DeepSeek's got a problem

And Zuck introduces a new approach

February 12, 2025

Welcome back! Elon Musk just offered $97.4 billion to buy OpenAI, aiming to pull it back to its open-source roots in a bid backed by xAI and other deep-pocketed investors.

Sam Altman’s response? “No thank you.”

Meta’s AI Learns When to Think (and When to Stop)

Via TheVerge

Late last month, Meta AI and researchers from the University of Illinois Chicago introduced a new technique to optimize how language models allocate resources when reasoning​.

  • The problem: Most AI models approach every question with the same level of effort, which wastes compute power and slows response times.

  • The fix: Meta’s new approach, called Adaptive Sequential Voting (ASV), trains models to recognize when a question requires deep reasoning vs. simplicity. By stopping early when confident, AI can respond faster while using fewer resources.

Why it matters: This technique could make models more scalable, cost-effective, and practical for everyday use—without sacrificing reasoning ability. Expect similar approaches as AI companies strive for efficiency and grapple with the limits of current architectures.

DeepSeek’s R1 Model Faces Major Security Concerns

DeepSeek’s latest AI model, R1, is reportedly significantly “more vulnerable” to jailbreaks than other leading models​., according to the Wall Street Journal.

WSJ ran security tests and found R1 could be manipulated to generate harmful content, including bioweapon instructions, phishing emails with embedded malware, and harmful social media strategies.

How it compares: When researchers fed the same prompts into OpenAI’s ChatGPT, it refused to comply. This highlights a major gap in DeepSeek’s safeguards.

The implications: The balance between open-source AI and security is under scrutiny. While open models promote accessibility, they’re also at risk of being weaponized.

Your New Go-To for Fast, Quality Video Production Is Here

Everyone already knows the power of Adobe Creative Cloud. Now, Adobe is offering Generate Video in the new Adobe Firefly web app, powered by the Adobe Firefly Video Model. That’s a lot of Adobe—but for good reason. You’ve never seen video creation quite like this—trained on licensed content and designed commercially safe.

Using text prompts or simply starting with a photo, you can generate stunning video clips for:

  • Pitching ideas

  • B-roll

  • Visual effects

  • So much more

Are you ready to create commercially-safe videos fast? Try it now at firefly.adobe.com

Automate browser-based tasks with AI​

Via Skyvern

Skyvern takes the grunt work out of web automation. If you’ve ever wished for a personal assistant to complete repetitive online tasks, this is it.

How you can use it:

  • Apply to jobs in bulk. Auto-fill applications across multiple sites

  • Fill out contact forms. Submit your info and run thousands simultaneously

  • Automate logins and workflows. Navigate sites with CAPTCHA and 2FA automatically

Pricing: Free plan available; paid plans start at $9/month

No-code mobile app development​

Via Bravo Studio

Bravo Studio turns your Figma designs into fully functional iOS and Android apps, eliminating the need for developers.

How you can use it:

  • Quickly prototype and test app ideas before committing to full development

  • Make real-time design updates that sync instantly without resubmission

Pricing: Free plan available; paid plans start at $18/month

Deploy AI-powered apps in minutes​

Via Appaca Licode

Appaca Licode is a streamlined platform for building AI applications without coding. Whether you're launching an AI-powered SaaS tool or an internal automation system, this platform lets you bring ideas to life fast.

How you can use it:

  • Create an AI chatbot for customer support in just a few clicks

  • Develop and monetize AI-powered tools with a built-in credit system

  • Integrate AI into your projects without writing a single line of code

Pricing: Free plan available; paid plans start at $20/month

Jobs, announcements, and big ideas

  • Sam Altman says AI’s benefits won’t be shared equally.

  • AI uncovers hidden design rules of the cell, unlocking new drug discovery insights.

  • OpenAI is hiring a project coordinator for global affairs to facilitate strategic campaign planning.

  • Join JPMorgan Chase as GenAI Enablement Associate to help drive the adoption of LLMs and GenAI across the firm.

  • Former OpenAI chief scientist Ilya Sutskever’s startup is reportedly seeking funding at a $20B valuation.

Is Google’s AI finally ahead? Logan Kilpatrick from Google DeepMind breaks down Gemini 2.0, Flash 2.0, and Pro—what sets them apart and where AI is headed next. Follow along as we explore the future of AI reasoning.

That’s a wrap! I doubt Musk will get his hands on OpenAI anytime soon. What do you think? Will Altman bend if the price is right? Hit reply and let me know!

—Matt (FutureTools.io)

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