Google just rolled out generative AI capabilities for Google Maps Platform and Google Earth, allowing developers to ground LLMs in real-world data from 250 million places. Google Earth is also getting Gemini-powered multi-step reasoning to slash urban planning analysis time from days to minutes. Next month, Rivian will begin rolling out AI place summaries to its vehicle screens.
Google just added Grounding with Google Search to Google AI Studio and the Gemini API, letting Gemini 1.5 fetch live web answers with inline citations. Testing is free inside AI Studio, while API access runs $35 per 1,000 grounded queries. Why it matters: developers can finally plug Google's web index directly into Gemini without messy custom retrieval pipelines.
OpenAI just rolled out its Advanced Voice Mode to ChatGPT desktop apps on macOS and Windows, bringing human-like, low-latency audio conversations directly to your computer. Users can chat back and forth while coding or brainstorming over uploaded files with five new voices. It is currently limited to paid Plus and Team subscribers, excluding the EU and UK.
Google CEO Sundar Pichai just revealed that more than a quarter of all new code at the company is generated by AI, which engineers then review and accept. The milestone highlights how deeply automated development has become inside the tech giant itself. Why it matters: Big tech isn't just selling AI coding tools to developers—they're letting AI build their own stack.
GitHub Copilot just added Anthropic's upgraded Claude 3.5 Sonnet as a model option directly within VS Code and GitHub.com. Powered via Amazon Bedrock, the top-tier coding model is rolling out in public preview to all Copilot Chat users and organizations over the coming weeks.
OpenAI just launched Canvas, a dedicated side-by-side workspace for writing and coding projects. Instead of getting stuck in long chat threads, users can highlight specific sections for inline feedback, restore previous versions, and use quick shortcuts to debug code or adjust reading levels. It marks the first major visual interface overhaul for ChatGPT.
Google just graduated Gemini 1.5 Flash-8B from experimental status to general availability. The ultra-lightweight model nearly matches the standard 1.5 Flash on key benchmarks while offering Google's lowest price per token yet and doubled rate limits of up to 4,000 requests per minute.
OpenAI just launched an integrated model distillation workflow directly in its platform, combining output storage, evals, and fine-tuning into one place. Developers can now capture completions from advanced frontier models and use them to train cost-efficient smaller models without juggling messy third-party pipelines.
Google just shipped a streamlined mobile experience for Google AI Studio, stripping away clutter and putting models front and center. The redesign makes prompting and testing models significantly easier and faster while on the go.
Microsoft just launched general availability for fine-tuning GPT-4o and GPT-4o mini on Azure OpenAI Service. The update also brings Meta's new Llama 3.2 vision models, the Phi-3.5 family, and a new Correction feature to catch hallucinations in real time before users see them. It makes tailor-made enterprise LLMs much easier and safer to deploy.
GitHub just opened a preview bringing OpenAI’s o1-preview and o1-mini to GitHub Copilot and Models. Developers can now swap out GPT-4o inside VS Code’s Copilot Chat to tackle complex logic bugs and edge cases with deliberate reasoning. Why it matters: Coding assistants are moving beyond basic boilerplate autocomplete toward actual architectural problem-solving.
Alibaba Cloud just released over 100 open-source Qwen 2.5 models, with sizes ranging from 0.5 to 72 billion parameters and sharper math and coding skills. The company also unveiled a new text-to-video AI model under its Tongyi Wanxiang family, alongside an upgraded Qwen2-VL vision model. It marks one of the biggest open-source AI drops from a major tech giant to date.
Anthropic just introduced Contextual Retrieval, a simple fix for broken RAG systems that lose context after chopping documents into chunks. By prepending chunk-specific context using Claude 3 Haiku, it slashes retrieval failures by 49%, and by 67% when paired with reranking. Prompt caching keeps the preprocessing cost down to $1.02 per million tokens.
Perplexity just rolled out a new reasoning focus mode in beta for Pro subscribers, tapping into OpenAI's o1-mini. It is built specifically for math, puzzles, and coding tasks, though it lacks web search integration for now and runs with strict rate limits.
Magic just announced LTM-2-mini, an ultra-long context model capable of processing 100M tokens—equivalent to roughly 10 million lines of code or 750 novels. The startup claims its sequence-dimension algorithm is 1,000x cheaper per decoded token than Llama 3.1 405B at that context scale, and revealed $320M in new funding alongside a Google Cloud partnership to build supercomputers.
Anthropic just made Artifacts generally available to all Claude.ai users, rolling out the interactive workspace to Free, Pro, and Team tiers across web and mobile apps. The feature lets you instantly render, iterate, and remix code, SVG graphics, flowcharts, and live prototypes in a dedicated side window without leaving the chat.
Google just rolled out three experimental models: an ultra-compact Gemini 1.5 Flash-8B, a significantly upgraded Flash, and a stronger Pro model that levels up coding and complex prompt reasoning. Both maintain 1M token context windows and are available to test for free in Google AI Studio and the Gemini API.
OpenAI is reportedly developing a reasoning-focused AI codenamed Strawberry, which tackles unseen math and coding problems and has already been shown to U.S. national security officials. While a distilled version could arrive in ChatGPT as early as this fall, OpenAI is actively using Strawberry to generate synthetic training data for Orion, its true GPT-4 successor.
NVIDIA just released NVIDIA Blueprints to help businesses build and deploy custom generative AI workflows. Each blueprint packages reference code, documentation, and Helm charts alongside NIM microservices, starting with customer service digital humans, drug discovery, and multimodal PDF extraction. It basically lets enterprise developers skip the plumbing and turn internal data into self-improving AI workflows right away.
xAI just overhauled its entire inference stack from scratch in three days using the open-source SGLang framework. Grok-2 mini is now running twice as fast with slightly improved accuracy, while the full Grok-2 model gained much smoother multi-host inference speeds.
