AI Safety is For Losers

Good morning.

This week’s news seems to be themed around AI risk and regulation.
The EU AI Act has jumped the final hurdle before it will become a reality.

In previous editions, we talked about China, Europe, and the US competing in a race towards (depending on who you talk to) Artificial General Intelligence (AGI) or more funds for Large Language Models pretending to be AGI.

But critics of the EU AI Act might be right to assume that Europe, by being first in regulating AI development, might also stifle innovation and drive startups to countries where there is less oversight.

The Biden administration has published an executive order regarding AI but it’s looking to be not anywhere near as stringent as what the EU is doing.

Meanwhile, our friends at OpenAI have dissolved their safety team after key players left the company in open disagreement with the company’s CEO, Sam Altman.

It’s a precarious balance between regulation and innovation and I’m not sure Europe is doing itself a favour by being quick to bolt things down.

Welcome to the Blacklynx Brief.

AI NEWS
AI Safety is for Losers

• As mentioned above OpenAI has dissolved its AI safety team focused on long-term risks after key leaders Ilya Sutskever and Jan Leike resigned last week. The team, called the Superalignment team, will now be part of broader research efforts. Leike mentioned disagreements over the company's priorities, saying that safety processes have been overshadowed by product development. However, OpenAI President Greg Brockman reassured everyone that the company remains committed to AI safety.

• As of this week - ChatGPT users can link their Google Drive or Microsoft OneDrive accounts to the ChatGPT app so you can upload documents from your cloud environment directly. There are some safety concerns about this however. Not sure if it’s wise to give OpenAI access to all your files (especially when you read the news item just above)
  

  

• Apple and OpenAI are set to announce a partnership at WWDC on June 10th. They plan to integrate OpenAI’s AI technology into iOS 18. Apple wants to improve its AI features, complementing its own improvements to Siri, auto-summarizing notifications, and proactive intelligence features.

• Apple has announced new accessibility features for iOS 18. These include AI-powered Eye Tracking, Music Haptics, and Vocal Shortcuts. Eye Tracking lets users control their devices with their eyes, Music Haptics provides a tactile music experience, and Vocal Shortcuts allow custom Siri commands. These features will be available with iOS 18 and iPadOS 18 updates later this year, enhancing accessibility for users with physical disabilities.

• OpenAI has signed a deal to access real-time content from Reddit’s data API. This will allow ChatGPT to link discussions from the site. In return, Reddit will use OpenAI's technology to develop new AI features for its users and moderators. OpenAI will also become a Reddit advertising partner. This partnership will enhance ChatGPT’s capabilities and support OpenAI's AI Search chatbot with real-time information from Reddit.

• Sony Music is sending warning letters to over 700 AI developers, including major companies like OpenAI, Microsoft, and Google. They want them to stop using their artists' content without permission for AI development. The letters prohibit using music from artists such as Harry Styles, Adele, and Beyoncé, and request Spotify and Apple to protect against unauthorized AI use.

• Microsoft announced several AI advancements before its BUILD conference. These include the first 'AI-first' PCs, enhanced by Copilot+ for faster and more efficient AI tasks, and new features like Recall for easy screen search and a speaking Copilot. Windows 11 has been updated for AI integration, with Copilot assisting directly within the OS, and the new AI-enhanced PCs will start at $999

• At the same event, Microsoft introduced new AI features, including Copilot+ Recall, which constantly takes screenshots to remember user actions, and the multimodal Phi-3-vision model. They also unveiled Team Copilot for assisting in Microsoft apps and Copilot Studio for developers to create AI agents. Additionally, Microsoft announced a partnership with Khan Academy to provide AI tools to U.S. educators.

• Dell is expanding its partnership with Nvidia and releasing new servers, PCs, and services to boost its AI capabilities. The new Dell PowerEdge servers, featuring Nvidia’s Blackwell GPUs, promise faster and more efficient AI task processing. Dell is also introducing an AI Factory to simplify and speed up AI adoption for businesses and a new GenAI solution for personalized digital assistants.

• Creative Artists Agency (CAA) - which is the management company for many Hollywood movie stars- has partnered with Veritone to protect and manage AI likenesses for its clients. The "CAA Vault" will securely store digital assets like names, images, scans, and voice recordings, ensuring only authorized use.

Quickfire News

• Stability AI is reportedly facing a cash crunch and has been discussing a potential sale as it struggles to generate revenue and manage expenses.

• Ampere Computing and Qualcomm announced a partnership to offer a combined AI chip solution aimed at lowering power consumption in data centers.

• Researchers in the Netherlands have built an AI-driven sarcasm detector, aiming to improve human-machine communication and potentially detect abuse and hate speech.

• OpenAI cofounder John Schulman predicted that AGI could arrive in two to three years, with AI models autonomously carrying out entire coding projects in one to two years.

• Researchers from MIT and University of Basel developed a physics-informed AI to automatically map out phase diagrams for novel physical systems, potentially aiding in the discovery of unknown phases of matter.

• Hugging Face commits $10 million in free shared GPUs through its new ZeroGPU program to help developers, academics, and startups create AI apps without the burden of GPU costs.

• Google released the Gemini 1.5 model family technical report, showing improvements in math, coding, and multimodal benchmarks.

• Gemini 1.5 Pro surpasses Ultra in most text and vision tasks.

• OpenAI is adding new data analysis tools in ChatGPT, including interactive and customizable charts.

• Tools to simplify merging datasets and creating pivot tables are also being added to ChatGPT.

• Synthesia partnered with ElevenLabs to offer natural-sounding voices and personalized voice cloning for Enterprise users.

• OpenAI might be opening an office in France, with job listings in Paris and a developer event in the region this week.

• Stability AI is in talks with a tech investor syndicate, including former Facebook president Sean Parker, for a major equity investment.
  

• OpenAI filed a trademark for GPT-4o on May 16th, mentioning virtual, autonomous, semi-autonomous, and intelligent agents.
  

• Elon Musk commented on how AI will impact X’s feed algorithm, saying that the platform will soon move to a ‘fully AI-based post recommendation system’.
  

• Microsoft’s Copilot appears to be sourcing some responses via advertisements, raising questions on the ethics of how LLM responses and ads will coexist in the future.
  

• OpenAI is pausing the use of the female ‘Sky’ voice featured during the company’s recent demo, following negativity surrounding its resemblance to Scarlett Johansson and the film ‘Her’.

• Inflection announced its new leadership team and pivoted focus on bringing empathetic, emotionally intelligent AI to enterprises via an API program.

• AI pioneer Geoffrey Hinton said that universal basic income will be a necessity to address job losses and inequality caused by AI.
  

• Lmsys introduced a ‘Hard Prompts’ category in its Arena chatbot testing, aiming to help better evaluate models on more complex, problem-solving-oriented questions.

• Google shared new demos of upcoming releases, including Project Astra multimodal agents solving math equations and Veo video generation model outputs.

• Elon Musk’s xAI is progressing towards multimodal capabilities for its Grok chatbot, anticipated in the upcoming 1.5V model release.

• French AI startup H announced $220M in funding to develop advanced AGI for automating complex business tasks, with investors Bernard Arnault and Eric Schmidt.

• Europe’s AI Act passed its final hurdle after EU endorsement, bringing comprehensive rules, strict fines, and new restrictions on high-risk AI apps.

• Suno raised $125M in funding for its AI music generation platform to increase product development and expand the company.

• Google’s new data analysis feature in Gemini Advanced is now live, allowing users to upload Google Sheets, CSVs, and Excel files for data exploration.

Closing Thoughts

That’s it for us this week.

Massive news from Google and OpenAI and it’s going to be really interesting to see how this will evolve in the coming weeks and months.

Have a nice weekend. Lots to absorb.

If you want a version of this newsletter with pictures etc. Be sure to subscribe below:

https://www.blacklynxbrief.com/subscribe