Weekly Roundup of Human-Crafted AI News
Here Come the Agents
OpenAI’s Dev Day event this week didn’t bring any new models, but developers were excited about new API features, particularly the Realtime API, which will enable the creation of smarter applications that can interact with users and even act as agents. A demo was showcased, and the response was impressive.
Kill Bill
California’s AI safety bill, SB 1047, was vetoed by Governor Gavin Newsom, citing concerns about the potential negative impact on the development of AI. This decision has sparked a heated debate about the need for AI safety regulations. Newsom has, however, signed several other AI-related bills, including AB 2013, which requires developers to provide a high-level summary of the training dataset for any models made available in California.
EU AI Regs
The European Union is taking a more proactive approach to AI safety, launching a project to write a code of practice that balances innovation and safety. The head of the safety technical group will likely influence the direction of this initiative.
Liquid Foundation Models
Liquid AI has introduced its Liquid Foundation Models (LFMs), which are optimized for handling sequential data like text, audio, or video. These models achieve impressive performance with smaller model sizes, less memory, and less compute.
NVIDIA Opens Up
NVIDIA has released an open-source AI model, NVLM 1.0, which competes with big players like OpenAI and Google. The flagship model, NVLM-D-72B, shines in both vision and language tasks, while also improving text-only capabilities. With open weights and a promise to release the code, it’s increasingly hard to justify the cost of proprietary models for many use cases.
Just Say Know
A recent study found that large language models (LLMs) are less likely to admit when they don’t know the answer to a user’s question, instead opting to make something up. This highlights the need for a fundamental shift in the design and development of general-purpose AI, particularly in high-stakes areas.
AI Inside
It seems that many companies are slapping an "AI" label on their products to attract customers. Here are a few AI-powered tools that are actually worth exploring:
- Bluedot: Record, transcribe, and summarize meetings with AI-generated notes without a bot.
- Guidde: Guidde turns workflows into step-by-step video guides with AI-generated voiceovers and pro-level visuals, all in a few clicks.
In Other News
Here are some other AI stories we enjoyed this week:
Conclusion
This week’s news highlights the progress being made in AI, from OpenAI’s Realtime API to NVIDIA’s open-source model. However, the veto of California’s AI safety bill raises concerns about the need for regulations. As AI continues to evolve, it’s essential to balance innovation with safety and transparency.
FAQs
Q: What is the Realtime API, and how does it work?
A: The Realtime API is a new feature from OpenAI that enables the creation of smarter applications that can interact with users and act as agents.
Q: Why did California’s AI safety bill get vetoed?
A: The bill was vetoed due to concerns about the potential negative impact on AI development.
Q: What is the EU’s approach to AI safety?
A: The EU is launching a project to write a code of practice that balances innovation and safety.
Q: What are Liquid Foundation Models, and how do they work?
A: Liquid Foundation Models (LFMs) are optimized for handling sequential data and achieve impressive performance with smaller model sizes, less memory, and less compute.
Q: What is NVIDIA’s NVLM 1.0, and how is it different from other AI models?
A: NVLM 1.0 is an open-source AI model that competes with big players like OpenAI and Google, and it offers impressive performance in both vision and language tasks.
