Perplexity just launched Labs where AI agents spend ten minutes building you entire websites and research reports automatically. 

A Wall-E looking robot hit the market for $3K while we accidentally told everyone it cost $250 because even AI experts hallucinate when tired. 

The federal government published an official health report stuffed with fake citations and OpenAI tags still visible like someone forgot to delete their browser history.

Yikes. 

(Infomercial voice.) But wait. There's more. (AI chaos.) 

The New York Times struck a licensing deal with Amazon in the middle of a 18-month long lawsuit against OpenAI for kinda exact same thing. 

Google rolled out email and video auto-summaries that actually work. Image AI got so realistic you literally cannot trust anything you see online anymore.

Voice AI learned when to shut up and listen. 

Annnnnnd Anthropic’s CEO warned half of entry-level jobs will vanish in five years right after launching their most powerful model yet.

And Apple?

They're taking what Bloomberg calls an "AI gap year" because they know they'd get absolutely destroyed if they tried to compete.

Wild week of AI updates, shorties. Here’s everything you need to know. 👇

Perplexity launched Labs, an agentic mode that spends about ten minutes on self-supervised tasks using deep web browsing, code execution, and visual generation to build you complete dashboards, websites, and slide presentations.

Zero coding required.

The AI agents organize data, apply formulas, generate charts, create documents, spreadsheets, and small websites automatically. Users access examples through a project gallery showcasing interactive war maps, stock portfolio dashboards comparing traditional versus AI-managed investments, and futuristic social media designs.

The assets tab organizes everything properly. Generated images, charts, CSV files, and code all sorted like a digital filing cabinet that makes sense. Unlike other tools that leave you scrambling to download creations.

Slide generation surprised us. Not design award material, but hits that "okay PowerPoint template" vibe for rapid business presentations. We tested it with ten news stories. The AI researched each topic, fact-checked claims, and delivered something actually usable.

What it means:

This pivot saves Perplexity from getting crushed by big tech.

We called it in January. They needed to evolve or die from being a traditional answers engine, as Google, OpenAI and now Claude closed that gap. 

Labs transforms them from search tool to productivity platform.

Hugging Face has introduced two new open-sourced robots:  the HopeJR, a fully programmable, open-source humanoid robot priced at $3,000. HopeJR can walk, manipulate objects, and features 66 independently controllable movements. Its open-source design allows users to modify both hardware and software.

They also announced the Reachy Mini, a desktop robot that moves its head, talks, and listens, priced at $300.

Tesla's Optimus Gen 2 costs around $25,000 and isn't even available yet. (If it is ever actually released as advertised.) 

Other advanced humanoids run hundreds of thousands.

Three grand hits the sweet spot where early adopters and universities can actually experiment.

What it means:

Mainstream home robotics adoption starts here.

At this price point, universities, small research labs, and ambitious hobbyists will push boundaries and create open-source improvements. Thousands of developers worldwide contributing beats waiting for one company to innovate.

We're predicting similar robots hit $1,000 within two years.

Google launched automatic email summary cards in Gmail. They appear at the top without tapping anything. Gemini proactively summarizes long threads and updates summaries as new replies arrive.

Finally.

Workplace admins control access through admin console. It's opt-in in EU, UK, Switzerland, and Japan but default in the US. The manual summary option still exists as a clickable chip.

Google also rolled out video summarization in Drive. Save a video, Gemini automatically summarizes it instantly. Users can chat with AI to request specific details like action items from recorded meetings or highlights from announcement videos.

Requires captions enabled. English only right now for Google Workspace and Google One AI Premium users.

What it means:

Google stopped building flashy demos and started solving real problems.

Shipping like the day after Black Friday, TBH. 

Email threads spiral out of control. Important videos bury crucial information in hours of footage. Manual summarization takes forever and misses details.

Knowledge workers could reclaim significant time currently lost to information processing. Even 30 minutes daily adds up.…..

And keep up with all these updates. Lolz. 

Whoopsies. 

The Department of Health and Human Services published an official health report led by Robert F. Kennedy Jr. packed with AI fabricated citations and made-up research studies. The Associated Press and the Washington Post exposed the SNAFU. 

Thirty-seven of 522 footnotes were repeated multiple times. Several URLs included "oai_site" tags linked to OpenAI. Digital fingerprints screaming AI-generated content.

Double doink. 

One study about overprescribing oral corticosteroids for children with asthma doesn't exist outside the report. Another claimed a 40-fold increase in childhood bipolar diagnoses tied to a psychiatric manual published years after the cited period.

The White House called it "minor formatting issues."

Minor?

Lolz. 

What it means:

This is catastrophic for federal credibility.

Government agencies are using AI to generate policy documents without verification. We're looking at a crisis of evidence-based governance extending far beyond health policy.

Someone fed prompts asking for supporting evidence, then copy-pasted results directly into official documents. Terrifying when you consider how this could spread across agencies.

The New York Times just reached a multiyear AI agreement with Amazon that allows the tech giant to use their editorial content across AI platforms. This is happening while they're simultaneously suing OpenAI and Microsoft for allegedly copying millions of their articles.

Talk about playing both sides of the AI revolution.

The deal enables Amazon to incorporate real-time summaries and short excerpts from The Times, New York Times Cooking, and The Athletic into Amazon products like Alexa. Amazon gets to train their proprietary foundational models on Times content, theoretically making Alexa less useless.

This represents The Times' first major licensing partnership with a tech giant. Remember, they filed that infamous lawsuit against Microsoft and OpenAI back in December 2023, asking courts to literally destroy the GPT technology entirely. Now they're licensing their content to another tech giant for the exact same purpose.

It's actually a pretty smart hedging strategy TBH. 

What it means:

Media companies finally realized there are only three realistic options when dealing with AI training.

Fight expensive lawsuits with uncertain outcomes. Block AI crawlers and watch your audience disappear as people stop finding your content. Or license your content and actually get paid for it.

Most AI crawlers ignore robots.txt files anyway. Even if you successfully block some crawlers, third-party datasets already contain your content and get used for training. Plus, if you want to show up in Google search results, you can't opt out of Google's AI training while staying in their search index.

You can't fight the tide, but you can ride the wave and get paid for it.

Black Forest Labs just launched FLUX.1 Kontext, and the image quality is so realistic that you should assume everything you see online is AI-generated from now on. We're not being dramatic here - the technology has genuinely crossed the line where distinguishing real from fake requires forensic analysis.

The new image generation suite includes two main models that deliver results up to eight times faster than leading competitors. Unlike previous open-source models from Black Forest Labs, these new versions aren't available for offline download. 

They're keeping this locked down in a private beta for safety and testing purposes, which tells you everything about how powerful and potentially dangerous this has become.

The company is offering 200 free credits through their model playground for testing. Black Forest Labs was founded by former StabilityAI team members and reportedly sought $100 million in funding at a $1 billion valuation.

Character consistency across multiple images is the real game-changer. Combined with video AI tools like Google's Veo 3, this becomes a complete deepfake creation pipeline that's accessible to basically anyone.

What it means:

The default assumption for any image you encounter online should now be "AI-generated until proven otherwise."

Seriously. This isn't hyperbole.

The technology has reached the point where distinguishing real from fake requires specialized knowledge and tools that most people simply don't possess. This adds fuel to the deepfake and misinformation epidemic that's already happening.

People are launching fraudulent GoFundMe campaigns using fake disaster photos. Blackmail schemes are using realistic fake compromising images. The general public took about 15 years to understand that things could be photoshopped.

ElevenLabs just launched Conversational AI 2.0, and the biggest improvement isn't what it says but when it knows to shut up and listen. Their new turn-taking model finally solves the awkward interruption problem that makes most voice AI feel robotic and frustrating.

This is actually a bigger deal than it sounds.

The updated platform features conversation flow recognition that can tell when humans are pausing to think versus actually finishing their thoughts. No more AI jumping in mid-sentence when you're just retrieving information or processing a complex question.

The new system includes integrated language detection for seamless multilingual conversations, built-in retrieval systems for accessing external knowledge bases, and multimodal communication supporting voice, text, or both simultaneously.

Enterprises can now automate large-scale outbound calling with batch capabilities for surveys, alerts, or personalized messages. Subscription pricing ranges from limited free tiers to business plans exceeding $1,300 monthly.

What it means:

Within a year, calling any customer service number will probably start with an AI voice agent.

This isn't necessarily a bad thing. It's definitely better than talking to someone in an overseas call center where you can't understand anything, or screaming "REPRESENTATIVE" at those terrible robotic phone trees.

The turn-taking advancement removes the primary friction point that prevented widespread adoption in customer-facing applications. Voice AI is genuinely transitioning from impressive demo technology to practical business infrastructure.

Companies can scale customer service cost-effectively without sacrificing quality. AI agents handle routine inquiries and seamlessly transfer complex issues to human agents with full context.

REPRESENTATIVE! 

Anthropic CEO Dario Amodei just warned that AI could eliminate up to 50% of all entry-level white collar jobs within five years. He made these predictions during a media tour that happened right after announcing Claude 4, which is... interesting timing.

Amodei predicted US unemployment could spike from the current 4.2% to potentially 10-20% due to AI-driven automation. He says society and lawmakers aren't acknowledging the threat to white collar jobs, and the impacts will hit faster than most people expect.

What it means:

The timing raises serious questions about corporate motivations versus genuine societal concern.

Whether these predictions prove accurate or not, they're already influencing business decisions and policy discussions around AI regulation. Corporate leaders are using job displacement warnings to justify accelerated automation and shape regulatory frameworks.

 We’ll save our takes on this one, as we’re doing a full episode on this topic for tomorrow’s show. 

According to Bloomberg's Mark Gurman, who basically never gets Apple predictions wrong, next week's WWDC will reveal only minimal AI advancements. Apple apparently realizes they're extremely far behind after going all-in on AI last year and failing spectacularly.

They're currently facing class action lawsuits because most of the AI features they announced and heavily marketed never actually materialized.

The biggest AI news expected? 

Apple opening their on-device foundation models with around 3 billion parameters to third-party developers. They're also renaming their operating systems to align with years (iOS 26, macOS 26) after reportedly spending millions on consultants to reach this groundbreaking revelation.

Gurman describes this as Apple's "gap year," which is legit hilarious. 

Apple is essentially sitting this one out because anything they announced would get laughed at compared to what Microsoft and Google just revealed at their recent conferences.

What it means:

This represents one of the biggest strategic failures in modern business history.

Apple had multiple years to see AI development coming, possessed more resources than virtually any competitor, and still completely missed the most important technology transition in decades. Their inability to deliver usable AI while falling years behind competitors is genuinely embarrassing.

The gap year strategy actually prevents them from making their position worse through overpromising and underdelivering again.