Big Tech isn't ready

The Cracks in Big Tech’s AI Ambitions Are Showing

Even the biggest of the big stumble sometimes. Meta and Microsoft both tried to flex their AI muscles this week—and both exposed their weaknesses.

Meta’s benchmarks aren’t the whole story. Meta dropped two Llama 4 models, Maverick and Scout, this week. Maverick is its flagship, supposedly competing with GPT-4.5 and Google’s Gemini 2.0 Flash. Scout’s the lightweight version, made to run on a single Nvidia H100 GPU but said to top Google's Gemini 2.0 Flash-Lite and open-source Mistral 3.1 on major benchmarks. 

• The catch? Meta has been accused of gaming the system. The Maverick model tested on LM Arena is a special “conversational” version—not the standard one developers can actually use. So when Meta claims to beat Google and OpenAI across various benchmarks? It’s based on a model the company isn’t releasing to the public. 

Microsoft’s AI-generated Quake II demo has hiccups. Microsoft rolled out a playable level of Quake II powered by its Copilot AI. It’s technically impressive—you can explore the game world, shoot, jump, and interact.

• But the AI struggles with the basics, like object permanence. Look away from an enemy, even for less than a second, and it forgets the enemy existed, rendering the AI off-limits for many game devs.

Can AI Therapy Tools Deliver?

The idea of AI therapy used to sound like a Black Mirror plot line. Now, it’s looking more like a reality. A new study out of Dartmouth found that a mental health chatbot helped reduce mental health symptoms as effectively as traditional human therapy.

It’s the first randomized clinical trial of its kind, and it could be a major unlock in a field desperately short on providers. Researchers trained the AI over five years to follow gold-standard psychological practices, then ran a controlled trial with around 200 participants. The results? Comparable to outcomes you'd see in top-tier human-led therapy sessions.

The appeal: Cost, accessibility, and 24/7 availability that no human therapist can match. But the risks are just as obvious—privacy issues, shallow interactions, and potential bias.

China’s AI Is Catching Up Fast

A new report from Stanford found that Chinese AI models have rapidly caught up with their US-made counterparts in quality. A year ago, US models had a clear lead on benchmarks. Now? China’s models are nearing parity.

Code that feels like magic

Windsurf (formerly Codeium) just dropped a major upgrade. Their coding agent was already top-tier and the Tab autocomplete experience has gotten has gotten considerably better. Now, with tools like Browser Previews and native App Deploys, it's never been easier to ship new projects. With Windsurf, you get:

• An insanely capable coding agent, Cascade, that understands your intent and carries out tasks on its own

• AI suggestions and completions to your code that are actually relevant, powered by an industry-leading context awareness engine.

• Access to the best models for coding

And best of all?

Generate landing pages like you’re sketching on a napkin

Sleek lets you describe your product or company, then builds production-ready code for landing pages. Perfect for AI startups, SaaS apps, or anyone tired of cookie-cutter templates.

How you can use it:

• Build clean, structured landing pages instantly

• Export usable code for frameworks like Next.js and Tailwind

• Streamline your workflow with customizable templates

Pricing: Freemium with paid plans available

Effortless video creation with AI

Higgsfield AI uses diffusion transformer tech to generate lifelike human characters and smooth motion video. Whether you’re a marketer, content creator, or just want to mess around, this tool can keep your workflow fast and your output polished.

How you can use it:

• Generate personalized videos for marketing or social media

• Produce professional-quality content on mobile or desktop

• Use AI to streamline video creation without sacrificing quality

Pricing: Paid plans available with a free trial