Welcome, humans.

Well, turns out that Tesla cameras vs LiDar story from yesterday was a biiiit biased. In the video, Mark Rober apparently didn’t even use Tesla’s newest “Full Self Driving” (FSD) tool for the comparison, so it wasn’t really a fair fight.

That said, some people were saying the video was sponsored by Luminar, but Mark Rober says it wasn’t—they just donated the car.

We still think the coolest part of the self-driving story is using genAI to simulate actual driving scenarios for training, as every self-driving car—be it Tesla, or anyone else—can benefit from that for finding edge cases.

We’ll let the real-world data of Tesla’s safety record (including Autopilot and FSD data, showing FSD has a much lower fatality rate—which is good!) vs Waymo’s speak for itself.

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

• New research found AI can actually make good memes.

• Roblox launched an open-source 3D object generation model.

• xAI bought a startup to develop Grok's video generator.

• Adobe says 55% of US consumers have used AI for research.

Turns out, AI models might be better at making memes than you are…

New research just dropped that might have your favorite social media meme account sweatin’ like the Daily Struggle guy.

In their paper “One Does Not Simply Meme Alone” (10/10 title game), European researchers tested memes about work, food, and sports created by three groups: humans working alone, humans collaborating with AI, and AI working solo.

The results? Pretty surprising:

• Memes generated entirely by AI scored highest across all three metrics (humor, creativity, shareability).

• Human-AI collaboration didn't improve quality over humans working alone.

• BUT humans still created the funniest top-performing memes.

While AI memes did better on average at creating broadly appealing, middle-of-the-road content, the absolute best memes still came from humans (for humor) and human-AI teams (for creativity and shareability).

The researchers suggest this happens because language models are trained on massive datasets showing what broad audiences find funny—essentially optimized for “crowd-pleasing” humor. Corporate memes about to be AI-ified.

How they did it: The researchers prompted GPT-4o to “generate 20 meme captions for this [image] about the topic of [topic]” and designed their own UI where the AI helped generate caption ideas.

What's also fascinating is that people who used AI assistance:

• Generated significantly more ideas.

• Reported less effort to accomplish their tasks.

• Still felt ownership over their creations (though less than those working alone).

Our take: The future likely belongs to those who can effectively partner with AI tools rather than compete against them. This research mirrors what we're seeing across other creative fields—AI's sweet spot is rapidly producing solid, broadly appealing content. Put another way: AI content is mid. 

This matters because memes help us cope. They're not just funny content, but actually activate our brain's empathy circuits and create shared experiences during stressful times—like, IDK, AI taking over the world??

The real magic of memes isn't their mass appeal—it's that moment when you think, “omg, THIS! 1000%” because someone else just gets it. Good luck automating THAT at scale!

FROM OUR PARTNERS

Google Cloud’s Future of AI: Perspectives for Startups

Want to build a real business with AI, but don’t know where to start?

Check this out: This new report from Google Cloud reveals how to transition AI projects from proof-of-concept to production, capitalize on underhyped opportunities, and create immediate value. 

Dive into 23 unique perspectives from top AI industry leaders, including… 

• Amin Vahdat (Google Cloud). 

• David Friedberg (Ohalo Genetics).

• Chamath Palihapitiya (Social Capital). 

• Crystal Huang (GV).

• Dylan Fox (Assembly AI). 

…and many more. 

Learn what areas investors are prioritizing for AI startups in 2025 and how to leverage AI to outpace competitors and establish a unique market position.

Download now

Prompt Tip of the Day

Whether you’ve used Claude before or not, today we recommend you try it out and ask it to help you visualize a concept you want to learn (example).

Want to take it a step further? This video from Grace Leung shows how you can combine Perplexity's research capabilities with Claude's visualization features to create powerful interactive dashboards and visual representations of complex info.

Treats To Try.

1. *Flow for Windows is finally here! Speak 3x faster than typing with AI-powered dictation that works everywhere—also available on MacOS. Try it today. 

2. Mistral Small 3.1 is a new small AI that’s quite good, can understand images and supports 128K tokens (equivalent to ~96K words or ~200+ pages)—try here or download this base version or instruct version.

3. Klangio converts your music recordings into editable sheet music as well as digital music formats like TABs, MIDI, and MusicXML files.

4. Freepik now lets you use Gemini’s image changer (which is going viral for its controversial ability to remove image watermarks) in its AI suite—you can also use it here or in Google AI Studio.

5. Currents scans Reddit (maybe others? seems mainly Reddit-based) to show you exactly what people are saying about your market, competitors, and trends.

6. Someone used Perplexity to analyze their own blood report—there’s also OpenEvidence, which organizes the latest medical research to help you make evidence-based decisions without drowning in information overload.

7. Cal AI calculates your food's calories and nutrients from just a photo—read more.

See our top 51 AI Tools for Business here!

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

Around the Horn.

• Roblox released and open-sourced Cube 3D, a new foundation model for generating 3D objects that is trained on publicly licensed data (along with Roblox data)—and will be available as part of Roblox Studio.

• Zoom announced it will update its AI Companion with new features that will roll out gradually from March through summer 2025, with immediate releases including meeting agendas and task management—read more.

• xAI acquired generative video company Hotshot, which will now likely be used to create Grok’s video generator (check out their models here).

• A top OpenAI exec in charge of post-training research left the company to create a new startup that uses AI for material sciences—and OpenAI will invest in it.

• According to a new survey from Adobe of 5K US consumers, 55% now use AI for research, while 39% use AI for shopping and 53% plan to do so soon.

FROM OUR PARTNERS

Sales teams using Attention are crushing their targets

Attention's platform is transforming how companies sell. HorizonIQ boosted post-call summaries from 50% to 99%, saving 25 minutes per call and doubling productivity. Unify increased win rates by 40% and reduced sales ramp time by 3x.

We’re big Attention fans at The Neuron and can confirm: if you sell literally anything, you need this tool. Book a free demo.

Under the Hood

• Here’s a great ~20 minute explainer on Anthropic’s MCP, which creates a unified communication layer between language models and external tools.

• L2D is a massive self-driving dataset that gives you 5K hours of driving footage from 6 HD cameras to train your own Tesla-like self-driving system (90TB)—use this to explore the diversity of scenarios.

• This blog explains how Cursor and other AI coding tools work—by combining a VS Code fork with language models that use tool-calling to read and edit files, optimized with prompts that help the models overcome their limitations.

• Here are the top 10 AI papers from last week, according to DAIR.AI: 

  1. Gemma 3.

  2. Traveling Waves in Visual Cortext.

  3. Transformers without normalization.

  4. Chain of thought monitoring.

  5. Improving planning of agents.

  6. Gemini Robotics.

  7. Search-R1.

  8. Auditing for hidden objectives.

  9. Survey of post-training.

  10. Block Diffusion.

  11. BONUS: Survey of Thinking LLMs.

A Cat's Commentary.