Daily AI Newsletter: OpenAI’s Sora video generator appears to have leaked

In today’s email:

🚓 Police departments across U.S. are starting to use AI to write crime reports
😮 Over ½ of Long Posts on LinkedIn are Likely AI-Generated Since ChatGPT Launched
👀 OpenAI moves to trademark its o1 ‘reasoning’ models
🧰 10 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI’s Sora video generator appears to have leaked

A group calling itself "Sora PR Puppets" leaked access to OpenAI's Sora video generator in protest of what they allege is OpenAI’s exploitative and misleading handling of the tool's early access program. The group published a project on Hugging Face tied to Sora’s API, enabling users to generate short 1080p videos via a public frontend. This unauthorized access lasted for three hours before OpenAI and Hugging Face disabled it. The group criticized OpenAI for leveraging unpaid labor from artists in testing the tool while tightly controlling outputs and narratives about Sora’s capabilities.

The group accused OpenAI of prioritizing PR over genuine artist collaboration. They claim Sora testers are pressured to frame the tool positively, with limited creative freedom and pre-approval required for sharing works. In their statement, the group called on OpenAI to be more transparent, artist-friendly, and supportive of the arts beyond promotional efforts. OpenAI has not yet commented on the incident.

Sora, which debuted earlier this year, has faced technical challenges, including long processing times and consistency issues in generated content. While a “turbo” variant of the model reportedly improves speed and customization, OpenAI has yet to overcome hurdles like safety, impersonation, and scalability. Meanwhile, competitors like Runway and Stability are making headway, with notable industry partnerships such as Lionsgate and James Cameron. OpenAI’s delays and restricted access have put it at a disadvantage in the rapidly advancing video generation space.

Scale your web scrapers with Apify’s Black Friday Boost plan

Get a 30% prepaid usage bonus on Apify every month for 6 months!

Join the largest ecosystem where developers build, deploy, and publish data extraction and web automation tools.

Here's why developers love Apify:

- Marketplace of 2,000+ scrapers.
- JavaScript, TypeScript, and Python code templates.
- Proxies and unblocking baked in.
- Export and integrate results in JSON, CSV, Excel, and more.
- Publish your own scrapers and reach thousands of new customers.

Hurry, it ends on December 5!

Claim your bonus now

Alibaba releases an ‘open’ challenger to OpenAI’s o1 reasoning model

Alibaba has introduced QwQ-32B-Preview, a new reasoning AI model designed to rival OpenAI’s o1 series. With 32.5 billion parameters, it surpasses o1-preview and o1-mini in certain benchmarks like AIME and MATH, which evaluate logic and math problem-solving. QwQ-32B-Preview also supports prompts up to 32,000 words, showcasing strong capabilities in solving complex logic puzzles and math tasks. However, it has limitations, including occasional language switching, looping errors, and struggles with common-sense reasoning.

The model incorporates a self-fact-checking mechanism, reducing common AI errors but increasing response time. Available for download on Hugging Face under an Apache 2.0 license, it can be used commercially but remains only partially open, with limited insight into its internal workings. Like other Chinese-developed models, QwQ-32B-Preview complies with regulations by avoiding politically sensitive topics or aligning with government narratives, such as affirming Taiwan as “inalienable” to China.

QwQ-32B-Preview reflects the growing focus on reasoning models as scaling traditional AI approaches like increasing data and compute power delivers diminishing returns. These models use test-time compute—extra processing during inference—to tackle problems more effectively. Big players like Google are also heavily investing in reasoning models, with expanded teams and resources, signaling this as a key direction for AI research and development.

Police departments across U.S. are starting to use AI to write crime reports

Police departments across the U.S. are increasingly adopting artificial intelligence tools to ease administrative burdens, with AI showing particular promise in drafting crime reports. Axon, known for its Tasers and body cameras, has introduced Draft One, an AI tool that generates police narratives from bodycam audio. Piloted in departments in Colorado, Indiana, and California, Draft One reduces report-writing time by over 60%, saving officers approximately 45 hours per month. Officers like Sergeant Robert Younger of Fort Collins, Colorado, commend the tool for producing accurate, fact-based drafts in under 10 seconds. However, Axon restricts its use to minor incidents for now, allowing agencies to build trust in the system before expanding its scope.

Despite the efficiency, legal experts voice concerns about AI’s transparency, accuracy, and potential bias. Defense attorney Stephanie Pitcher and trial attorney David Schwartz argue that while AI-generated reports might save time, they raise questions about evidence admissibility and officer diligence. AI-generated inaccuracies, or “hallucinations,” could lead to inconsistencies that defense attorneys might exploit in court. To address these issues, Axon includes safeguards such as mandatory officer review, editing, and supervisor oversight before reports are finalized.

Other companies like Truleo and 365Labs focus on alternative approaches, with Truleo using real-time dictation and 365Labs providing grammar and error correction tools. Truleo CEO Anthony Tassone criticizes AI-driven transcription of bodycam footage as inefficient, emphasizing ethical, conversational AI instead. Experts like Cassandra Burke Robertson highlight the need for clear documentation of AI processes to ensure accuracy and maintain public trust, particularly in sensitive investigations.

While skeptics question the need for AI in police reporting, supporters like Younger believe it could transform law enforcement by reducing administrative workloads. Younger notes that these tools might help retain officers who are disillusioned by paperwork demands, describing AI as part of a broader, evolving process to modernize policing while preserving accuracy and oversight.