Gemma 3 is here: The most powerful AI model that runs on your laptop

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

👉 Google released its most capable model, Gemma 3, that runs on your laptop

👉 Gemini 2.0 Flash just got a very powerful image generator

👉 Snapchat launched its first video generative AI Lenses

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 5 minutes

What’s going on: Google announced the release of Gemma 3, the latest iteration of its open model family, designed to be the most advanced and portable. Building on the technology that powers the Gemini 2.0 models, Gemma 3 introduces a collection of lightweight, state-of-the-art models that excel in performance while being optimized to run efficiently on a range of devices, from smartphones and laptops to workstations. This helps you to create engaging user experiences that can fit on a single GPU or TPU host.

What does it mean: This release follows the remarkable success of the Gemma family, which has seen over 100 million downloads and spawned more than 60,000 variants within its vibrant "Gemmaverse" community since its debut. With this new version, Google aims to help developers create AI applications that are accessible wherever they are needed, emphasizing portability and speed.

More details: 

• Gemma 3 comes in multiple sizes: 1B, 4B, 12B, and 27B parameters, offering flexibility for various use cases and hardware capabilities.

• The model has enhanced features such as multimodality and a 128k-token context window for handling complex tasks. Gemma 3 supports over 140 languages, thanks to a new tokenizer, and delivers improved performance in math, coding, and instruction-following.

• Gemma 3 was trained with advanced techniques such as distillation, reinforcement learning from human and machine feedback, and model merging.

• Additionally, quantized versions of Gemma 3 are available, reducing computational demands while maintaining accuracy, making it the world’s top-performing single-accelerator model, surpassing competitors like Llama-405B and DeepSeek-V3 in benchmarks like LMArena.

• For learning more about Gemma 3’s capabilities and benchmarks, read Google’s official blog post.

• Looking for a more developer-friendly guide on Gemma 3? Read this.

What’s going on: Google has added a new experimental native image generation feature to the Gemini 2.0 Flash that is now available for developer experimentation across all regions currently supported by Google AI Studio. You can test this new feature by choosing the (gemini-2.0-flash-exp) model in Google AI Studio and via the Gemini API.

What does it mean: Gemini 2.0 Flash combines multimodal input, enhanced reasoning, and natural language understanding to create images. You can use this model to generate visuals for a story keeping the characters and settings consistent throughout the whole process. Give it feedback and the model will retell the story or change the style of its drawings. You can also upload your own image or use an AI-generated image by the Gemini and add or subtract items/props/details.

More details: 

• Gemini 2.0 uses its extensive knowledge base (world understanding) and advanced reasoning capabilities to produce appropriate and contextually relevant images. The model excels in generating lifelike and intricate visuals.

• Many image generation models face challenges when it comes to accurately depicting lengthy text sequences, which can lead to issues such as illegible characters, poor formatting, or spelling errors. However, internal assessments indicate that Gemini 2.0 Flash outperforms leading competitors in rendering quality. This makes it particularly effective for producing advertisements, social media posts, and invitations.

• We tested this feature and it blew or mind! If you are also interested in giving it a shot, login to your Google AI Studio account, choose the “gemini-2.0-flash-exp” from the models dropdown, and start experimenting for free!

🕵️‍♂️ Dario Amodei, CEO of Anthropic, is concerned that spies, probably from China, are accessing valuable "algorithmic secrets" from leading US AI companies and is calling for intervention from the US government.

📄 Sakana, a Japanese startup, claims its AI Scientist-v2 system, generated a paper that passed peer review at an ICLR workshop; however, the achievement is nuanced, as the paper was withdrawn before publication, contained citation errors and underwent less scrutiny than typical peer-reviewed publications.

👻 Snapchat has launched its first video generative AI Lenses which is powered by its in-house generative video model, for Snapchat Platinum subscribers ($15.99/month) starting with three Lenses and planning to add more weekly.

🔒 Pentera, an automated security validation platform that simulates network attacks to train security teams, has raised $60 million at a $1 billion valuation.

🎥 Moonvalley has launched Marey, an AI video-generating model trained on openly licensed data, distinguishing itself from competitors by aiming for lower legal risk and offering features like fine-grained camera and motion controls.

⚖ Meta is facing a copyright lawsuit in France, filed by the National Publishing Union (SNE), the National Union of Authors and Composers (SNAC), and the Society of People of Letters (SGDL), accusing the company of unlawfully training its AI models on protected content.

☁ Microsoft has introduced OpenAI’s two innovations in Azure AI Foundry: the Responses API, which integrates AI tools like function calling and file search into a single API for seamless automation, and the Computer-Using Agent (CUA), an AI model capable of navigating GUIs, automating tasks across applications, and adapting dynamically to UI changes.

AI + Mobile app design

OpenAI - Software Engineer, Backend

CSC - AI/ML Software Engineer

Nuro - Sr. Data Scientist

Ramp - Staff Machine Learning Engineer

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!