Google, meet Claude

Microsoft’s AI just ditched the GPU

April 18, 2025

Welcome back! If your parents complain that you never call them, a startup called inTouch has a fix: send an AI to do it for you. The service uses voice models to spark "mind-stimulating conversations" with your family—and it sounds exactly like an automated cold call. Will grandma be able to tell the difference?

Claude Just Became Your Smartest Coworker

On Tuesday, Anthropic launched a massive upgrade to its Claude assistant, giving it the ability to search your entire Google Workspace and autonomously conduct research.

What’s new: Claude can now dig through your emails, calendar, and docs to prep for meetings, summarize launch plans, or answer complex questions with source citations. Its new Research mode chains together live searches—no human prompt needed.

How it stacks up: Claude is gunning for the sweet spot between depth and speed.

  • OpenAI’s Deep Research is still the most rigorous. It questions assumptions, parses source conflicts, and offers academic-level reports but takes 15–30 minutes per task.

  • Google’s Gemini is fast and clean but leans on summaries, not synthesis. It can’t handle file uploads or the kind of critical reasoning Claude and GPT deliver.

  • Claude? It’s now offering comprehensive answers in under a minute, pulling from both your internal documents and the web, and citing sources as it goes.

The bigger picture: Claude is part of the broader shift away from passive tools and toward AI that reasons, summarizes, and gets things done like a teammate.

OpenAI’s Next Move: Coding Domination

As OpenAI debuts newer, smarter models, it’s also pursuing AI expertise outside of its own company. OpenAI is reportedly in talks to buy AI coding startup Windsurf for around $3 billion, which would be its biggest acquisition to date.

What Windsurf does: Formerly Codeium, the tool helps devs write code faster—think GitHub Copilot, Cursor, or Replit. It’s popular among programmers who want to "vibe code," aka write software with minimal typing and maximum AI help.

Why it matters: The deal would bring OpenAI deeper into the AI developer stack and put pressure on Microsoft, Anthropic, and Anysphere. AI coding assistants are becoming essential—and this acquisition signals OpenAI wants to be the default tool.

Microsoft’s AI Breakthrough…With a Catch

Microsoft introduced a new 1-bit model called BitNet b1.58 this week, designed to run on CPUs instead of GPUs. That’s a big deal because most high-performing models today are GPU hogs, eating up memory and cash. 

It’s faster and leaner. BitNet uses dramatically less memory and can run efficiently on Apple’s M2 chip. That makes it ideal for laptops or edge devices where GPU access is limited.

But it doesn’t work with GPUs. The model only works with Microsoft’s custom framework and doesn’t support GPUs—still the backbone of most AI infra today.

The takeaway: Everyone’s racing to make AI smaller, faster, and cheaper. But unless it runs on the right hardware, even the most efficient model could be stuck in the lab.

AI product photos without the studio shoot

BetterStudio

BetterStudio transforms basic product images into polished, conversion-optimized fashion shots using AI-generated models and backgrounds.

How you can use it:

  • Swap flatlays for model photos in minutes

  • Bulk-edit product shots for Shopify drops

  • Get pro-level images without the production cost

Pricing: Paid and free plans available

Offline subtitles in 19 languages

Submarine Player

Submarine Player auto-generates accurate English subtitles for any video—no internet required.

How you can use it:

  • Translate international films instantly

  • Export captions in multiple formats for YouTube or TikTok

  • Dual-subtitle mode for language learners

Pricing: Paid and free plans available

Jobs, announcements, and big ideas

  • I’m hiring a full-time Production and Operations Assistant to help with managing comms, doing with research, building relationships with brands, and more. Learn more and apply here.

  • OpenAI launches Flex API to handle slower tasks without burning your budget.

  • Wikipedia uploads article data to Kaggle to stay ahead of AI scraping.

  • Google previews Gemini 2.5 Flash with faster speed, smarter reasoning, and lower cost.

  • AMD takes an $800 million hit after new US rules tighten AI chip exports.

  • ChatGPT photo trend raises alarms as users start reverse searching locations.

  • Kling AI 2.0 launches with video and image editing and racks up 22 million users.

Llama 3 is making noise. Follow along as I unpack the hype, the specs, and the controversy.

That’s a wrap! See you next week.

—Matt (FutureTools.io)

P.S. This newsletter is 100% written by a human. Okay, maybe 96%.