Welcome, humans.

AI slop is coming at us hot and heavy these days. For example, check out this video of straight fuel being pumped into the ocean—can you tell this is AI? 

Or what about these “immersive” movie theater experiences?

It’s gotten so bad that some people need advice on how to tell their friends or family why images like this adorable (but very fake) heart cat are not real.

This is an issue, because OpenAI’s image generator just got A LOT better, and this means higher quality AI content will be even easier to create.

Example: OpenAI’s new image generator doesn’t restrict generating celebrities, though Sam says this might change.

As we covered on Monday, Google has its own AI top-tier image generator, and couldn’t help but take a few shots at OpenAI’s launch with it. V petty. V impressed.

With AI images getting this good, we may have to abandon the internet entirely.

Here’s what you need to know about AI today:

• Gemini 2.5 Pro launched and topped the AI leaderboards.

• OpenAI released a wild new image generator.

• There’s a new test to stump today’s top AI.

• Earth AI found rare minerals in an unexpected place.

The AI arms race just made a new turn, with Gemini 2.5 vs DeepSeek V3.1 at the front.

We've got a serious AI showdown happening this week. In one corner: Google DeepMind's new Gemini 2.5, their “most intelligent model” yet.

In the other: DeepSeek's V3-0324 (let's call it V3.1), a Chinese open-source challenger that can be run on prosumer-level hardware. 

Thoughts and Prayers go out to OpenAI right now...this week is shaping up to be their nightmare incarnate.

First up, Gemini 2.5 = the new Benchmark King. 

1. This new model is now #1 on the LM Arena by a HUGE margin—the largest score jump ever (+40 points versus Grok-3 and GPT-4.5).

2. It leads in math and science benchmarks like GPQA and AIME 2025.

3. It also scores a state-of-the-art 18.8% on Humanity's Last Exam (an expert-level benchmark designed around the frontier of human knowledge).

To try it, go to Google AI Studio, and select “Gemini 2.5 Pro Experimental 03-25.” It has a massive context window of 1M tokens (~700K words), which is ridiculously good (for FREE, we might add).

Up next, DeepSeek V3.1 shows open-source means business. The Chinese AI startup quietly released their 641-GB model on Hugging Face.

Two things make this release a big deal: 

1. It comes with an MIT license—making it free for commercial use. 

2. It can run directly on a $10K Mac Studio at about 20 tokens (~15 words) per second—that’s not exactly cheap, nor is it fast, but it runs! 

Performance-wise, researcher Xeophon proclaimed it's “now the best non-reasoning model, dethroning [Claude] Sonnet 3.5.”

You can try DeepSeek on their cloud or on OpenRouter (thanks, Simon Willison!) 

Why this matters: According to Lee Kai-fu, CEO of 01.AI and former head of Google China, China has narrowed the AI development gap with the US to just three months in some areas. 

This shift is accelerating China's AI capabilities at a pace that has shocked Western observers, as Alibaba, Tencent, and Baidu have all released open-source models (for extremely cheap, if not free).

Our take: The race is no longer just about who has the smartest AI—it's about who can get it into the most hands.

We now understand why Sam recently said he thinks being a billion user consumer tech company is more valuable than creating the next state of the art model… the sharks are circling.  

FROM OUR PARTNERS

💼 Want to become an AI Consultant? 

This stat is WILD: The AI consulting market is projected to explode from $6.9 billion to $54.7 billion by 2032. Want to capture this opportunity?

So how do you, an AI enthusiast, become a handsomely compensated AI consultant?

Well, The AI Consultancy Project from Innovating with AI has already helped 300+ students build successful consulting businesses. 

Current students love:

• Tools and frameworks to find and serve clients.

• 6-month roadmap to building a 6-figure business.

• Methods to land your first client in as little as 3 days.

As a Neuron reader, you get early access to the next enrollment cycle!

Take advantage of the AI boom—request early access to The AI Consultancy Project ASAP!

Prompt Tip of the Day

Want six years of AI prompting knowledge in less than an hour? Watch this video from Nick Saraev, who has a whole host of recommendations for you.

Our favorite tip? Use the “playground” versions of your favorite AI tool, NOT the consumer version everybody uses. Why? Playgrounds let you see “under the hood” and tweak the controls that the main consumer apps hide from you.

That would be ChatGPT Playground, Google AI Studio (shared above), or the Anthropic Console for Claude and La Plateforme for Mistral.

Treats To Try.

1. *Promptimize is like Grammarly, but for AI prompts—it's a browser extension that transforms your basic prompts into expert ones, helping you get way better results from platforms like ChatGPT, Claude, and Gemini. Check it here.

2. You can use ChatGPT-4o’s new image mode to merge two images together—here’s a demo of it generating the anatomy of a typical Redditor and 9 other examples from X.

3. AiSDR books meetings with your ideal prospects through personalized conversations using their LinkedIn activity and your HubSpot data.

4. Agora is a search engine for Shopify stores (since standard search engines aren't optimized for product discovery across smaller retailers)—now it has an API to help you search, purchase, and track products across 25K online stores.

5. Method helps financial institutions connect to their customers' credit accounts and loans across 15K+ institutions so you can instantly transfer balances, pay off loans, or refinance debt (raised $42M).

6. Wordly provides live translation and captions in 60+ languages—its now being used to offer real-time translations for a live theater production in NYC.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

• Alibaba’s Chairman Joe Tsai says the estimated ~$1T AI data center buildout is “the beginning of some kind of bubble” (or is it the end?).

• There’s a new reasoning test, called ARC-AGI-2, meant to challenge AI systems with problems that are easy for humans, but hard for today’s AI—here is the current ARGI-AGI Leaderboard, showing where today’s models land between % correct and cost to complete.

• A startup called Earth AI’s algorithms discovered valuable critical minerals in Australian regions that mining companies had overlooked for decades.

FROM OUR PARTNERs

⛽Fuel MCP with data from anywhere

Join Unstructured’s live session to learn

The key components of an MCP server and how to build and deploy one

• How MCP standardizes data access and enhances interoperability between AI tools

• How to integrate MCP with the Unstructured API for automated document processing

Register here to join live or to get the recording.

A Cat's Commentary.