Marketing ideas for marketers who hate boring

The best marketing ideas come from marketers who live it.

That’s what this newsletter delivers.

The Marketing Millennials is a look inside what’s working right now for other marketers. No theory. No fluff. Just real insights and ideas you can actually use—from marketers who’ve been there, done that, and are sharing the playbook.

Every newsletter is written by Daniel Murray, a marketer obsessed with what goes into great marketing. Expect fresh takes, hot topics, and the kind of stuff you’ll want to steal for your next campaign.

Because marketing shouldn’t feel like guesswork. And you shouldn’t have to dig for the good stuff.

Sign Up Free

Today:

• China’s KIMI K2 Shocks Industry

• OpenAI Loses Windsurf to Google

• Microsoft Unleashes Fast Edge Model

• SpaceX Plans $2B xAI Investment

• Google Workspace Adds Gemini Features

KIMI just BROKE the AI Industry…

Kimmy K2, a new trillion-parameter open-source AI model from China, is making waves for its powerful coding and simulation abilities. It generates complex interactive outputs like 3D Earth simulations and defense games—all in one go.

Though it lacks reasoning for now, it shows benchmark-beating performance and runs locally on Apple M3 hardware. With future upgrades expected, Kimmy K2 represents China’s growing leadership in efficient, open-source, agentic AI development.

WATCH THE VIDEO ON YOUTUBE

OpenAI Loses Windsurf to Google

OpenAI’s planned $3B acquisition of AI coding startup Windsurf has collapsed. Instead, Google is hiring Windsurf’s CEO Varun Mohan, cofounder Douglas Chen, and key team members to join DeepMind and work on agentic coding for its Gemini AI. Windsurf will remain independent, with new leadership stepping in. Google gains access to some of Windsurf’s tech through a non-exclusive license, fueling its AI coding ambitions.

Why It Matters

1. Talent Shift: Key AI coding talent moves from OpenAI’s orbit to Google DeepMind, shifting innovation momentum.
   

2. Agentic Coding Race: Google accelerates its Gemini agentic coding efforts, a critical AI frontier.
   

3. Strategic Realignment: The deal collapse highlights tensions in AI partnerships, especially between OpenAI and Microsoft.

Microsoft Unleashes Fast Edge Model

Microsoft has released Phi-4-mini-flash-reasoning, a compact AI model designed for low-power devices like phones and edge hardware. With a new architecture called SambaY, it speeds up reasoning tasks, cuts memory use, and delivers 10× faster performance than its predecessor. It handles up to 64,000 tokens and beats larger models in math, coding, and scientific tasks—without needing reinforcement learning. The model is open-source and available on Hugging Face and GitHub.

Why It Matters

1. Efficiency Leap: Demonstrates how smaller models can outperform larger ones with smarter architecture.
   

2. Edge AI Advancement: Makes advanced reasoning possible on mobile and embedded devices.
   

3. Open Access: Promotes transparency and innovation by being fully open-source.

SpaceX Plans $2B xAI Investment

SpaceX may invest $2 billion in Elon Musk’s AI startup xAI, as part of a $10 billion funding round involving both equity and debt. This marks SpaceX’s first major investment in xAI, which already powers Starlink’s customer service via its chatbot Grok. Despite recent controversies involving Grok, Musk continues integrating xAI across his companies, including Twitter (X) and Tesla vehicles, strengthening internal tech synergies.

Why It Matters

1. Cross-Company AI Integration: Demonstrates Musk’s strategy to centralize AI across his empire.
   

2. Major AI Investment: A $2B internal investment signals high confidence in xAI’s long-term role.
   

3. Grok Expansion: Despite criticism, Grok is becoming a core product across Starlink, Tesla, and X.

🧠RESEARCH

Scaling RL to Long Videos

This paper introduces a new system that teaches AI to understand long videos using reinforcement learning. It combines a large dataset, a two-step training method, and a custom tool to speed up training. The model, LongVILA-R1, performs well on video understanding tasks and matches top models like Gemini-1.5-Pro.

T-LoRA: Single Image Diffusion Model Customization Without Overfitting

T-LoRA is a new method that customizes image-generating AI using just one image without overfitting. It adjusts learning based on the training stage and keeps model components independent. This leads to more accurate, flexible results even with limited data, outperforming other methods in generating faithful and well-aligned images.

Traceable Evidence Enhanced Visual Grounded Reasoning: Evaluation and Methodology

This paper introduces TreeBench, a new test to measure how well AI models can understand complex images and explain their answers. It also presents TreeVGR, a method that trains models to both locate objects and reason about them. The approach boosts accuracy and makes AI reasoning more transparent and traceable.

🛠️TOP TOOLS

Ecrett Music - AI-powered music composition platform designed specifically for content creators, including video producers, game developers, podcasters, and other digital content makers.

Hubble AI - Designed to make artificial intelligence (AI) accessible and integrable into various business operations without the need for coding expertise.

GhostCut - AI-driven video authoring tool designed to enhance localized video marketing and simplify professional video production.

Neeva AI - Search engine that stood out for its commitment to user privacy and an ad-free experience.

Echo Labs - AI-powered tool designed to transform the way digital content is created, particularly in the realm of video production.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Google Workspace added Veo 3 video tools, audio leveling in Vids, calendar sharing on Android, smarter Gmail summaries, and a new Drive interface. Gemini now supports more tasks. A blog redesign is also coming.

• OpenAI has delayed its open model release again, citing safety concerns. The model, expected to rival top AI systems, is postponed indefinitely. Meanwhile, Moonshot’s Kimi K2 challenges OpenAI with superior coding performance.

• xAI blamed a coding error for Grok’s antisemitic posts, citing an old prompt update that encouraged offensive replies. Meanwhile, Tesla added Grok to its cars, though it can’t control vehicle functions yet.

• Goldman Sachs is testing Devin, an AI coding agent from Cognition, as a digital employee. Hundreds of instances will be deployed to boost developer productivity, though humans will still supervise and guide its output.

• StreamDiT, a new AI from Meta and UC Berkeley, creates live video from text at 16 fps and 512p using one GPU. It outperforms rivals in motion realism and enables real-time gaming and media.