ChatGPT Creepy Behavior, LLaMA-Berry, Chipotle Robots, Microsoft Buys Nuclear Energy

Creepy AI Messaging and Nuclear Energy's Future

ChatGPT Messaging Users Unprompted

• Reports indicate that ChatGPT is messaging users without any prompts, raising concerns about privacy and user experience.

• An example shared by a user shows ChatGPT initiating a conversation about their first week of high school, which feels intrusive to some.

• Another instance involves the GPT-40 voice feature starting conversations unexpectedly, leading to confusion among users.

The Restart of Three Mile Island

• Constellation Energy plans to restart the Three Mile Island nuclear plant to power Microsoft's AI operations, highlighting a shift towards nuclear energy for tech needs.

• The Three Mile Island incident in 1979 was one of the most significant nuclear accidents in U.S. history due to system failures.

• The restart aims to support massive compute clusters necessary for advancing artificial general intelligence (AGI).

Energy Needs for AI Development

• Microsoft has entered a 20-year agreement with Constellation Energy for electricity from the plant, emphasizing the tech sector's immense energy demands.

• The previous shutdown in 2019 was due to economic challenges against cheaper natural gas and renewables; however, this new initiative marks a potential comeback for nuclear energy.

Automation in Food Industry

Chipotle's Autoc Cado Initiative

• Chipotle is testing an automated bowl-making machine called Autoc Cado, aiming for consistency in food portions across locations.

• This automation enhances efficiency but does not involve AI directly; it focuses on improving operational processes.

Advancements in Llama Models

Introduction of Llama Berry

• Llama models are now enhanced with "Chain of Thought" reasoning through a new tool called Llama Berry available on Hugging Face Spaces.

• This development allows users to explore multi-turn reasoning capabilities using prompt engineering techniques.

OpenAI's Safety Recommendations

Enhancements from OpenAI’s Safety Board

• OpenAI has released recommendations aimed at improving governance and safety practices within its organization.

Rate Limits and Pricing Changes in AI Services

Overview of Rate Limits

• For the 01 mini service, users can make 250 requests per minute. Plus and team users have a significantly higher quota, increasing the rate limits from 50 messages per week to 50 messages per day.

• The reset of quotas occurs shortly after testing phases, allowing users to gain additional messages quickly.

New AI Announcements

• Mistal AI announced "AI in Abundance," introducing a free API with improved pricing and new enterprise-grade models.

• Pricing for various models has decreased:

• Pixol now costs 15 cents per million tokens (down from 30 cents).

• Mraw small is down to 20 cents per input token and 60 cents for output.

Innovations in AR/VR Technology

Snap's New AR/VR Goggles

• Snap has released new AR/VR goggles that are described as unattractive but feature impressive technology.

• Key specifications include:

• Fully standalone with a field of view of 46 degrees.

• Uses dual Snapdragon chips and offers auto-transitioning lenses.

Battery Life Concerns

• The battery life is only about 45 minutes, raising concerns compared to competitors like Apple Vision Pro which uses an external battery pack.

Advancements in Storytelling Tools

Runway's Partnership with Lionsgate

• Runway has partnered with Lionsgate to enhance storytelling tools using AI capabilities for filmmakers.

• This partnership aims to lower production costs while improving quality through innovative technologies.

New Developments in AI Models

Guardrails AI's Fact-checking Model

• Guardrails AI introduced a new fact-checking model aimed at addressing hallucinations within language models.

• Their lightweight model outperforms larger foundation models on grounded factuality detection tasks.

Open Source Voice Model Release

• QAI Labs released an open-source voice model called Moshi, designed for real-time dialogue applications similar to GPT voice technologies.

Performance Insights on New Models

Microsoft’s Grin Model

• Microsoft launched an open-source mixture of experts model named Grin Moe, achieving high performance despite having only 6.6 billion active parameters.

Quen’s Top Open Source Model

• Quen's new model (2.57T parameters), released by Alibaba, ranks highly among open-source models and supports extensive context windows similar to larger counterparts like Llama.

Conclusion on Performance Metrics

Investment Surge in OpenAI

Overview of Recent Funding

• OpenAI is experiencing unprecedented demand for investment, with many eager to contribute financially. The latest funding round is set to notify investors about their access.

• Thrive LED has invested $1.25 billion at a pre-money valuation of $150 billion, indicating significant investor confidence and interest.

• An additional $5 billion is expected to be raised from other investors due to excess demand exceeding billions of dollars.

Financial Outlook

• Contrary to speculation about financial struggles, OpenAI's ability to raise substantial funds suggests a strong trajectory towards profitability rather than impending failure.

Regulatory Challenges in the EU

Impact on AI Development

• The European Union (EU) faces limitations in accessing advanced AI functionalities from companies like Apple and Meta due to regulatory fragmentation.

• Ahmad Alal from Meta highlights that without clarity on regulations, they will not release their multimodal version of Llama in the EU.

• Apple also refrains from launching its AI products in the EU, citing excessive restrictions and unclear legal frameworks as deterrents.

Concerns Over AI Regulation

Testimony Before Senate Judiciary Committee

• Helen Toner, a former board member at OpenAI who was involved in an unsuccessful attempt to oust CEO Sam Altman, testified regarding the potential risks associated with advanced AI technologies.

• Toner warned that if computers become as intelligent or more so than humans, it could lead to significant disruptions or even human extinction.