Twice a week, Data Points brings you the latest AI news, tools, models, and research in brief. In today’s edition, you’ll find:
- Nvidia’s new Blackwell chips face months-long delays
- Mistral adds fine-tuning and other custom tools
- LangChain Studio IDE is built to build agents
- Character.AI open-sources its in-house prompt developer
But first:
GPT-4o gives developers more control over its outputs
OpenAI introduced Structured Outputs in their API, allowing model outputs to reliably adhere to developer-supplied JSON schemas. The new feature works with function calling on all models that support it, as well as with a new response_format parameter on the latest GPT-4o models. Supporting Structured Outputs may helo solve challenges developers face in generating structured data from unstructured inputs, achieving nearly perfect reliability in matching output schemas through a combination of model training and constrained decoding techniques. (OpenAI)
Figure unveils robot with enhanced AI capabilities
Figure introduced its second-generation humanoid robot, Figure 02, featuring significant hardware and software improvements. The robot incorporates speech-to-speech conversation abilities, an onboard vision language model, a 2.25 KWh battery pack, integrated wiring, and advanced hands with 16 degrees of freedom. Figure recently tested the robot at a BMW manufacturing plant and plans to develop humanoid robots for both industrial and domestic applications in the future. (IEEE Spectrum)
Design flaws push back release of Nvidia’s next-gen chips
Nvidia informed customers that its upcoming Blackwell series AI chips will be delayed by at least three months due to design flaws discovered late in the production process. The delay affects chips ordered by major tech companies like Microsoft, Google, and Meta, who collectively placed orders worth tens of billions of dollars for use in developing advanced AI models. This setback could impact AI development timelines for these companies and raises questions about Nvidia’s ability to meet high revenue projections for its new chips in 2025. (The Register)
Mistral AI offers fine-tuning, agents, a new SDK, and more
The company now allows customization of its flagship models like Mistral Large 2 through fine-tuning, few-shot learning, or base prompts on their La Plateforme service. Mistral also introduced an alpha version of Agents for creating custom AI behaviors and workflows, and released version 1.0 of its client SDK for Python and TypeScript. These additions simplify the process of tailoring large language models for specific use cases and integrating them into applications. (Mistral)
New IDE simplifies development of agent systems
LangChain introduced LangGraph Studio, an integrated development environment (IDE) for building and testing AI agents. The tool allows developers to create, visualize, and debug complex multi-agent systems using a graphical interface, supporting both code and no-code approaches. LangGraph Studio aims to simplify the development of AI agents by providing features like step-by-step execution, state inspection, and easy integration with existing LangChain components. (LangChain)
Character.AI unveils Prompt Poet for streamlined prompt creation
Character.AI developed Prompt Poet, a tool (now released under an MIT license) that simplifies the creation of complex, dynamic prompts for large language models. The system uses a combination of YAML and Jinja2 templating to allow both technical and non-technical users to design prompts efficiently. Prompt Poet offers features like template composition, custom encoding functions, and cache-aware truncation to optimize prompt performance and GPU usage. This approach shifts focus from manual string manipulation to a more intuitive, design-focused method of crafting AI prompts. (Character.AI)
Still want to know more about what matters in AI right now?
Read this week’s issue of The Batch for in-depth analysis of news and research.
This week, Andrew Ng introduced his new sequence of courses, AI Python for Beginners, aimed at teaching anyone to code with the help of AI:
“These courses teach coding in a way that is aligned with these trends: (i) We teach how to write code to use AI to carry out tasks, and (ii) Unlike some instructors who are still debating how to restrict the use of ChatGPT, we embrace generative AI as a coding companion and show how to use it to accelerate your learning.”
Read Andrew’s full letter here.
Other top AI news and research stories we covered in depth: Google gets Character.AI co-founders, how employers and prospective employees are embracing automated hiring tools, Ukraine's aquatic drones, and ArtPrompt, a technique to test the impact of text rendered as ASCII art on LLM performance.