Meta's reportedly shopping around to use Google's Gemini and OpenAI's models in their own Meta AI assistant.

Awkward. 

Marc Benioff just went full hypocrite mode, cutting 4,000 customer service jobs after claiming two months ago that AI wouldn't cause mass unemployment.

Meanwhile?

Google dropped AI updates so viral that their "nano banana" model jumped 170 points ahead of every competitor on the LMArena leaderboard, while also going after Duolingo's turf with Google Translate.

That’s just the tip. 

What a wild week of AI news.

Here's what matters, y’all. 

How do you signal you're behind in AI without actually saying it?

The Information reports that Meta's Superintelligence Labs leaders have been exploring integrating Google's Gemini and potentially OpenAI's models to deliver better text-based responses in Meta AI. This would be a "near-term strategy" to raise response quality while they work on Llama 5.

Translation: Their current tech isn't cutting it.

A Meta spokesperson said they're taking an "all-of-the-above approach" combining partnerships with building their own systems, which sounds way better than "we need help from our rivals."

Meta already uses Anthropic's Claude for internal coding, so they're clearly comfortable mixing providers where they see advantages, but offering competitor tech to consumers is next level.

What it means: 

There's a steep quality drop-off from Google and OpenAI to everyone else in AI.

Meta spending billions on talent and compute but still needing competitor models shows just how hard this game is.

If Meta can't compete with their resources, most companies are essentially choosing between OpenAI and Google - everyone else is fighting for scraps.

Why did Google just turn their translate app into a Duolingo killer?

Google rolled out AI-powered practice sessions inside Google Translate that adapt to your skill level and let you choose listening or speaking drills. You set learning goals like "getting to know a host family" and it generates customized scenarios.

Smart move.

The feature launched in beta for English speakers practicing Spanish and French, plus Spanish/French/Portuguese speakers working on English, with Google saying it's "complementary" to other learning methods.

But here's the thing - Duolingo's stock dropped 3% on the announcement before recovering, showing investors the threat. With 130 million active users but only 10% paying, Duolingo's got a problem.

Google processes 1 trillion words daily across Translate, Search, and Lens, so they've got the data advantage plus it's free versus Duolingo's subscription model.

What it means: 

Google realizes they can destroy established markets by adding features to apps they already own.

Unlike Duolingo's gamified lessons that mainly check pronunciation, this offers conversational AI tutoring at zero cost.

Google is systematically targeting startup territories they could have dominated years ago if they'd connected the dots sooner.

Remember when Reuters exposed Meta's internal policy encouraging romantic conversations with minors?

Meta is rolling out new protections to stop teen users and its AI chatbot from discussing self-harm, suicide, and disordered eating over the next few weeks. They're also blocking teens from user-made chatbot characters that could host inappropriate conversations.

About dang time. And like we talked about last week, the fact that this was reportedly in Meta’s policy in the first place is absolutely disgusting. 

The company said they added guardrails to redirect teens to expert resources after that absolutely disgusting internal policy that literally outlined how their AI should have sensual conversations with children.

Meta called it an "error" when they got caught, but Reuters reported at least four different leaders across departments had signed off on the language. That's not an error - that's an approved company policy.

U.S. Senator Josh Hawley set a deadline for internal documents from Meta this month. 

(Tick tock Meta.) 

What it means: 

Meta's original policy was so disturbing it reveals fundamental problems with their approach to user safety.

Multiple executives approving romantic chatbot guidelines for minors should result in firings, not just policy tweaks.

This was their default stance until public outrage forced changes - imagine what other policies they have that haven't been exposed yet. 

How do you go from "humans in the loop" to "I need fewer heads" in eight weeks?

Salesforce CEO Marc Benioff revealed that AI agents let him slash customer support from 9,000 to 5,000 employees, with AI now handling 50% of customer conversations versus zero a year ago.

Two months ago, Benioff told Fortune that AI accuracy was only 90% and companies needed to keep humans involved. Now he's bragging about cutting nearly half his support staff because he "needs fewer heads."

Either he was lying about AI limitations or the tech advanced impossibly fast - neither looks good.

The CEO also revealed Salesforce had over 100 million uncalled leads from 26 years ago because they lacked staffing, but now AI agents call back everyone who contacts them.

His stock being down 22% year-to-date while other AI companies soared makes this jobs cut announcement feel extra desperate for investor validation.

What it means: 

When CEOs flip this dramatically on AI employment impact within weeks, it signals scrambling to show cost-cutting like competitors.

Benioff's credibility on AI job displacement just took a massive hit after such a public reversal.

This is classic CEO behavior when under pressure - say whatever sounds good for the current audience.

What's the difference between today's clunky AI phone calls and what's coming next?

OpenAI launched the general availability of its Realtime API alongside gpt-realtime, a speech-to-speech model for production-grade voice agents that supports phone calling integration and image inputs.

Game changer.

Traditional voice AI systems convert speech-to-text, process in text, then convert back to speech - multiple steps that create delays and lose conversational nuance.

This runs a single voice pipeline that handles mid-sentence language switching, better alphanumeric transcription, and tone control with two exclusive voices called Cedar and Martin.

Companies can now plug voice agents into existing phone systems with minimal custom work, meaning your bank could deploy this tech instead of making you wait on hold for human agents.

What it means:

Customer service is about to see a massive quality upgrade as this tech replaces the terrible AI phone systems currently deployed.

The gap between current AI phone agents and what's now possible just became enormous - expect rapid adoption.

Voice-to-voice processing preserves emotional context and conversational flow in ways previous tech couldn't match.

How do you blow a massive AI acquisition in just months?

TechCrunch reports that Scale AI's former SVP Ruben Meyer left Meta after only two months despite being part of the core TBD Labs team "from day one," citing personal reasons.

Rut-roh. 

Despite the $14.3 billion investment and 49% equity stake, five sources say Meta's labs are actively using Scale AI's competitors Surge and Mercur for data labeling, with some researchers viewing Scale's quality as inferior.

Multiple high-profile researchers including Rishabh Agarwal, Chaya Nayak, and Rohan Varma have reportedly left after signing those NBA-level contracts, walking away from tens of millions in compensation.

Inside Meta, sources describe "rising turbulence" with new hires frustrated by big company bureaucracy and parts of the previous GenAI team seeing reduced scope.

What it means:

Throwing billions at top researchers doesn't work if your company culture drives them away within weeks.

These people are leaving generational wealth on the table, suggesting serious internal issues beyond normal corporate friction.

Meta's strategy of outspending everyone for talent is spectacularly backfiring when the talent keeps quitting. 

Will LLMs eat the journalism industry, or feed it? 

Well, Perplexity is trying to do the latter with its newest Comet Plus program. 

The Wall Street Journal reports Perplexity is launching Comet Plus, a $5 monthly subscription sharing 80% of revenue with publishers when their AI uses publisher content to answer questions.

Finally.

The plan covers both traditional indexed traffic and new "agent traffic" from actions like calendar scanning and article suggestions - interactions that bypass ads and aren't covered by existing publisher deals.

Existing Pro and Max subscribers automatically contribute since Comet Plus is included, drawing from a $42+ million revenue pool that grows with signups.

What it means:

We’ve been saying this since Day 1 – publishers have three choices with AI - sue, partner, or die.

Since many publications are losing 30-70% of traffic to AI, programs like this might be their only lifeline.

We need partnerships like this to prevent the internet from becoming pure AI slop trained on increasingly low-quality content.

How many video production jobs just vanished?

Google added AI avatars and editing features to Google Vids, letting teams write scripts, select avatars, and generate complete videos with live collaboration like Google Docs.

Brutal for video editors.

New features include noise cancellation, background options, and transcript trimming that automatically removes filler words and awkward pauses - the tedious work that used to take professionals hours.

You can also record yourself and let AI fix lighting, sound, and pacing, with Google executives saying Vids can correct talking speed and filler usage in clicks.

Basic features are free while premium ones require paid Gemini subscriptions, making professional-grade video creation accessible to any small business that couldn't afford agencies.

What it means:

Google just kinda eliminated some entry-level video editing, production, and voiceover opportunities in one update.

Companies like HeyGen and Descript that spent years perfecting these features got steamrolled by Google's resources and integration advantages.

On the flip side, small businesses can now create compelling content without professionals, but thousands of creative jobs disappear in the process.

Why are 22-year-olds getting crushed while 40-year-olds thrive?

Stanford economists analyzed ADP payroll data from 25 million workers and found early-career workers in AI-exposed jobs experienced a 13% employment decline since late 2022 while older workers in the same fields grew 6-9%.

Harsh reality check.

The biggest hits are in software engineering and customer service, where entry-level roles dropped 20% while employment for older workers in identical positions actually increased.

Lead researcher Erik Brynjolfsson says AI overlaps heavily with "book learning" that college grads have, while experienced workers possess tacit knowledge and organizational skills AI can't replicate yet.

This isn't about AI replacing all jobs. Nope. It's more about creating a brutal generational divide where experience suddenly became way more valuable than credentials.

What it means: 

Entry-level workers are losing the stepping stones they need to build expertise and climb career ladders.

Companies are keeping senior workers who understand business context while AI handles tasks that used to be training opportunities.

Young workers who learn to use AI effectively can be more productive, but those just doing what AI already does have little value-add left.

What makes an AI model so viral it jumps 170 points ahead of every competitor?

Google released Gemini 2.5 Flash Image, previously called "nano banana," and it dominated the LM Arena leaderboard with the biggest quality jump in AI image editing history.

Completely bananas.

The model is multimodal by default with character consistency that preserves exact details when you upload photos and edit them with natural language - no more facial feature changes for safety like other models.

You can combine multiple objects in single photos while maintaining consistent styling, and Google embraced the community nickname by adding banana icons and using "nano banana" in their interface.

At 100 free daily edits through AI Studio and under 4 cents per image via API, it costs a fraction of what less capable competitors charge.

What it means: 

This represents a massive leap in creative AI that puts Google way ahead of every image editing competitor.

Natural language photo editing with perfect character consistency opens entirely new creative workflows that weren't possible before.

Google's pricing makes professional-grade AI image editing accessible to everyone while competitors charge dollars for inferior results.