Welcome, Pioneers!
Can AI make a movie you actually care about? OpenAI thinks yes. Critterz is sprinting to Cannes on a sub-$30M budget and a nine-month timeline. If that lands, the way films get made changes overnight. Meanwhile, publishers are feeling the squeeze from AI Overviews, MIT just taught bio models to show their work, and Alibaba’s speech tech is getting fluent in the noisy real world.
📌 In today’s Generative AI Newsletter:
OpenAI backs Critterz for a 2026 release after Cannes
Publishers vs Google AI Overviews blamed for vanishing clicks
MIT + IBM protein models that explain their picks
Alibaba Qwen3-ASR handles 11 languages in the wild
Special highlight from our network
Swedish Startup Releases AI That Takes 1,000+ Actions at Once
A new Swedish AI Startup, Incredible, just released Agentic LLMs that do real work. The models have access to 200+ apps out of the box, and can take thousands of actions in parallel.
Instead of handling one task at a time, their models execute 1,000+ tasks simultaneously using live-code. This is massive!
This approach allows you to:
Connect to 200+ apps out of the box.
Process massive datasets (up to 5GB instead of 2MB).
Take hundreds of actions at once (e.g., update 1,000 CRM records instantly).
It’s a powerful drop-in replacement for major LLMs like OpenAI, Anthropic, and Google, at a significantly lower cost.
Try it now at incredible.one
✨ OpenAI Brings AI to Cannes With $30M Animated Film
Image credit: Critterz TV
OpenAI is backing Critterz, an animated feature made largely with its AI tools, in a high-profile attempt to prove generative models can deliver cinema faster and cheaper than Hollywood. The film, developed with Vertigo Films and Native Foreign, is targeting a 2026 global release after a planned debut at the Cannes Film Festival.
What sets Critterz apart:
Lean production: Built on a budget under $30M and a timeline of just nine months. This is a fraction of what major animated features typically require.
AI-human blend: Artists sketch, AI tools generate visuals with GPT-5 and image models and human actors provide voices.
Industry play: The film is pitched as a case study to convince studios and executives that AI can reduce costs and accelerate production.
Creative roots: Originated as a short by OpenAI’s Chad Nelson, later expanded into a feature-length project with a script from writers behind Paddington in Peru.
The project lands as Hollywood battles AI firms in court over training data and copyright. Previous AI-made films, including DreadClub: Vampire’s Verdict, drew skepticism over quality and emotional impact. If Critterz finds acceptance with audiences and survives industry scrutiny, it could mark the moment AI filmmaking goes mainstream.
📰 Publishers Warn Google’s AI Search Is Cutting Traffic
Newspapers are banking on online revenue to replace falling circulation Credit: BBC News
News publishers are sounding alarms over Google’s AI Overviews, claiming the summaries at the top of search results are siphoning off traffic that once flowed to their sites. For outlets already battling falling print circulation and shrinking ad revenue, losing Google clicks could prove devastating.
The concerns mounting:
Traffic collapse: DMG Media reported click-through rates falling by as much as 89% when AI Overviews appear.
Revenue threat: Publishers argue Google is using their content without compensation, while reducing the incentive to click through.
AI Mode expansion: Now live in Hindi, Indonesian, Japanese, Korean and Brazilian Portuguese, after expanding to 180 markets in English last month.
Legal action: Groups including the Independent Publishers Alliance have filed a complaint with the UK competition regulator, asking for interim measures to curb Google’s practices.
Google insists overall traffic has stayed “relatively stable,” pointing to “quality clicks” as proof. But publishers are preparing for a future where default search delivers fewer referrals and are scrambling to shift audiences to newsletters, apps and direct alerts before the search gatekeeper changes the rules entirely.
⚛️ AI in Pharma Learns to Explain Its Predictions
Image Credit: IBM
Protein language models can predict which molecules might make good drug or vaccine targets, but their reasoning has been a black box. A new study from MIT, paired with IBM’s open-source biomedical models, shows how to pull back the curtain: by forcing AI to reveal the biological features it uses to make predictions.
How the method works:
Sparse autoencoder: MIT’s model takes a protein’s compressed representation and expands it into a sparse map, where specific features correspond to real biological functions such as binding, metabolism, or stability.
Feature alignment: Many of these extracted features match known protein families or molecular roles, letting scientists see which patterns drive a prediction.
Plain-language summaries: The features are further translated into natural language, turning complex sequence patterns into interpretable explanations.
Validation boost: By showing which features matter, researchers can more quickly decide whether a candidate molecule is worth pursuing.
Instead of handing over inscrutable predictions, these models can now highlight the biological signals behind their choices. The visibility makes it easier to win trust from regulators, funders and research teams. In drug discovery, where credibility can be as important as accuracy, AI that explains itself may finally move from a promising tool to a trusted partner.
🎤 Alibaba Releases Qwen3-ASR for Multilingual Speech Recognition
Image Credit: Alibaba
Alibaba has launched Qwen3-ASR-Flash, a multilingual speech recognition model built on Qwen3-Omni and trained on tens of millions of hours of multimodal data. The model supports 11 languages, including Chinese (with dialects like Cantonese, Wu, and Minnan), English, French, Spanish, Arabic, and more.
Highlights of Qwen3-ASR-Flash:
High accuracy: Surpasses rivals on benchmarks across 11 languages, including Chinese, English, French, Spanish, and Arabic.
Handles music and noise: Accurately transcribes singing voices and conversations, even with heavy background sound.
Contextual prompts: Users can bias results by providing text in any format, from keyword lists to full documents.
Accents and dialects: Supports regional accents in English and Chinese, as well as dialects like Cantonese, Minnan, and Wu.
Noise rejection: Identifies languages precisely and ignores silence or irrelevant background sounds.
Qwen3-ASR-Flash positions Alibaba as a serious player in advanced voice tech, where transcription must move beyond clean audio clips and into the messy reality of multilingual conversations, music, and noisy environments.
🚀 Boost your business with us. Advertise where 13M+ AI leaders engage!
🌟 Sign up for the first AI Hub in the world.
