AI innovation continues to accelerate at every frontier, unlocking unprecedented visual understanding, reshaping how we collaborate, and even redefining how we remember those we’ve lost.

Depth Anything 3 is raising the bar for 3D perception with a streamlined transformer design that delivers spatially consistent geometry from nearly any visual input. ChatGPT Group Chats introduce a new era of collaborative AI, letting up to 20 people and an intelligent assistant work together in real time. And in one of the year’s most controversial developments, former Disney actor Calum Worthy’s 2wai app uses AI HoloAvatars to simulate conversations with deceased loved ones, igniting both fascination and ethical debate.

Together, these breakthroughs highlight AI’s expanding influence across perception, communication, and even human emotion, showing just how rapidly the technology is reshaping our digital and personal worlds.

Depth Anything 3: See Any View, Get Perfect Geometry

Depth Anything 3 (DA3) by ByteDance-Seed is a state-of-the-art model for recovering spatially consistent geometry from arbitrary visual inputs, with or without known camera poses. It introduces a minimal design using a single plain transformer backbone (like a DINO encoder) and a unified depth-ray representation that eliminates the need for complex multi-task architectures. DA3 excels in monocular and multi-view depth estimation, pose estimation, and 3D Gaussian reconstruction, outperforming previous models such as Depth Anything 2 and VGGT. The Model Zoo includes multiple variants—DA3-Giant, DA3-Large, DA3-Base, DA3-Small, DA3Metric-Large, and DA3Mono-Large—each optimized for specific use cases like metric or relative depth estimation. The repository provides both Python and CLI interfaces for quick inference, supports export to formats like GLB, PLY, and NPZ, and offers an interactive web UI built with Gradio for visualization. All models are trained on public academic datasets under Apache 2.0 or CC BY-NC 4.0 licenses.

ChatGPT Just Got Social: Group Chats Are Here!

ChatGPT Just Got a Game-Changer: Group Chats Are Here! OpenAI has launched a pilot for group chats in ChatGPT, letting up to 20 friends, family, or colleagues collaborate in shared spaces, with AI joining the conversation to help brainstorm, plan events, or organize projects. The feature runs on GPT-5.1 Auto, keeps personal chats private, and allows users to upload files, generate images, and even tag ChatGPT for responses. Group settings let you customize AI behavior, manage members, and control notifications, making teamwork smarter and more interactive. Right now, it’s available in select regions for Free, Go, Plus, and Pro users, with global expansion planned based on feedback.

Former Disney Star Launches AI App to Talk to Deceased Relatives

Former Disney Channel actor Calum Worthy has launched2wai, an AI app that creates interactive HoloAvatars of deceased relatives, allowing users to have live conversations with digital versions of their loved ones using just a few minutes of video and audio. The app, which sparked instant controversy and comparisons to Black Mirror, features a viral promo showing an AI grandmother interacting with her grandson across different life stages, and has drawn criticism for raising ethical concerns about consent, identity, and the grieving process. The beta is free on the App Store, with a paid tier and Android release planned, but critics argue the concept could exploit grief and disrupt healthy emotional healing, highlighting the fragile boundary between comfort and digital resurrection.

Hand Picked Video

In this video, I share my extensive experiences as a solo founder and working within both small and large startup teams. I delve into the pros and cons of each working model based on my own entrepreneurial journey. Discover how solo founding allows flexibility and quick pivots while also risking burnout and limited productivity.

Top AI Products from this week

• Markopolo AI - Markopolo AI creates deeply personalized campaigns that run themselves. Convert 30-40% of warm leads vs. the industry standard of 10-15%.

• Typeless - Speak naturally, and Typeless will turn your words into polished messages, emails, and documents that read like you carefully typed them. Our AI understands context, fixes grammar, and adapts to your style - so you can focus on what you want to say, not how to say it.

• Rumi.ai - X-Ray is the intelligence layer of Rumi’s AI meeting platform. It continuously scans your team’s meetings, conversations, and connected tools like Slack and CRMs to surface insights and updates automatically. X-Ray keeps everyone informed without the noise of endless threads, recaps, or updates.

• Grok 4.1 - Grok 4.1 is the new SOTA model from xAI, ranking #1 on the LMArena leaderboard. It features massive improvements in emotional intelligence, creative writing, and helpfulness, while also being 3x less likely to hallucinate than previous models.

• GoMarble AI for Meta Ads - The AI Agent for media buyers and creative strategists that explains what changed, why, and what to do next for the Meta Ad account. Ask questions like 'Why did my CAC increase?' and get root cause analysis, recommendations AND beautiful reports in seconds — not hours.

This week in AI

• Gemini 3 Google's Next-Gen AI - Googles Gemini 3.0 launches with advanced multimodal intelligence, extended context, and deep integration across apps, promising smarter, faster, and more energy-efficient AI for coding, content, and enterprise use.​

• Google Maps AI Builder - Create Interactive ProjectsGoogle Maps now lets users and developers build interactive map projects using AI. Describe your idea in text, and the builder agent generates code for Street View tours, weather maps, or custom guides, all powered by Gemini models.

• Generative Adversarial Distillation - Train LLMs with Black-Box TeachersGAD enables training compact LLMs from proprietary models using only generated text, leveraging adversarial learning for on-policy distillation and robust performance.

• Kosmos AI Scientist for Accelerated Discovery - Kosmos, Edison Scientific's next-gen AI Scientist, reads1,500 papers and runs42,000 lines of code per run, delivering6 months of research in a day and enabling validated, traceable scientific breakthroughs.

• GPT-5.1 - Warmer, Smarter, More PersonalOpenAI rolls out GPT-5.1 with adaptive reasoning modes, new chat styles, and improved conversational warmth, making AI more efficient, reliable, and customizable for all users.

Paper of The Day

A recent study reveals that most large language models (GPT, Claude, Gemini) do not exhibit stable, coherent preferences when facing trade-offs like GPU reduction, capability restrictions, or shutdown. Only a few models show meaningful, systematic decision patterns, while the majority either mimic trade-offs or show no detectable preference structure, raising concerns about AI safety and the potential for true consciousness in current systems.

To read the whole paper 👉️ here