Welcome, AI Trendsetters!

Amazon’s new Nova Sonic voice model is outperforming GPT-4o in tough environments—and costs 80% less. Mira Murati is reuniting OpenAI’s original minds under a bold new lab. Snapchat is automating ad creation with AI-powered lenses. And NVIDIA, with Stanford, just cracked the code for longform AI cartoons. The battleground for next-gen media is shifting fast.

In today’s Generative AI Newsletter:

• Amazon Nova Sonic beats GPT-4o on speed, cost, and clarity
• Murati’s Thinking Machines reunites OpenAI’s original architectsr
• Snapchat ditches 3D pipelines for selfie-based AI ads
• NVIDIA and Stanford unlock coherent longform AI video

🗣️Amazon’s New AI Speaks Sharper, Films Smarter

Amazon has released Nova Sonic, a new speech-to-speech model that delivers more accurate, responsive voice interactions than OpenAI’s latest offerings at a fraction of the price. The launch was accompanied by Nova Reels 1.1, an upgraded video generation model with improved fidelity and extended output length. Both are now available through Amazon Bedrock, the company’s platform for foundation model access.

Key developments:

• Nova Sonic responds with a latency of just over one second and achieves a word error rate of 4.2 percent across multiple languages
• In noisy, multi-speaker environments, Nova Sonic outperformed GPT-4o by nearly 47 percent in transcription accuracy
• Amazon claims the model costs approximately 80 percent less than comparable voice systems from OpenAI
• Nova Reels 1.1 allows up to two minutes of video generation, with support for both automated prompts and manual shot-by-shot editing
• The video model introduces improved visual quality and greater consistency in character and style across scenes

Amazon’s latest launch signals a more aggressive push into generative AI infrastructure. With enhancements in voice, video, and its expanding suite of developer tools, the company is positioning Bedrock as a viable alternative to more established AI platforms. These advances also point to Amazon’s growing ambitions in agentic computing, digital content creation, and multimodal interaction.

🧠 Ex-OpenAI Execs Reunite Under Mira Murati’s New AI Lab

Thinking Machines Lab, the AI startup launched by former OpenAI CTO Mira Murati, has quietly expanded its leadership circle with two more high-profile hires from OpenAI’s inner circle. Bob McGrew, former chief research officer, and Alec Radford, one of the original architects of GPT, have both joined the company as advisers.

Key developments:

• 19 of 38 founding team members have previously worked at OpenAI, including co-founder John Schulman, who now serves as chief scientist
• McGrew joined OpenAI in 2017 and left in 2024 after serving as VP of research and CRO
• Radford, who helped create GPT, Whisper, and DALL·E, left OpenAI last year to pursue independent work before quietly joining Murati's venture
• The company has not disclosed its product roadmap but is focused on building more “customizable and generally capable” AI systems
• Reports suggest Thinking Machines is seeking to raise up to $1 billion at a $9 billion valuation

Thinking Machines now houses many of the minds behind ChatGPT and OpenAI’s early breakthroughs. While the company remains tight-lipped on its direction, its growing roster of elite talent hints at significant ambitions and raises the stakes for the next generation of AI labs emerging from OpenAI’s shadow.

🪞 Snapchat Launches AI-Powered Ad Lenses

Snapchat is rolling out Sponsored AI Lenses, a new ad format that uses generative AI to create immersive, selfie-based brand experiences. It’s a strategic shift from bespoke 3D production toward scalable, algorithmically generated creativity — aimed at making high-impact ads faster and more cost-effective.

What’s New in Sponsored AI Lenses:

• Selfie-driven storytelling: Users take a photo, and Snap’s AI integrates them into one of up to 10 preset, stylized scenes
• No 3D pipelines: Brands bypass VFX and modeling with Snap’s AI templates
• Higher engagement: Early campaigns by Uber and Tinder saw above-average playtime
• Increased visibility: Placement in the Lens carousel boosts daily reach to over 300M users

Snap’s move reflects a broader trend: the automation of creative work in advertising. By replacing traditional asset production with AI-generated templates, the company is not only lowering the cost barrier for brands, but also reinforcing the shift toward identity-based, participatory media. In a feed-first world, Snapchat is doubling down on faces, not formats.

🎬 NVIDIA and Stanford crack the code for longform AI cartoons

A new AI technique from NVIDIA, Stanford, and collaborators is pushing past the limits of short-form generation. By introducing Test-Time Training layers, the team has produced minute-long animated videos with scene and character consistency which is something no current model has achieved at this scale.

Key Points:
• Test-Time Training acts as a live memory system, helping the AI stay coherent across time
• Built on top of CogVideo-X, the system evolves from generating three-second clips to full narrative minutes
• Demo clips recreate Tom and Jerry-style cartoons with fluid motion and persistent visual logic
• The architecture combines global learning with local attention, keeping computation efficient
• Users can guide generation using anything from single-sentence prompts to full paragraph-level storyboards

AI video has dazzled with fidelity but faltered on narrative. With this breakthrough, models no longer need to cheat coherence through editing or stitching. The door is now open for native, longform storytelling where a scene flows naturally, a character remembers, and a minute of animation actually makes sense.

🚀 Boost your business with us—advertise where 10M+ AI leaders engage

🌟 Sign up for the first AI Hub in the world.

📲 Our Socials