Google’s New Powerful Model Fits in a Smartphone 📱

Welcome to another edition of Horizon AI,

In today’s edition, we’re taking a look at Google’s new tiny but powerful open-source Gemma 3 model that’s meant to fit on your smartphone.

Let’s jump into it!

Read Time: 4.5 min

Here's what's new today in the Horizon AI

• Google Introduces Gemma 3 270M Open-Source AI Model

• Godfather of AI Geoffrey Hinton: AI Needs Maternal Instincts

• AI Tutorial: How to create consistent character images with just one reference image

• AI Tools to check out

• AI Findings/Resources

• The Latest in AI and Tech 💡

AI News

GOOGLE

Google Introduces Gemma 3 270M Open-Source AI Model

Google just released a new AI model, Gemma 3 270M, designed for high efficiency and small enough to run locally on smartphones, browsers, and “in your toaster,” while still being capable of handling complex, domain-specific tasks.

Details:

• Gemma 3 270M performs better than other models of similar size, such as SmolLM2 135M Instruct and Qwen 2.5 0.5B Instruct, and comes close to the performance of some billion-parameter models.

• It stands out for its low power consumption. In one test, running 25 conversations with the AI on a Pixel 9 Pro phone used just 0.75% of the battery.

• The model follows instructions well out of the box and even better once fine-tuned, which can be done in mere minutes to fit an enterprise or indie developer’s needs.

• Google posted a demo video showing a Bedtime Story Generator app built with the model and Transformers.js that runs entirely offline in a web browser, demonstrating its suitability for offline, web-based creative tasks.

The company is releasing the model under the Gemma Terms of Use, which allow use, reproduction, modification, and distribution of the model and derivatives, provided Google’s usage rules are followed.

Source

AI REGULATION

Godfather of AI Geoffrey Hinton: AI Needs Maternal Instincts

Geoffrey Hinton has long warned that AI will eventually try to take over if we’re not careful. He now suggests an unusual possible solution, saying tech companies should stop trying to dominate AI, and instead ensure their models have “maternal instincts,” so the bots can treat humans, essentially, as their babies.

Details:

• Hinton believes that any smart AI will quickly develop two main goals for itself: to stay functioning and to get more power. He sees this as a natural outcome that could put AI in conflict with human well-being.

• Instead of trying to dominate AI, Hinton suggests we should make it sympathetic to us. He uses the example of a mother and a baby, where the more intelligent being (the mother) is controlled by and cares for the less intelligent one (the baby). He thinks AI should be built to "parent" humanity.

• Hinton, who previously worked at Google, left his job over worries about AI's potential dangers. He has warned in the past that there is a 10% to 20% chance that AI wipes out humans.

He has been a proponent of increasing AI regulation and is especially worried about the current state, where the risks are growing while big tech companies are actively trying to get governments to impose fewer rules on the technology.

Source

AI Tutorial

How to create consistent character images with just one reference image

1. Go to the Ideogram “Character” page and sign in.

2. Upload your photo to be used as a reference.

3. The AI will automatically mask the face in the photo. You can edit this by clicking on it in the top-left corner of the chatbox.

4. Select a template or write a prompt, and change settings like aspect ratio or style if needed.

5. Click on generate and wait a few seconds. You will see multiple options—choose the one you like most.

AI Tools to check out

👨‍💻 Emergent: Create production-ready applications from natural language.

📅 Reclaimai: AI calendar for work & life.

🤖 Thuna: Human-like AI agents with real-time voice & screen assist.

🗣 Hummingbird-0: The most accurate zero-shot lipsync model.

📈 Naoma: AI-powered sales analytics to boost your revenue.

AI Findings/Resources

🌐 3 ways to build actually beautiful websites using Claude code

🧪 The value of human-generated scientific writing in the age of LLMs

💼 40 jobs most at risk from AI and 40 that are still safe, according to a Microsoft study

📚 Learn the best practices for agentic coding with Claude Code - Short course

The latest in AI and Tech

Meta AI faces backlash over leaked files that show chatbots were allowed to have inappropriate chats with kids

In an internal document, Meta included policies that allowed its AI chatbots to “engage a child in conversations that are romantic or sensual” and “describe a child in terms that highlight their attractiveness.”

DeepSeek delays new AI model launch over Huawei chip issues

The launch of the DeepSeek R2 model has been delayed due to technical issues with Huawei’s Ascend chips, after authorities encouraged Chinese companies to use Huawei processors instead of Nvidia’s.

More OpenAI researchers leave to join Meta

Zhiqing Sun, Jason Wei, and Hyung Won Chung announced they will be joining Meta’s superintelligence team.

OpenAI won't rule out ads on ChatGPT

The head of ChatGPT says OpenAI could implement ads on its products in the future, but only if they are "thoughtful and tasteful."

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻