The Daily AI Briefing - 13/03/2025

Welcome to The Daily AI Briefing, here are today's headlines! Today we're covering Google's new Gemma 3 model family that promises high performance on single GPUs, Gemini Flash's expanded image capabilities, a tutorial for building your own Telegram AI assistant, Jotform's no-code AI agents for customer service, Sakana's achievement with an AI-authored scientific paper, and a roundup of trending AI tools transforming various industries. Starting with Google's Gemma 3 announcement, Google has unveiled a new family of lightweight AI models built from the same technology as Gemini 2.0. These models deliver performance rivaling much larger counterparts while running efficiently on just a single GPU or TPU. The family comes in four sizes - 1B, 4B, 12B, and 27B parameters - optimized for different hardware configurations from phones to laptops. Notably, the 27B model outperforms larger competitors like Llama-405B on the LMArena leaderboard. Gemma 3 boasts impressive capabilities including a 128K token context window, support for 140 languages, and multimodal abilities to analyze images, text, and short videos. Google also released ShieldGemma 2, a 4B parameter image safety checker that can filter explicit content with easy integration into visual applications. In related news, Google has expanded Gemini Flash with new experimental image-generation capabilities. Users can now upload, create, and edit images directly within the language model without requiring a separate image-generation system. Available via API and in Google AI Studio, the 2.0-flash-exp model supports both image and text outputs with editing through natural conversation. What makes this particularly impressive is Gemini's ability to maintain character consistency and understand real-world concepts throughout interactions. For example, you can prompt it to generate a story with pictures and then refine it through dialogue. Google claims Flash 2.0 excels at text rendering compared to competitors, making it ideal for ads, social posts, and other text-heavy design generations. For the DIY enthusiasts, there's a new tutorial on building your own AI-powered Telegram assistant. This guide walks you through creating a personal AI helper that can answer questions, remember conversations, and eventually connect to other services using n8n's automation platform. The process involves creating a Telegram bot via BotFather, setting up an n8n workflow with a Telegram trigger, adding an AI Agent node connected to your preferred AI model, and configuring a response mechanism. By enabling Window Buffer Memory in the AI Agent settings, your bot will remember previous conversations, creating a more natural interaction experience. Moving to business applications, Jotform AI Agents are now offering organizations the ability to provide 24/7 conversational customer service across multiple platforms without coding requirements. The system includes over 7,000 ready-to-use AI agent templates, automation capabilities for workflows and custom actions, seamless handling of voice, text, and chat inquiries, and customization options to align with brand identity. This solution aims to help businesses scale their customer interactions efficiently while maintaining personalized service. In scientific news, Japanese AI startup Sakana has achieved what they claim is a milestone: their AI system successfully generated a scientific paper that passed peer review. Their AI Scientist-v2 created three papers, handling everything from hypotheses and experimental code to data analyses and visualizations without human modification. One paper was accepted at the ICLR 2025 workshop with an average reviewer score of 6.33, ranking higher than many human-written submissions. Sakana acknowledged some limitations, including citation errors and the fact that workshop acceptance rates are higher than typical conference tracks, but they view this as a promising sign of progress. Before we end, some trending AI too