Google went NUTTY with Gemini 2.0 Flash Thinking.

Apple execs called their internal AI meetings "ugly" and pushed features to 2027, while employees report widespread burnout and frustration over features failing one-third of the time in testing. 

DeepSeek is literally taking employee passports away to prevent AI knowledge from leaving China.

Columbia's study caught Grok hallucinating a staggering 94% of the time, making it essentially an expensive random text generator masquerading as an authoritative source.

And that’s barely the surface of this week’s AI news. 

Sheeeesh.

Don’t waste hours each day trying to keep up and understand what it means for your business. 

On (most) Monday, we do that for you with our weekly AI news that matters. 

Ready to sort through it all? 

Let's get into this week's AI feast, shorties.👇

Google unexpectedly released a MASSIVE suite of AI updates on this past week. 

Nobody saw it coming.

Gemini 2.0 Flash Thinking brings improved reasoning, multimodal capabilities, and a one MILLION token context window for paid users who need to process extremely long documents or conversations without losing context.

The inline image generation is legit game-changing. Say "write a blog about Chicago tourism with images for each spot" and boom. Done.

Google's Deep Research got completely rebuilt to be significantly more accurate than its four-month-old predecessor, examining a handful of high-quality sources before determining the next research steps rather than casting an indiscriminate wide net across the internet.

Paid users get personalization that uses your search history. 

Some love it. Others find it creepy AF.

Best yet? 

Their tiny 24-billion parameter Gemma 3 model is crushing competitors 20 TIMES its size while using just a fraction of the compute power required by larger models that have hundreds of billions of parameters.

What it means

Google finally stopped playing catch-up.

They’ve entered the chat with OpenAI as the top tier AI providers after languishing behind in 2023 and the first half of 2024.

Gone are the days of Google marketing without their AI without substance. 

Google is straight shipping like a shipyard. 

Small language models are about to make "bigger is better" obsolete in the race for AI efficiency, potentially disrupting everything we thought we knew about large language model development and deployment costs.

Edge AI will go mainstream next year because of this breakthrough from Google. Real AI power without cloud dependency.

Elon Musk and OpenAI agreed to fast-track their trial to December 2025 and Northern California's U.S. District Court will host the showdown.

Musk claims OpenAI betrayed its "for humanity" mission when they restructured to raise billions for corporate profits instead of developing AI for the good of humanity as originally intended.

OpenAI needs that restructuring desperately if they want to continue competing in the increasingly expensive AI development race where their last fundraising round brought in $6.6 billion and they're discussing an additional $40 billion from SoftBank.

The judge rejected Musk's request to pause OpenAI's for-profit transition, suggesting the legal argument may be weaker than Musk's team had hoped.

(But like…. Just about everyone with their head above the Twitter waters knew this.) 

Elon’s $97 billion bid for OpenAI wasn't serious. At all.

With Musk pretty much running the U.S. government alongside President Trump, this fight is about to get political. 

Extremely political.

What it means

This lawsuit is theater. Pure theater.

If you want the full breakdown, we’ve covered it before.

Musk isn't protecting humanity.

He's hoping to use the legal system against a company that succeeded without him and now directly competes with his xAI venture, which has every incentive to slow OpenAI's progress through any means available.

A nonprofit changing course and turning into a for-profit is a tale as old as time. Nothing illegal about it.

So in retaliation?

Elon’s trying to make OpenAI DOGE his new political power. 

Musk hopes his government influence tips the judicial scales. That’s his only hope, as the rest of the dust he’s kicking up is baseless. 

Bloomberg's report on Apple's AI failures reads like a soap opera.

Executives are taking "intense personal accountability."

Senior director Robbie Walker called the situation "ugly" in a Siri team meeting. He acknowledged widespread employee burnout.

Features promised at the 2023 WWDC keynote are now delayed to 2026 or 2027, despite Apple having spent what reports indicate was upwards of a million dollars daily trying to develop their own internal models before eventually partnering with other AI providers.

Yikes.

Also, we told you Apple would fail in AI YEARS ago. So we’re not surprised. 

According to reports, internal testing showed Apple’s AI features failing ONE-THIRD of the time. Completely unacceptable by Apple standards.

Marketing made everything worse by hyping nonexistent features in commercials that consumers couldn't actually access, leading Apple to pull iPhone  ads showcasing capabilities that don't work in the real world.

What it means

Apple completely fumbled the AI bag.

So. 

Hard. 

Someone should make a movie about this disaster. They reportedly burned $1 MILLION A DAY on AI development that produced virtually nothing beyond basic summarization features for unread text messages.

Got nothing to show for it.

Their "high standards" excuse is BS masking the fact that they failed to anticipate how quickly generative AI would evolve and now find themselves years behind competitors who moved faster and accepted imperfect initial releases.

The delay to 2027 is a death sentence for Apple in the AI race. 

They'll be three generations behind by then.

The Apple Intelligence SNAFU might end up being the biggest misstep in the company’s history. 

OpenAI and Google are asking the government to basically skip copyright law for AI training.

Straight up.

Both submitted proposals to the White House as part of Trump's AI Action Plan, arguing that allowing access to copyrighted data is critical for national security in an increasingly competitive global AI landscape.

They're playing the China card hard — if they can’t openly train their models on copyrighted data, China will. And then the U.S. will lose. 

Google says copyright and privacy policies shouldn't block AI training data, emphasizing that fair use policies have been vital for developing cutting-edge AI systems that maintain America's technological edge.

Anthropic ignored copyright issues entirely in their proposal and focused on national security instead.

The New York Times lawsuit looms large over the entire industry because if OpenAI loses, courts could in theory instruct them to destroy the GPT technology that powers tens of thousands of American businesses, creating potential economic chaos overnight.

(Yes, really. But it’s very unlikely to happen, but read/listen/watch more here if you want.

What it means

• Is Copyright antiquated? 

• Should Big Tech be able to basically ignore copyright in the name of AI? 

• And if AI labs can’t train on copyrighted data, then what happens to the whole AI industry?

• And America’s move-first advantage? 

The answers to all of those questions are messy AF. 

(Glad we’re not the ones deciding.) 

DeepSeek is reportedly taking employees' passports away.

Literally. 

According to reports, China wants to prevent data leaks and keep AI knowledge from leaving the country, reflecting growing concerns over data security and geopolitical risks in the increasingly competitive AI industry.

Chinese authorities in Zhejiang province are reportedly screening investors before they can even MEET Deepseek company management, adding another layer of control over sensitive AI-related operations and potential foreign influence.

They kinda fibbed about spending only $5.5 million to train their model when in reality they only disclosed the cost of the final training run, not the massive infrastructure and development costs required beforehand.

Oh, don’t forget Deepseek's parent company is a hedge fund called High Flyer. They make money through algorithmic trading, and the U.S. economy dropped $3 trillion in market cap after the Deepseek “news.” 

Google DeepMind and other research companies quickly debunked the $5.5 million claim.

But the damage was done. 

What it means

Using DeepSeek's API or website?

You're giving your data directly to the Chinese government.

Their terms of service explicitly allow this under Chinese laws that require companies to share user data with the government, creating a massive security risk for any Western company foolish enough to upload sensitive information.

Passport confiscation proves they're state-controlled. Not just a regular AI company.

The "$5.5 million training cost" story was state-sponsored disinformation. Designed to crash Western tech markets.

It worked.

We’re based in the U.S. and help U.S. companies win with AI. 

So is our take slanted that way?

Sure.

But, you have to understand AI is MUCH more about geopolitical power than it is about LLM outputs and benchmarks. 

OpenAI just released a game-changing agents SDK.

Building AI agents just got way simpler.

Their platform provides a complete stack. No more juggling multiple frameworks and tools for companies that want to deploy AI agents without hiring an army of specialized developers to handle complex integration challenges.

The Responses API integrates with web search, file search, and computer use tools. Same tech powering ChatGPT, now available to businesses.

Reliability has been AI agents' biggest problem.

OpenAI's solution?

Opening their ecosystem to external developers who can contribute innovative solutions to overcome agent failures and inconsistencies that have plagued early adoption.

Their new file search tool directly challenges specialized RAG database vendors by offering built-in retrieval augmented generation capabilities that allow enterprises to upload data directly into OpenAI's system, gaining the transparency and traceability that regulated industries like finance and healthcare demand.

What it means

This is one of OpenAI's most business-focused moves yet.

Absolutely brilliant.

The RAG startup ecosystem is toast when hundreds of vector database companies that raised millions in venture funding now find themselves competing directly with OpenAI's native capabilities that offer seamless integration with the world's most popular AI models.

Good luck. 

Enterprise AI adoption will explode by Q4 2025. Technical barriers? Gone.

OpenAI is becoming the AWS of AI. The infrastructure layer everyone builds on.

Columbia's journalism school just dropped receipts on AI chatbots.

They're terrible at retrieving accurate news, according to their new study. 

Nearly 60% of queries across eight AI tools produced INCORRECT answers when tested on basic factual questions that should have been straightforward to verify through reliable sources.

Grok 3? Hallucinating 94% of the time.

Yikes. Not a typo.

Premium chatbots gave MORE confidently incorrect answers than free tools. While charging subscription fees from customers who expect higher accuracy in exchange for their monthly payments.

ChatGPT incorrectly identified 134 articles but only declined to answer in 15 out of 200 instances, revealing a fundamental inability to recognize its own limitations or signal uncertainty when faced with questions it couldn't accurately address.

Five chatbots bypassed publishers' robot exclusion protocols, accessing content from websites that explicitly blocked their crawlers, with Perplexity Pro being the worst offender by correctly identifying excerpts from restricted articles nearly a third of the time.

What it means

The entire AI search category is murky. 

Perplexity's $7 billion valuation doesn’t seem super serious when you realize their premium product uses other AI labs’ models and is wrong more often than right, despite positioning itself as a reliable research assistant worth paying for.

Grok is making stuff up 94% of the time, according to the study. 

Gulp. 

We'll see lawsuits next year from businesses who lost money from hallucinated AI search results that led to costly mistakes and decisions based on completely fabricated information.

Humans — stay in the loop. 

(Also, never solely rely on AI as your single source of truth. Just like you would never rely on a single internet website as a single source of truth. Always verify.) 

OpenAI is calling for the U.S. government to BAN DeepSeek.

No subtlety there.

They accused DeepSeek of being "state-sponsored and state-controlled" in a letter to the White House, raising serious concerns about security risks tied to Chinese laws requiring companies to share user data with the government.

They warned DeepSeek's R-One model could be weaponized by the Chinese Communist Party against critical infrastructure and high-risk applications in the United States and other Western countries.

DeepSeek is already banned in Italy, Ireland, South Korea, Australia, and parts of the U.S. government due to similar concerns about data security and potential state influence.

OpenAI says DeepSeek violated their terms by distilling knowledge from OpenAI's models. Freeloading off their compute costs.

What it means

The AI cold war just went hot.

OpenAI isn't competing with DeepSeek. They're trying to get Chinese AI banned entirely from the U.S. as part of a broader geopolitical struggle over AI dominance and data control.

The security concerns are legit. Using DeepSeek's API or online chatbot means handing your data to the Chinese government. 

This split will create two separate AI ecosystems by 2026. 

Companies will have to pick sides.

Zoom announced 45+ new AI features at Enterprise Connect.

With the latest releases, Zoom REALLY wants to be more than just video calls.

Their AI Companion now handles complex tasks. Calendar management, clip generation, meeting notes, writing assistance.

A $12/month add-on launches in April with customized meeting templates and personalized avatars, targeting businesses that want to enhance their virtual collaboration capabilities without switching to Microsoft or Google's ecosystems.

They're creating industry-specific solutions for healthcare workers, clinicians, and educators who have unique professional needs that generic AI tools might not adequately address.

Small language models paired with third-party AI systems power the whole thing, reflecting Zoom's broader strategy to integrate more efficient AI models rather than building massive systems from scratch.

What it means

Zoom is clawing uphill to be included in the AI collaboration party after spending most of the pandemic focusing on video improvements.

Now, they wanna play in the enterprise AI and collaboration sandbox with Microsoft and Google.

Their $12 monthly could be dead on arrival when companies already pay for Microsoft 365 Copilot or Google Workspace AI that integrate seamlessly across multiple applications their employees use daily.

Truth be told — we love Zoom and have always used their paid products. 

But, we’re not sure what companies are going to pay for their AI when the overwhelming majority of users are already getting these built-in AI features in Microsoft or Google.