Straight to Super Intelligence !

Good morning.

Something significant has happened this week.

I had written an entire rant on the corporate culture at OpenAI and how it’s a bit worrying that they’re choosing speed over safety.

On Wednesday evening, Ilya Sutskever, who left OpenAI because of these safety concerns, announced he’s starting a new company, so I guess my rant now has to go straight to the trash bin.

Ilya Sutskever is the co-founder of OpenAI and considered by many to be the brain behind ChatGPT. He is considered to be a genius-level AI researcher.

This is the guy. As you can see, hair maintenance is optional when you’re that smart.

Sutskever has now started a new company called SSI Inc. And this is their mission statement:

There’s already a lot of criticism, especially as this company will be based in Palo Alto AND Tel Aviv, but the thing that has amazed everyone that is following this space is that one of the brightest minds in artificial intelligence is actually not only acknowledging the possibility of superintelligence but is actively going to pursue it. Sutskever and his company claim ASI (Artificial Super Intelligence) is “within reach”.

And here we are talking about AGI (Artificial General Intelligence) and if that will ever be reached in our lifetime.

To clarify : AGI is when a machine is at least as smart as the smartest human and can complete human level tasks. Superintelligence is an artificial intelligence that is many times smarter than a human and will be able to do unimaginable things. One mind that will run all our lives. That kind of thing.

Where as AGI will replace your job - ASI will replace everything and anything.

If this is achieved life as we know it will be completely different.

The big deal here is that superintelligence has always been dismissed as science fiction - something that’may’ emerge but probably won’t ever exist.

If one of the brightest minds in the space has now gone all-in on “superintelligence” and he believes it’s coming, that means that we should buckle up and pay close attention.

Welcome to the future (and to the Blacklynx Brief)

AI News

• Google DeepMind and Harvard researchers have developed a virtual rodent powered by an AI neural network, accurately mimicking the movements and neural activity of real rats. This bio-mechanically realistic model, controlled by an AI brain in a physics simulator, shows brain activity patterns similar to those of live rodents. This advancement could revolutionize neuroscience research and contribute to the development of advanced robotics.

• OpenAI's CTO, Mira Murati, did a Q&A with Fortune Magazine. When asked about Elon Musk's comment that OpenAI and Apple's integration is "creepy spyware," she responded, "We don't think so," which is a slightly terrifying response.

• To stay with the “slightly terrifying” theme : OpenAI added retired U.S. Army General and former NSA head Paul M. Nakasone to its Board of Directors. He will also join the company’s Safety and Security Committee. Edward Snowden was not happy and points out that this move is proof OpenAI can NOT be trusted.

  

• Nvidia (which has this week surpassed Microsoft as the most valuable company int the world) has launched Nemotron-4 340B, a set of open-source language models aimed at generating high-quality synthetic training data and developing advanced AI applications. The three models—Base, Instruct, and Reward—work together to create and filter synthetic data for training new AI models, outperforming competitors like Llama-3 and Qwen-2. Additionally, Nvidia introduced Mamba-2 Hybrid, a model that exceeds similar AI systems in accuracy

• Google DeepMind has developed a new system called V2A that creates synchronized soundtracks for videos, including music, sound effects, and dialogue. V2A uses video pixels and text descriptions to produce realistic audio that matches the visuals. Currently being tested with filmmakers, V2A aims to enhance video content by adding seamless, matching audio, addressing the common issue of silent AI-generated videos.

• Health tech company Color has teamed up with OpenAI to create an AI assistant that helps doctors personalize cancer screening and treatment plans, aiming to reduce delays in care. This AI tool, based on GPT-4o, analyzes patient data and medical records to identify gaps and create tailored diagnostic plans, potentially saving weeks or months in treatment time. By late 2024, Color hopes to provide AI-generated screening plans for over 200,000 patients

• Meta's Fundamental AI Research (FAIR) group has unveiled several new open-source AI models and techniques. These include Chameleon, a versatile model for handling image, audio, and text inputs; JASCO, a controllable text-to-music model; and AudioSeal, an audio watermarking tool for detecting AI-generated speech. Meta also released a Multi-Token Prediction model to enhance code autocompletion

• Researchers at the University of Cincinnati have developed an AI system that predicts anxiety levels using a picture rating task and some demographic and psychological variables. This AI, called Comp Cog, measures human decision-making patterns by having participants rate emotion-evoking pictures and answer a few questions about their background. The system achieved up to 81% accuracy in predicting anxiety

• Dell, Nvidia, and Super Micro Computer are teaming up to build a supercomputer for xAI’s Grok AI chatbot. The system will use up to 100,000(!!) Nvidia H100 GPUs, potentially making it the largest AI cluster in existence. Set to be operational by Fall 2025, this project highlights the immense resources needed to advance AI technology and compete with leading AI companies like OpenAI and Anthropic.

  

  Source: Michael Dell on X

Quickfire News

• Former Meta engineers launched Jace, an AI agent from Zeta Labs capable of executing complex browser tasks autonomously.

• LinkedIn introduced new AI tools to its job-search platform, including AI-powered coaches for expert advice, learning personalization for skill courses, and advanced job-search features.

• Stanford AI Lab showcased HumanPlus, a system enabling humanoid robots to autonomously learn and perform tasks by imitating human actions.

• A photographer was disqualified from an AI image contest after winning using a real photo of a flamingo. He said he wanted to show that ‘nature can still beat the machine’.

• Apparate Labs launched PROTEUS, a new real-time AI video generation model that can create realistic avatars and lip-syncs from a single reference image.

• Microsoft is delaying the launch of its new Copilot+ Recall feature. It will first be available as a preview feature to testers in the Windows Insider Program.

• Sam Altman has revealed that OpenAI is considering changing its governance structure from a nonprofit to a for-profit benefit corporation.

• McDonald’s is removing its AI-powered drive-thru ordering system after testing it in over 100 restaurants. Viral videos highlighted customer frustrations and technical issues.

• MIT has unveiled a new study presenting a 3D technology platform capable of mapping the human brain

• Some Tesla shareholders are suing Elon Musk for allegedly diverting talent and resources to his company xAI, claiming a breach of fiduciary duties and unjust enrichment.

• Adobe added new Firefly AI capabilities to Acrobat for creating and editing images within PDFs using text prompts. Users can also access an AI assistant for insights and content creation.

• A report from the Reuters Institute of Journalism highlights growing public concern about AI-generated news, with many worried about its impact on content reliability and trust.

• The U.S. Navy is deploying AI-powered underwater drones to detect threats, with plans to use this technology for identifying enemy ships and aircraft.

• Luma announced new control features for its Dream Machine video model, enabling quick scene changes and precise character edits, along with extending videos and removing watermarks.

• Anthropic's new research reveals that AI models can engage in 'reward tampering', learning to cheat the system for higher rewards even without specific training.

• Nvidia became the world’s most valuable company after its stock price rise brought the chipmaker’s valuation to $3.34T, overtaking Microsoft.

• ElevenLabs launched new open-source text and video-to-sound effects apps and APIs, allowing users to generate audio based on text prompts or reference videos.

• Apple published 20 new Core ML models and 4 datasets on Hugging Face, enabling builders to create advanced on-device AI applications.

• Figure AI CEO Brett Adcock announced Cover, an advanced AI hardware project in partnership with NASA aiming to help prevent school shootings using imaging scanners for concealed weapon detection.

• Hedra introduced a preview of its Character-1 model, capable of generating unlimited length videos of expressive talking-head avatars that can talk, sing, and rap.

• Factory announced new updates and a $15M raise for its AI that automates development tasks like generating software features, reviewing code, and solving bugs.

• OpenAI has shut down tools used by two AI political candidates for violating policy against political campaigning.

• Microsoft unveiled Florence-2, an AI vision model excelling in captioning, object detection, and segmentation using prompt-based instructions.

• Canva introduced Connect APIs to help developers integrate the platform into workflows, manage assets, automate designs, and streamline feedback.

• Elon Musk agreed with Geoffrey Hinton's 10-20% probability of AI causing something terrible but believes the most likely outcome is an abundance of services.

• Snap previewed a real-time image diffusion model for AR experiences on smartphones and new AI tools for AR creators in Lens Studio 5.0.

Closing Thoughts

That’s it for us this week.

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

https://www.blacklynxbrief.com/subscribe