Twice a week, Data Points brings you the latest AI news, tools, models, and research in brief. In today’s edition, you’ll find:
- SmallThinker builds a 3 billion parameter reasoning model
- Alibaba cuts prices on its Qwen models
- Google unveils the FACTS model benchmark
- Smolagents orchestrates smaller open source agents
But first:
Texas proposes far-reaching AI regulation with new liability and compliance rules
Texas legislators formally introduced the Texas Responsible AI Governance Act (TRAIGA), a comprehensive AI regulation bill that imposes strict requirements on AI developers, distributors, and deployers. The bill creates a powerful new AI regulator, mandates extensive compliance documentation for high-risk AI systems, and establishes negligence liability for algorithmic discrimination against protected classes. While some provisions have been refined since the initial draft, TRAIGA remains one of the most aggressive AI regulations proposed in the U.S., with potential to significantly impact AI development and deployment across many sectors. (Hyperdimensional)
OpenAI explores restructuring to secure funding for AGI
OpenAI’s Board of Directors announced a potential change to the company’s structure to better support its mission. The proposed plan would transform OpenAI’s for-profit arm into a Delaware Public Benefit Corporation, allowing it to raise capital with conventional terms. This restructuring aims to secure the substantial funding needed for AGI development, estimated to be in the hundreds of billions of dollars, while also creating a well-resourced non-profit arm to pursue charitable initiatives in sectors like healthcare and education. (OpenAI)
New compact model shows promise for reasoning tasks at the edge
Powerinfer researchers presented SmallThinker 3B Preview, a three billion parameter o1-like language model designed to excel at reasoning tasks. The model shows improved performance over its base Qwen2.5-3b-Instruct model on several benchmarks, including outperforming GPT-4 on some tests. SmallThinker 3B Preview’s small size makes it suitable for edge deployment on devices with limited computing power, potentially enabling more accessible and efficient AI applications in various fields. (Hugging Face)
Alibaba Cloud slashes visual AI model pricing by 85 percent
Alibaba Cloud reduced the price of its most advanced visual AI model, Qwen-vl-max, to 3 yuan ($0.41) per million input token uses, marking an 85 percent cut. This price reduction, announced on the last day of 2024, is Alibaba Cloud’s third AI price cut of the year. The move matches ByteDance’s recent pricing for a similar visual model, intensifying competition in China’s AI market. With Chinese regulators having approved 252 generative AI services for public use as of November, companies are aggressively lowering prices to attract customers and encourage adoption. (South China Morning Post)
New benchmark measures LLMs’ ability to ground responses in source material
Google DeepMind researchers presented FACTS Grounding, a comprehensive benchmark for evaluating large language models’ ability to generate factually accurate responses based on provided source documents. The benchmark includes 1,719 examples across various domains and uses multiple AI judges to assess responses for eligibility and factual accuracy. Google 2.0 Flash topped the initial leaderboard, followed by two other Google models, Claude 3.5 Sonnet, and GPT-4o. (Google DeepMind)
Open source models rival closed ones in AI agent tasks
A new Hugging Face library called smolagents enables developers to create AI agents using open source language models. The library supports “code agents” where AI models write executable code actions rather than JSON-like snippets, which research shows improves agent capabilities. In benchmark tests, leading open source models like Mixtral-8x7B performed comparably to closed models like GPT-4 on agent tasks, demonstrating that open AI systems can now match proprietary ones for building autonomous AI assistants and workflows. (Hugging Face)
Still want to know more about what matters in AI right now?
Read this week’s special issue of The Batch for an inspiring glimpse into AI’s potential in 2025, featuring insights from leading experts on generative AI, cinematic creativity, generalized intelligence, and the future of prosocial platforms.
In this week’s letter to readers and learners, Andrew Ng highlighted the excitement around AI’s potential in 2025, emphasizing the ease of building software prototypes with AI-assisted coding and its impact on productivity, creativity, and learning. He encouraged readers to make a learning plan, build prototypes, and embrace the fun and educational journey of creating with AI.
“One aspect of AI that I’m particularly excited about is how easy it is to build software prototypes. AI is lowering the cost of software development and expanding the set of possible applications. While it can help extend or maintain large software systems, it shines particularly in building prototypes and other simple applications quickly.”
Read Andrew’s full letter here.
Our New Year special issue explores the transformative potential of AI in 2025: generative AI liberating artists to focus on creativity while ensuring safety and accessibility; video models revolutionizing cinematic storytelling with integrated audio and video; AGI driving personalized and contextual interactions; data-efficient models enabling broader accessibility and sustainability; autonomous agents taking meaningful actions to simplify our lives and enhance productivity; and AI-powered platforms fostering empathy, collaboration, and unity in digital spaces.