- Homescreen
- Posts
- 🏠 Gemma 3
🏠 Gemma 3
The One-GPU Wonder
GM! It’s Brett again. Remote work is a disaster for young people. Jamie Dimon had a meltdown about it in that leaked video a few months ago.
Early in your career, you need to optimize for learning over everything else - work life balance, city, role, pay, etc. - because later in your career, it’s what you know that will determine your success.
People working remote, especially young people, get put on fewer assignments, know less of what’s going on, and fall behind.
We moved a junior team member to NYC to work in person and it has been game changing for him (and us).

🏠 AI
Google Strikes Back

Google released Gemma 3, and it's a pretty solid deal. These new models are built from the same tech as their Gemini 2.0 but optimized to run on a single GPU or TPU - making them a great for everything from phones to laptops.
Gemma 3 is here! Our new open models are incredibly efficient - the largest 27B model runs on just one H100 GPU. You'd need at least 10x the compute to get similar performance from other models ⬇️
— Sundar Pichai (@sundarpichai)
11:06 AM • Mar 12, 2025
The 27B parameter model outperforms monsters like Llama3-405B while needing just one GPU.
Gemma 3 comes in four sizes (1B, 4B, 12B, and 27B), and the feature set is impressive: 128K token context, support for 140 languages, and multimodal capabilities to process text, images, and short videos.
Developers are already jumping on it. One user built a fully local OCR app that extracts text from images as structured markdown - all running on their own computer via Ollama.
I just built my own OCR app powered by Google's Gemma 3.
It runs 100% locally on my computer using Ollama and extracts text from images as structured markdown.
The code is open-source, and you can find it in the next tweet.
— Akshay 🚀 (@akshay_pachaar)
12:30 PM • Mar 16, 2025
Google also released ShieldGemma 2, a 4B parameter safety filter for images that detects explicit, dangerous, or violent content.
Since launching the original Gemma last year, Google has seen over 100 million downloads and 60,000+ community-built variants. To keep the momentum going, they're offering academic researchers $10,000 in cloud credits.
Gemma 3 is best in class for a VLM that runs on 1 GPU. Should make RL fine tuning feasible. Also Academic researchers can apply for Google Cloud credits (worth $10,000 per award) to accelerate their Gemma 3-based research.
— Kevin Patrick Murphy (@sirbayes)
12:11 PM • Mar 14, 2025
The big picture - This marks a good for AI accessibility. When powerful models can run on standard hardware, suddenly AI isn't just for tech giants with massive compute budgets - it's for everyday developers with regular machines. That's going to accelerate innovation across the board.

Automate Prospecting Local Businesses With Our AI BDR
Struggling to identify local prospects? Our AI BDR Ava taps into a database of 200M+ local Google businesses and does fully autonomous outreach—so you can focus on closing deals, not chasing leads.
Ava operates within the Artisan platform, which consolidates every tool you need for outbound:
300M+ High-Quality B2B Prospects
Automated Lead Enrichment With 10+ Data Sources Included
Full Email Deliverability Management
Personalization Waterfall using LinkedIn, Twitter, Web Scraping & More

🤝 THE LATEST IN…
TECH
Live translation is coming soon to AirPods.
Elon claims X was down due to a cyberattack.
McDonald’s gives its restaurants an AI makeover.
Oracle is reportedly a top choice for helping run TikTok.
AI
OpenAI asks White House for relief from state AI rules.
Google owns a 14% of Anthropic.
OpenAI has signed a $11.9B agreement with the GPU-heavy cloud service provider CoreWeave.
Snapchat is rolling out AI-powered video lenses.
Meta begins testing its first in-house AI training chip.
SCIENCE
Firefly’s Blue Ghost keeps wowing us from the lunar surface.
Engineers turn skin cells directly into neurons for cell therapy.
Robotics and spinal stimulation restore movement in paralysis.
The first OS for quantum networks is here.
CRYPTO
South Korea's central bank rules out bitcoin reserve possibility.
Global crypto funds' worst-ever weekly outflow.
Russia leans on cryptocurrencies for oil trade.

🏃♀️ QUICKIES
Raise: Ditto, a company specializing in resilient connectivity for edge devices, raised $82M in a Series B funding round to enhance data synchronization from the edge to the cloud.
Stat: $3.5B. That’s the price Niantic is selling its video game (Pokémon Go) division to Saudi-owned Scopely as it pivots to geospatial AI.
Rabbit Hole: Climbing the Wrong Hill (Chris Dixon)

🤩 MONDAY MOTIVATION

🛠️ FOUNDERS CORNER
The best resources we recently came across that will help you become a better founder, builder, or investor.
🎥 Screen Studio lets you create stunning recordings & share them instantly via link.
⚙️ Fewsats gives you the payment infrastructure for AI agents.
💯 Dover helps you review job applications 10x faster.

🎯 The 80 / 20
Simple tricks to get a ton of value with minimal effort for work and life.
The OHIO Hack
Only Handle It Once.
If a task takes less than two minutes - just do it. Don’t write it down. Don’t push it to later. Don’t let it sit in your inbox. Knock it out and move on.
Every time you touch a task but don’t finish it, you’re doubling the mental load. Reading an email now and responding later? That’s two context switches instead of one. Sorting mail but not dealing with it? Wasted energy. Every extra step is friction.
OHIO keeps your mental RAM clean. It works great for emails, DMs, small admin tasks - anything quick. If it’s a bigger task, at least decide when you’ll handle it instead of leaving it in limbo.
The less you revisit small things, the more brainpower you have for actual work.

❓️ AI GENERATED OR NOT

Is this AI generated? |
Last week’s answer: If you guessed AI-generated… you were wrong.
🟨🟨⬜️⬜️⬜️⬜️ Yeah (~29%)
🟩🟩🟩🟩⬜️⬜️ Nope (~71%)
Here’s the original.

👋 A Quick Note About Micro
Micro is a CRM that doesn't feel like homework and an email client that knows what matters. Simply put, it's an all-in-one productivity tool that organizes itself so you don't have to.
I built Micro with some friends because I was tired of getting guilt tripped for not hitting inbox 0 and having to bounce around between so many tools to do my job.
We're launching in 2025, but if you want early access, sign up for the waitlist at micro.so and respond to the confirmation email saying you came from Homescreen.

HOW WAS TODAY'S NEWSLETTER? |

REACH 40K+ FOUNDERS, INVESTORS & OPERATORS
If you’re interested in advertising with us, send an email over to [email protected] with the subject “Homescreen Ads”
P.S. If you're thinking about starting your own newsletter, we highly recommend beehiiv.