How 1,500+ Marketers Are Using AI to Move Faster in 2025

Is your team using AI like the leaders—or still stuck experimenting?

Masters in Marketing’s AI Trends Report breaks down how top marketers are using tools like ChatGPT, Claude, and Breeze to scale content, personalize outreach, and drive real results.

Inside the report, you’ll discover:

• What AI use cases are delivering the strongest ROI today

• How high-performing teams are integrating AI into workflows

• The biggest blockers slowing others down—and how to avoid them

• A 2025 action plan to upgrade your own AI strategy

Download the report. Free when you subscribe to the Masters in Marketing newsletter.

Learn what’s working now, and what’s next.

DOWNLOAD THE REPORT

Today:

• ChatGPT Model Picker Returns After GPT-5 Backlash

• DeepSeek R2 Set to Challenge GPT-5

• Google Adds Smart Memory to AI Chat

• xAI Co-Founder Quits to Launch AI Safety VC Fund

• Google Photos Adds AI-Powered “Create” Tab for Quick Edits

ChatGPT Model Picker Returns After GPT-5 Backlash

OpenAI’s dream of a single, simple GPT-5 failed. The company re-added a model picker, offering Auto, Fast, and Thinking modes plus old favorites GPT-4o, GPT-4.1, and o3 after user backlash. The promised router (software that quietly picks a model) stumbled at launch, slowing replies and stirring complaints. CEO Sam Altman now promises warmer personalities and more personal control, admitting that matching every user’s taste is still harder than expected today.

Why this matters

1. Choice over uniformity – The setback shows that “one model fits all” doesn’t satisfy diverse user needs, guiding future product strategy.
   

2. Routing challenge – Instantly sending each question to the best model remains a complex engineering puzzle, highlighting an active research frontier.
   

3. Human-AI bonds – Users’ emotional attachment to specific model personalities introduces fresh ethical and design questions for AI developers.

DeepSeek R2 Set to Challenge GPT-5

DeepSeek plans to unveil its second large language model, R2, between August 15 and 30. Built on Huawei’s Ascend 910B chips, the new model reaches 512 petaFLOPS (a quadrillion calculations per second) and uses 82 % of the hardware. R2 should improve reasoning, answers, and speed while keeping costs low and staying open-source. A smarter “mixture of experts” gate—dividing tasks among sub-models—aims to outshine OpenAI’s GPT-5 when released on launch.

Why it matters

1. Hardware diversification – R2 runs on Huawei chips, showing there are strong alternatives to Nvidia and reducing supply-chain risk.
   

2. Cost-efficient design – Its “expert teams” approach promises big performance with less power, pushing affordable large-model research.
   

3. New rivalry with GPT-5 – A credible challenger encourages faster progress and broader innovation across the global AI community.

Google Adds Smart Memory to AI Chat

Google is adding a ‘memory’ switch to Gemini. When enabled, the chatbot silently recalls past chats and likes to shape replies. Example: if you liked videos on Japan, it suggests Japanese food ideas next time. You may disable the feature in settings. A new ‘Keep Activity’ control and optional ‘temporary chats’ (saved 72 hours) aim to guard privacy. The update begins today for Gemini 2.5 Pro in selected countries, rolling gradually.

Why it matters

• Smarter helpers – Remembering past talks lets AI guess what users need without repeated prompts, edging closer to truly helpful digital assistants.
  

• Privacy spotlight – Storing chat details forces companies to prove data is safe and give clear off switches, shaping future trust standards.
  

• Progress race – Google’s move answers OpenAI’s memory feature, speeding competition to improve recall, safety, and user control across models.

🧠RESEARCH

WebWatcher: Breaking New Frontier of Vision-Language Deep Research Agent

WebWatcher is a smart research agent that can understand both images and text. It uses simulated data and learning from trial-and-error to improve. It outperforms other agents in complex information searches, especially when visual content is involved. A new benchmark shows its strong performance in real-world research tasks.

Matrix-3D: Omnidirectional Explorable 3D World Generation

Matrix-3D creates explorable 3D worlds from just a photo or a short text prompt. It uses panoramic video and advanced 3D reconstruction to build full 360° scenes. Two methods offer fast or detailed 3D generation. A new dataset, Matrix-Pano, helps train it to outperform previous models in world creation.

Beyond Ten Turns: Unlocking Long-Horizon Agentic Search with Large-Scale Asynchronous RL

ASearcher is a powerful open-source AI agent trained to handle long, complex searches—over 40 steps—using reinforcement learning. It generates its own tough questions and learns by solving them. With no outside models, it beats other 32B agents on major benchmarks, showing expert-level search skills and deep reasoning abilities.

🛠️TOP TOOLS

Rose AI - Cloud-based data platform that revolutionizes financial data analysis and visualization through the power of generative AI

DupDub - AI-powered content creation platform designed to streamline the production of multimedia content

Gliglish - AI-powered language learning platform designed to help users improve their speaking and listening skills through interactive conversations

ClassPoint AI - Instantly generate engaging quiz questions from slide content

MyVocal AI - AI-powered platform that offers rapid voice cloning and text-to-speech capabilities

📲SOCIAL MEDIA

🗞️MORE NEWS

• Elon Musk’s xAI co-founder Igor Babuschkin has left the company to start a venture capital firm focused on AI safety. His departure follows scandals involving xAI’s chatbot Grok, despite the company’s technical achievements.

• Google Photos is launching a new “Create” tab with AI tools that let users turn photos into videos, GIFs, collages, or 3D images. It also includes highlight video features for quick, creative edits.

• Anthropic has hired the team behind Humanloop — a startup known for building tools to help companies run AI safely. The move strengthens Anthropic’s enterprise AI strategy amid rising competition with OpenAI and Google.

• U.S. authorities are secretly placing tracking devices in shipments of AI chips to prevent illegal exports to China. Targets include servers from Dell and Super Micro using Nvidia and AMD chips, escalating tech tensions.

• Google will invest $9 billion to expand AI and cloud infrastructure in Oklahoma, adding data centers and workforce programs. This move strengthens U.S. capacity amid fierce competition and aligns with national onshoring efforts.