Introducing the first AI-native CRM

Connect your email, and you’ll instantly get a CRM with enriched customer insights and a platform that grows with your business.

With AI at the core, Attio lets you:

• Prospect and route leads with research agents

• Get real-time insights during customer calls

• Build powerful automations for your complex workflows

Join industry leaders like Granola, Taskrabbit, Flatfile and more.

👉 Try Attio Pro for free

Today:

• Moonshot’s Kimi K2.5 Challenges Top Models with "Agent Swarm" and High-Level EQ

• Google DeepMind’s "Genie" Lets You Turn Text Prompts into Playable Games

• xAI Challenges Sora with New "Grok Imagine" Video Generation API

• Nvidia, Microsoft, and Amazon in Talks to Pump $60 Billion into OpenAI

• Apple Acquires Israeli Startup Q.AI for $2 Billion to Supercharge Siri

• Musk Plotting Massive Merger of SpaceX and xAI Ahead of Planned IPO

KIMI K2.5 AGENT SWARM is INSANE

Kimi K2.5 is the newest open-source AI model from Moonshot, offering standout performance in coding, visual replication, and creative writing. It can recreate websites from videos, generate full games in a single prompt, and handle 100+ self-directed agents in parallel via "agent swarm" mode. 

Unlike previous Chinese models that often overperformed on benchmarks but underdelivered in practice, Kimi K2.5 appears to match top Western models like Claude and Gemini in real-world tasks. It's also #1 on emotional intelligence benchmarks. While early, its strong showing and free access through Kilo Code make it a rising contender worth watching closely.

WATCH THE VIDEO ON YOUTUBE

Google DeepMind’s Project Genie: “infinite” interactive worlds you can sketch, explore, remix

This one is pure “future-feels-like-a-game-engine” energy.

Google is rolling out Project Genie as an experimental prototype that lets you create, explore, and remix interactive worlds. It’s powered by Genie 3 and also integrates Gemini plus Nano Banana Pro for a “world sketching” flow where you can preview and fine-tune the world before you jump in.

It’s built around three core moves:

• World sketching (text + generated/uploaded images; choose first/third person; tweak visuals)

• World exploration (it generates the path ahead in real time as you move)

• World remixing (build on others’ prompts, explore curated worlds, download videos)

A few real constraints (because prototypes are prototypes): worlds may not always follow prompts perfectly, characters can be finicky/laggy, and generations are currently capped at 60 seconds.

Availability is also limited: it’s starting with Google AI Ultra subscribers in the U.S. (18+), with expansion “in due course.”

Why it matters anyway: this is basically a public “touch it yourself” step toward world models—systems that can simulate an environment and how it changes when you act inside it. Even if you never use it, it’s a clue about where AI is heading: not just making images/videos, but generating spaces you can move through.

xAI launches Grok Imagine API: video generation + video editing, built to iterate fast

xAI dropped the Grok Imagine API positioning it as an “end-to-end creative workflow” bundle that can generate video (including from text or from an image) and edit video (restyle scenes, add/remove objects, control motion).

The part I think most people will feel immediately: they’re pushing hard on speed + cost + ability to iterate, not just “look how pretty the demo is.” The announcement explicitly says quality isn’t enough if latency and cost make experimentation painful.

They also lean on public benchmark-style comparisons, referencing Artificial Analysis and LMArena rankings/plots and stating the snapshot is “as of 2026-01-28.”

Practical takeaway: if you’re building content, ads, product demos, or anything “creative-but-repeatable,” this is another sign that video is turning into a prompt → preview → tweak → publish loop—more like design software than filmmaking. 

The “infrastructure era” gets louder: OpenAI is reportedly in talks for up to $60B

The chatter right now is that Nvidia, Microsoft, and Amazon are in discussions that could total as much as $60 billion into OpenAI.

The rough breakdown being floated:

• Nvidia: up to $30B

• Microsoft: less than $10B

• Amazon: “significantly more than $10B,” possibly over $20B

And it’s not just a “write the check” situation — there’s mention that Amazon’s piece could tie into bigger negotiations around cloud rentals and selling enterprise subscriptions through Amazon.

The feeling I can’t shake: we’re watching AI turn into a national-scale infrastructure game in real time — where “funding round” starts to look like “power plant + data center + supply chain.”

🧠RESEARCH

DeepSeek-OCR 2: Visual Causal Flow

DeepSeek-OCR 2 introduces "Visual Causal Flow," a new way to process document images using a visual encoder followed by a large language model. This method significantly improves text recognition and formatting in complex documents by treating visual data as a sequential flow, outperforming previous models on standard benchmarks.

Linear representations in language models can change dramatically over a conversation

This study reveals that the internal "linear representations" (how concepts are stored) in language models are not static but shift significantly as a conversation progresses. This "drift" means that the same concept might be encoded differently later in a chat, challenging current methods that assume fixed representations for interpreting model behavior.

Group Distributionally Robust Optimization-Driven Reinforcement Learning for LLM Reasoning

Standard training for AI reasoning often wastes effort on easy problems while neglecting hard ones. This paper proposes a new framework, GDRO, that uses "adversaries" to dynamically focus training on difficult prompts and reasoning paths. This targeted approach boosts the reasoning accuracy of models like Qwen3 by over 10% compared to traditional methods.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

ChatGPT Writer – Chrome Writing Assistant for Web Page - a lightweight browser assistant that helps you draft and refine emails and messages, “chat with” any web page to summarize or extract answers, fix grammar, translate, and rewrite in different tones.

Chatling – No‑Code Chatbots for Support, Sales & Lead Gen - a no‑code platform for building AI chatbots and agents that you can deploy on your website and WhatsApp to automate customer support, lead capture, and sales.

Chatmind – Turn PDFs, Videos & Webpages into Mind Maps - AI mind‑mapping tool that converts long‑form content—like PDFs, webpages, YouTube videos, audio recordings, images, and pasted text—into clear, editable mind maps.

📲SOCIAL MEDIA

🗞️MORE NEWS

Apple Acquires Q.AI Apple bought Israeli startup Q.AI for $2 billion to improve Siri. The team focuses on AI that reads facial expressions. This tech helps devices understand unspoken cues, making your interactions with computers feel more human.

SpaceX and xAI Merger Elon Musk may merge his rocket and AI companies before selling stock to the public. The bold plan includes launching computer servers into orbit, aiming to use the space environment to cut energy costs for AI.

Gemini for Maps Google Maps now lets walkers and cyclists talk to its Gemini AI. You can ask for directions or spot cool places nearby without looking at your screen, keeping your eyes safely on the path.

Music Industry Sues Anthropic Music companies are suing Anthropic for $3 billion, claiming its AI stole lyrics from 20,000 songs. The lawsuit argues the company used copyrighted music to teach its system without paying the artists who created it.

Cloudflare’s MoltWorker Cloudflare released MoltWorker, a way to run a personal AI assistant on the internet instead of a home computer. It lets you host a private digital helper cheaply using their global web infrastructure.

Open Source AI Risks Experts warn that free, public AI models are helping criminals. Bad actors are tweaking the shared code to build undetectable computer viruses and fake content, bypassing safety locks found in private, paid AI systems.

Tesla Pivots to Robots Tesla is ending production of its Model S and X cars to focus on AI. The company will repurpose factories to build "Optimus" robots, signaling a major shift from electric vehicles to humanoid assistants.

AI Toy Leak A smart toy exposed 50,000 private chats between kids and its AI. A flaw let anyone with a Google account view the logs, sparking fears about how safe these "intelligent" playthings really are for families.