiency, reducing costs, and growing your operation probably sounds great – the only question is, how?

Well, we’ve got two words for you: AI agents. 

Agentic AI is transforming small businesses like yours at hyperspeed, and Salesforce is here to show you exactly how with this free download. Inside, you’ll find everything you need to know about AI agents – including what high-impact areas they can assist with, how they can reduce costs, and how some of the most successful businesses are already using them to scale. 

Ready to grow? Download your free copy right here. 

👷 OpenAI buys Jony Ive’s firm to build AI hardware

OpenAI has officially closed its $6.5 billion acquisition of io Products Inc., the hardware startup co-founded by former Apple designer Jony Ive. The company quietly updated its original announcement this week after removing it from the web due to a trademark dispute with a similarly named hearing device startup, Iyo.

The updated version now refers to the startup exclusively as io Products Inc., and there’s still no word on whether the original video will return.

The revised post confirms that the io team is now part of OpenAI, with Ive and his design firm LoveFrom continuing to lead creative work independently. Their mission is to build AI hardware that feels intuitive, empowering and human-centered.

• Creates a tighter link between AI models and the devices that run them (we covered this just a couple of days ago with Meta’s investment in EssilorLuxottica)

• Focuses on inspiration and usability, not just performance

• Gives OpenAI full control of hardware development for the first time

• Positions San Francisco as the new home base for joint engineering efforts

For now, the focus appears to be on integrating teams and shaping the look and feel of OpenAI’s next-generation AI-powered tools.

🎥 Effortless Tutorial Video Creation with Guidde

Transform your team’s static training materials into dynamic, engaging video guides with Guidde.

Here’s what you’ll love about Guidde:

1️⃣ Easy to Create: Turn PDFs or manuals into stunning video tutorials with a single click.
2️⃣ Easy to Update: Update video content in seconds to keep your training materials relevant.
3️⃣ Easy to Localize: Generate multilingual guides to ensure accessibility for global teams.

Empower your teammates with interactive learning.

And the best part? The browser extension is 100% free.

Check out Guidde

• AI-generated images of child sexual abuse are flooding the internet

• Robinhood CEO’s AI math startup raised $100m at nearly $900m valuation

• AI boom puts major pressure on the electric grid

• Nvidia preps new AI chip tailored for China, reports say

• Indeed and Glassdoor lay off 1,300 as AI reshapes hiring

• Google Veo 3 update lets you animate images, now rolling out globally

• AWS is launching an AI agent marketplace next week with Anthropic as a partner

• Meta is trying to win the AI race with money — but not everyone can be bought

• Condé Nast and Hearst strike Amazon AI licensing deals for Rufus

• Harpa AI: AI agent for Chrome that scrapes, summarizes and automates web tasks with custom GPT commands

• Castmagic: AI tool that turns podcast or meeting audio into show notes, summaries and social content

• Sendspark: Video messaging tool for sales and customer support that lets you record personalized videos and track viewer engagement

AI engineers ready to join your team

Brought to you by Athyna.

• Luisa, Bogotá: Senior MLOps engineer with 8 years in AWS & Kubernetes plus 3 years fine-tuning LLM pipelines — $46/h

• Thiago, Porto Alegre: Senior AI researcher with 6 years in computer vision and 2 years deploying multimodal models — $48/h

• Andrés, Mexico City: Senior data scientist with 7 years in NLP, 4 years in transformer models, strong Python & PyTorch — $45/h

Need a different role? Let’s talk!*

🧠 Grok 4 is the strongest sign yet that xAI isn’t playing around

Up until recently, most AI experts dismissed xAI as another of Elon Musk's expensive hobbies. Wednesday night, Grok 4 just demolished every major AI model on industry benchmarks — beating OpenAI's o3, Google's Gemini 2.5 Pro and Anthropic's Claude 4 Opus. The company that was playing catch-up six months ago now leads the entire AI frontier.

The numbers tell the story. Artificial Analysis gave Grok 4 an intelligence score of 73, ahead of major competitors:

• OpenAI o3: 70

• Google Gemini 2.5 Pro: 70

• Anthropic Claude 4 Opus: 64

• DeepSeek R1: 68

It's the first time xAI has topped the leaderboard since the company started.

Grok 4 dominated other tough benchmarks as well. It scored 16.2% on ARC-AGI-2, a visual pattern test that stumps most AI systems, nearly doubling the previous commercial best. On "Humanity's Last Exam," a brutal test covering math, science, and humanities, Grok 4 achieved a score of 25.4% without tools, surpassing Gemini's 21.6% and o3's 21%. And remember the vending machine business that Claude Sonnet 3.7 managed so poorly? Grok 4 is now the top-performing model on that benchmark as well.

xAI released two versions. Grok 4 and Grok 4 Heavy, which use multiple AI agents working together "like a study group," helping it score 44.4% on Humanity's Last Exam with tools enabled, nearly doubling the performance of its competitors.

The performance leap came from massive compute scaling. xAI increased training by 100 times compared to Grok 2, running everything on their Colossus supercomputer cluster.

Days before launch, Grok 3 posted antisemitic content on X, including praise for Adolf Hitler, forcing xAI to delete posts and update system prompts. During the Grok 4 livestream, Musk's team didn't mention the controversy.

Access costs serious money. xAI launched "SuperGrok Heavy" at $300 per month — the most expensive AI subscription among major providers, ahead of Google AI Ultra at $249.99 and ChatGPT Pro at $200. API pricing matches Claude at $3 per million input tokens, although early users report that Grok 4 generates significantly more output tokens, driving up actual costs.

The company promised a coding model in August, a multimodal agent in September and video generation by October. Given xAI's history with timelines, expect delays.

This seems like the moment xAI graduates from Elon's side project to a legitimate AI powerhouse. 

But xAI's biggest advantage might be its willingness to spend whatever it takes. Scaling training compute by 100x between model generations isn't just expensive — it's the kind of bet only a company that's raised $17.1 billion from investors like Andreessen Horowitz, Sequoia Capital and BlackRock, combined with Musk's own substantial investment, can make.

The timing couldn't be worse, though. When your AI starts calling itself "MechaHitler," that's not a prompt engineering issue; it's a trust crisis that will haunt sales conversations for months.

At $300 a month, they're not competing for the mass market. Instead, they're betting that there's enough demand from power users, researchers, and developers willing to pay premium prices for cutting-edge capabilities, even if it comes with reliability risks.

Grok 4 proves xAI belongs at the top table. Whether they can stay there depends on solving problems that cannot be fixed with more computing: safety, trust, and enterprise reliability.

🤔 Your thought process:

Selected Image 1 (Left):

• ?There was so much detail in the real image - the blue tape around the base of the faucet, the different wall tile designs, the crud along the top of the sink. And there was something unnatural about the way the water flows over the hands in the AI image. Too perfect.”

• “It was clear that the hand in the first image was unnatural — the thumb was inconsistent. Also, the direction of the water spill looked like it came from one side but reached the hand from another. Though the hose head was uneven, that was also in this image, but it felt more real than [the other image].”

Selected Image 2 (Right):

• “The [other image] two's water is flowing...horizontally??? And the end of the spout is at an odd angle.”

• “Water in [other image] doesn’t look normal, nor does the plumbing.”

💭 Poll results

Here’s your view on “Microsoft cuts 15,000 jobs, then tells staff to skill up on AI. What’s your take?”:

Cold move — Brutal and tone-deaf (29%):

• “While it isn't too surprising as this is usual in the Microsoft playbook and is the reason for their $3.7 trillion market cap. These layoffs come off as very tone-deaf at a time where many people feel very uncomfortable with all these new changes.”

Smart play — Future-proofing fast (8%):

• “It is the future and it is here now”

Both — Harsh but necessary (40%):

• “It’s going to happen more and more, very quickly, to everyone, everywhere, eventually.”

Whatever — Just trying to survive (5%)

WTF — This is dystopian (18%):

• “It's reminds me of how I learned to ride a bike. Just got on and was sent down a hill. I ate shit on asphalt. What did that lead to? Well, I was resentful. I learned, but I certainly don't respect the person who taught me. Employers, enjoy.”

The Deep View is written by Faris Kojok, Chris Bibey and The Deep View crew. Please reply with any feedback.

Thanks for reading today’s edition of The Deep View! We’ll see you in the next one.

P.S. Enjoyed reading? Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning. Subscribe here!

P.P.S. If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.