This Week in AI | Week 11
Neha
Last updated on October 19, 2026
Microsoft’s Phi-3 Mini is a language model small enough to run on a phone, and Microsoft says it performs close to much larger models. This week also brought the first Abu Dhabi Autonomous Racing League event, where driverless cars showed how far the technology still has to go, and new Firefly tools in Photoshop that let designers steer results with images instead of long text prompts.
Together, the three stories point the same way: AI is moving out of the lab and into tools people use every day. Below, we cover what Phi-3 Mini means for developers, where the A2RL cars struggled, and which Firefly features are worth trying first.
Microsoft launched the Phi-3, a 3.8-billion-parameter language model designed to run on laptops and phones instead of in a data center. Most capable models need expensive cloud hardware. Phi-3 Mini is built for cases where you want to avoid that cost, keep data on the device, or work offline.
Key Features of Phi-3 Mini:
Limitations: A small model stores less factual knowledge, so Microsoft notes weaker results on trivia-style tasks and recommends pairing it with search or retrieval. It also performs best in English.
Who should care: developers building mobile or edge apps, startups that want low inference costs, and teams whose privacy rules rule out cloud APIs. Phi-3 Mini is available on Hugging Face, Azure AI, and Ollama.
The Technical University of Munich (TUM) won the first Abu Dhabi Autonomous Racing League (A2RL) race at Yas Marina Circuit on 27 April 2024. Four cars qualified for the final, which was meant to be the first wheel-to-wheel race with no human intervention. The result was less polished than the headline suggests, and that is what makes the event worth a closer look
Key Takeaways from A2RL:
A2RL’s debut signifies the ongoing evolution of autonomous vehicle technology, with each race serving as a milestone in the journey towards enhanced performance and reliability.
Adobe introduces Firefly, a new generation of generative AI tools integrated into Photoshop, aimed at streamlining creative workflows and empowering users with enhanced control over design elements.
Key Features of Firefly:
Adobe’s Firefly model is an advancement in AI-assisted designing tools, allowing users to easily and accurately express their ideas through intuitive features.
For creators looking to explore new concepts beyond image editing, the Adobe Firefly AI design generator can help transform ideas into engaging visual content.
AI continues to be integrated into various fields, bringing potential improvements in efficiency, creativity, and performance to multiple industries.
Yes, in most cases. With 4-bit quantization the model needs roughly 1.8 GB of memory, so a recent laptop or phone can run it offline. Tools such as Ollama make local setup simple on a computer. Phone performance depends on your device, and Microsoft’s reported speed came from an iPhone 14.
Not for everything. It works well for summarizing, drafting, and simple Q&A, and Microsoft’s benchmarks put it near much larger models. Because it is small, it stores less factual knowledge, so it is weaker on trivia-style questions. For fact-heavy work, pair it with search or retrieval, and test it on your own tasks before relying on it.
They launched in the Photoshop desktop beta, and Adobe said general availability would follow “later this year.” If you have a Creative Cloud subscription, you can install the beta from the Creative Cloud app. Check Adobe’s current release notes for the latest status.
Team TUM from the Technical University of Munich won the first A2RL race in Abu Dhabi, with a last-lap overtake. In a demonstration run against former F1 driver Daniil Kvyat, the AI car finished 10.38 seconds behind him. That was not a flat-out race, so treat the gap as a progress marker, not proof of parity.
Better inputs and real-world testing drove this week’s progress, not bigger models. Phi-3 got its results from cleaner training data, Firefly lets designers steer with image references instead of longer prompts, and A2RL showed in public, with a spin and a red flag, what lab tests miss.
To act on this week’s news, try Phi-3 Mini on a task of your own and compare it with Microsoft’s benchmark claims, or test the Firefly tools on a real image in the Photoshop beta. Check back next week for Week 12.
YourGPT has progressed from production deployments reaching around 80% AI resolution to 90%+ on eligible repeated requests in maintained deployments. Here is what changed, how we measure resolution, and how we apply that work with organizations that want to raise their own rates.
TL;DR Claude Fable 5 launched on June 9, 2026, was pulled offline three days later under a US export control order, and returned on July 1 after the order was lifted. For support teams, Fable 5 is best suited for long-horizon tickets across billing, CRM, shipping, and documents not basic FAQ deflection. Its biggest support […]
At YourGPT, we’ve always believed businesses need one platform for customer support, sales, and engagement. These are not separate parts of the customer journey. They shape how businesses communicate, build relationships, and move conversations forward. That belief continues to guide how we evolve YourGPT, building toward a more connected way for teams to manage communication, […]
TL;DR YourGPT Copilot SDK is an open-source SDK for building AI agents that understand your application state, user context, and what is happening inside your product. Instead of working like isolated chat widgets, these agents connect directly with your product and can take real actions within existing workflows. This helps teams build contextual AI experiences […]
Happy New Year! We hope 2026 brings you closer to everything you’re working toward. Throughout 2025, you’ve seen the platform evolve. We shipped the AI Copilot Builder so your AI could execute actions on both frontend and backend, not just answer questions. We added AI assistance inside Studio to help you generate workflows without starting […]
Grok 4 is xAI’s most advanced large language model, representing a step change from Grok 3. With a 130K+ context window, built-in coding support, and multimodal capabilities, Grok 4 is designed for users who demand both reasoning and performance. If you’re wondering what Grok 4 offers, how it differs from previous versions, and how you […]