• Everyday AI
  • Posts
  • Google’s surprise AI releases: What’s new and how it changes the LLM race

Google’s surprise AI releases: What’s new and how it changes the LLM race

AI search tools are getting citations wrong, OpenAI asks feds to ban DeepSeek, Elon wants to replace government workers with AI.

Outsmart The Future

Sup y’all! 👋

Google went wild. Make sure to check out today’s episode. Sheeeesh.

Also, I’ll be LIVE at NVIDIA’s GTC conference next week. So our normal livestream/newsletter times might be a lil different.

FYI — you can attend virtually for free if you wanna dive into it for yourself. As you might know, NVIDIA literally powers most of the world’s GenAI, so there should be some HUGE news.

Are you or your company gonna go? If so, make sure to hit me up!

Are you going to NVIDIA's GTC Conference next week?

Are you going to NVIDIA's GTC Conference next week?

Login or Subscribe to participate in polls.

✌️

(Let’s connect on LinkedIn. Just tell me you’re from the newsletter. Stranger danger.)

Today in Everyday AI
7 minute read

🎙 Daily Podcast Episode: Without fanfare, Google just went all AI March Madness on us. Here’s what they announced. (A ton for free!) Go listen.

🕵️‍♂️ Fresh Finds: What to expect at NVIDIA GTC, Google's new Deep Research 2.0 (somewhat) available to free users and a new more ethical AI video model. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: AI search tools are getting citations wrong, OpenAI asks feds to ban DeepSeek, Elon wants to replace government workers with AI. Read on for Byte Sized News.

🧠 Leverage AI: OK: Google went crazy with AI releases outta nowhere. Here’s what it means and how you can take advantage. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about Google drops even more Gemini 2.0 updates, OpenAI pressures U.S. government for less AI regulation, Adobe not cashing in on AI and more.. Check it here!

Google’s surprise AI releases: What’s new and how it changes the LLM race

Google woke up mid March and chose madness.

  • Gemini is refreshed.

  • Deep Research is supercharged.

  • Gemma 3 is a tiny heavyweight.

Whuuuuuuut.

Who saw this coming?

Also on the pod today:

Google’s Deep Research goes 2.0 💪
Multimodal text and image is easy 🖼️
Google Robotics drops 🤖

It’ll be worth your 53 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Freepik released an AI video upscaler, AgentAI is a professional network for AI agents and Amiry is your AI-powered city guide.

Prompt EngineeringCarnegie Mellon’s LCPO technique trains AI to reason smarter, not longer—saving costs without sacrificing accuracy. Is less really more?

NVIDIA GTC What to expect at NVIDIA’s GTC Conference. Remember, you can attend for free virtually by signing up here.

AI and HumansOverconfidence in AI tools like ChatGPT may reduce critical thinking, a study finds—could trusting generative AI too much be sabotaging your skills?

Free AI Tools – Google’s new version of Deep Research, powered by Gemini 2.0 Flash Thinking, includes a few free searches a month for non-paid users.

OpenAI Updates — OpenAI opened up the ‘Work with Apps’ feature to free users on Mac.

AI VideosCould Moonvalley’s new Marey model redefine AI video with ethical, Hollywood-grade innovation?

AI Assistants – Android users can now ditch Google Gemini for ChatGPT as their default assistant—get quick access with a simple swipe or press. Ready to make the switch?

AI Voice Models — A new open-source AI voice model is causing a stir.

1. AI Search Tools Under Fire for Broken Links and Syndication Issues 🔗

According to Columbia Journalism Review, Google's Gemini and Grok 3 are under scrutiny after 154 out of 200 citations from Grok led to broken or fabricated URLs—a glaring issue for publishers navigating licensing agreements with AI companies.

This raises broader concerns about the future of content attribution and the balance between innovation and fair traffic distribution for original publishers. Time magazine COO Mark Howard acknowledged these challenges but pointed out user responsibility in trusting free tools, while OpenAI and Microsoft provided vague assurances without addressing the core issues.

2. OpenAI asks Gov to ban DeepSeek 🐋

OpenAI has accused Chinese AI startup DeepSeek of being "state-sponsored" and a potential security risk, urging the U.S. government to ban its models as part of the Trump Administration’s AI Action Plan.

OpenAI claims Chinese laws could force DeepSeek to share user data with the CCP, raising fears of privacy breaches and manipulated infrastructure. The showdown, fueled by concerns about global AI dominance, comes after DeepSeek’s founder met with President Xi Jinping, sparking fresh scrutiny over political ties.

3. China Tightens Grip on Rising AI Star DeepSeek 🥹

China is reportedy ramping up oversight of its AI darling DeepSeek after the startup gained global attention in January with its groundbreaking reasoning model, R1. According to The Information, Beijing has imposed stricter controls, including restricting some employees from traveling abroad and screening potential investors.

DeepSeek’s parent company, High-Flyer, is reportedly holding onto key staff passports to enforce these measures. These developments reflect China's growing concerns over intellectual property leaks amid rising AI competition, particularly with the U.S.

4. Google: AI Rules? Trust Us, Bro 🧑‍🎓

Google has released a bold new policy stance urging lighter federal regulation on AI, claiming current policymaking is overly focused on risks. The tech giant argues that a national AI framework is essential for innovation and warns against state-level "patchwork" laws that complicate compliance.

According to reports, Google opposes holding developers liable for AI misuse and views global transparency rules like the EU's AI Act as threats to trade secrets, striking a similar tone as OpenAI. The company calls for U.S. diplomatic efforts to promote business-friendly AI policies worldwide, aiming to shape regulations that align with its vision of innovation without heavy oversight.

5. Musk's AI Overhaul of U.S. Government Sparks Controversy 😵

Elon Musk, leading the Department of Government Efficiency (DOGE), is reportedly slashing federal jobs and replacing them with AI tools, raising alarm among experts, according to Al Jazeera.

Critics warn that AI-driven decisions about employment and social services could lead to bias, errors, and harm, especially for marginalized communities. Experts also highlight the lack of transparency and testing in these AI systems, pointing to past failures in Michigan’s unemployment system and biased policing algorithms as cautionary tales.

🦾How You Can Leverage:

Holy processing power, Batman. 

Google just dropped a March Madness-worthy AI barrage without warning. 

No flashy event. No hype. Just new and updated models and features.

Sheesh. 

The most jaw-dropping part that many of us may not use but will impact us all? 

Their new 27B parameter Gemma 3 model is BEATING models 30 TIMES its size in human preference tests. We're talking about a model so efficient you could run it on a $3,000 NVIDIA device.

Not millions of dollars. Three thousand.

Google’s silent drops this week weren’t incremental update. 

It's Google quietly proving everyone else has been playing the wrong game.

We broke it all down on today’s show. Here’s what you need to know. 👇

 

1. Gemma 3: The Little Giant-Killer ⚔️

Forget everything you thought you knew about model size and performance.

Seriously. Ctrl+Alt+Delete that knowledge shorties. 

Gemma 3 just delivered the most decisive proof yet that we've been looking at AI all wrong. This "tiny" 27B parameter OPEN SOURCE! model is outperforming DeepSeek v3's massive 671B parameters in human preference tests.

More than 20X smaller, yet humans consistently prefer its responses.

The Elo scores don't lie.

Gemma 3 sits in the top 10 models GLOBALLY despite being a fraction of the size of its competitors.Like…. In size it’s miniscule. 

In performance? Masssssssive. 

Three years ago, we would have said running a top-10 LLM locally would cost millions. 

Now? It’s a small model and you can run it for three grand on DIGITS from NVIDIA

This demolishes the "bigger is better" myth. 

Google proved you can achieve breakthrough performance without needing a country's power grid.

Try This:

Download the 4B parameter version of Gemma 3 through HuggingFace or Ollama for local testing. 

Compare it against your current cloud API tasks for speed and quality.

 If you're feeling adventurous, start fine-tuning it on your domain-specific data. This wasn't possible before without massive budgets.

ow it is.

2. Your Content Team Just Got Downsized ⌨️

Remember creating a blog post with images?

  • Writing and researching: 6 hours.

  • Finding photos you can use: 2 hours.

  • Editing and placing images: 2 hours.

  • Proofreading: 30 minutes.

That workflow just collapsed into seconds.

Gemini 2.0 Flash Thinking with in-line image generation doesn't just save time. Inside AI Studio, it has image+text outputs. 

It obliterates an entire category of creative busywork.

We tested it with a simple prompt about Chicago tourism. In seconds—not hours—it produced a comprehensive article with perfectly matched AI-generated photos of landmarks, positioned exactly where they belonged.

We were all like…. whuuuuuuu?!?

The million-token context window for paid users means you can feed Gemini hundreds of pages of company content first. It'll understand your brand voice before generating anything new.

This isn't automation. It's creative multiplication.

Try This: 

Take your most successful blog post and tell Gemini 2.0 Flash Thinking to create a visually enhanced version with integrated images. 

Publish both versions and compare engagement metrics. Then recalculate your content calendar timeline with this new capability factored in. 

You'll be shocked at the hours saved.

3. Robot Revolution: Physical World AI Is Here 🤖

While everyone obsessed over chatbots, Google quietly became the first major AI lab to release a robotics model.

Google just released Gemini Robotics, running on Gemini 2.0, which integrates multimodal reasoning with physical actions. Robots can now understand natural language commands and adapt to environments in real-time.

The most fascinating part?

Embodied reasoning.

The new Gemini Robotics ER model adds spatial understanding that lets robots perform complex tasks requiring fine motor skills. Folding origami. Handling coffee mugs. Things that seemed impossible last year.

What's really happening is more strategic than "Google making robots." They're building their world model by collecting physical interaction data from robotics companies that use their freshly updated model. 

Every robot interaction teaches their AI how the real world works. That knowledge improves everything from text-to-video generation to creative tools.

Try This: 

Identify three repetitive physical tasks in your workflow that require fine motor skills but minimal complex decisions. 

Calculate the hours spent on these annually. 

Research which emerging robotics platforms might fit these needs. 

Understanding your "automation readiness" now gives you a massive strategic advantage for when these tools become mainstream.

Our Takeaway? 

While others bragged about parameter counts, Google solved problems that actually matter.

Edge AI is no longer second-best. It might be superior to cloud models.

Multimodal integration isn't a feature. It's becoming the default.

Physical world understanding isn't future tech. It's happening now.

The biggest shock? The democratization timeline just warp-speeded forward.

Two years ago, local AI seemed like a 2030 milestone. Now it's a $3,000 purchase away.

This means every company can build AI capabilities without massive infrastructure investments.

The future arrived early. And it runs on your laptop.

Numbers to watch

€1 billion

The Dutch government plans €1 billion in cuts while turning to AI for efficiency—fewer rules, less bureaucracy.

Reply

or to participate.