Japanese AI Company Delivers Breakthrough

Good morning.

In what seems a race between the US and China, a breakthrough in AI comes to us quite unexpectedly from the land of the rising sun: Japan.

We already reported on this last week but I dove a little bit deeper on the topic this week since this new development seems to have quite a few people perking up and paying attention.

The company “Sakana” developed a tool called “AI Scientist”. The hope for ‘AGI’ or Artificial General Intelligence is that it’ll help us develop new technologies.

AI Scientist is a collection of AI ‘Agents’ that work together to accomplish a complex task. A very complex task in this case.

The AI Scientist starts by coming up with some new ideas and then checks if they’re really unique.

After that, the Scientist works on coding these ideas using advanced tools that help with writing code automatically. Next, they run tests to see how well the new ideas work, gathering numbers and visual results. Then, the Scientist writes a detailed report explaining what they found and why it matters.

Finally, the AI Scientist uses an automated system to review the work, making sure it meets high standards. This review helps improve the project and gives new ideas for future research.

It runs in an open loop so it keeps on spewing new research and new ideas. The hope is that one of these ideas will be groundbreaking.

The ideas and consequent papers that roll out of this are quite good:
Check it out here and here (and also here)

The fact that it runs in a loop is intriguing.

This will be the future of work. You tell an army of agents to accomplish a task for you. You leave the thing running and go to bed and when you wake up in the morning- it’s ready for you.

The options are intriguing.

Welcome to the Blacklynx Brief

AI News
Robots Made in China

• Walmart's CEO Doug McMillon announced that the company is using generative AI to significantly boost productivity, updating 850 million product catalog entries 100 times faster than human methods. Walmart has also integrated AI into customer search, shopping assistants, and seller support, with plans to expand AI experimentation across all business areas, highlighting the growing influence of AI in large-scale retail operations.

• SoftBank's Project Izanagi, which aims to develop AI processors to compete with Nvidia, is facing setbacks after Intel failed to meet volume and speed requirements. SoftBank is now considering partnering with TSMC, despite TSMC's current production challenges. This development underscores the intense competition in the AI chip market as companies vie to challenge Nvidia's dominance.

• Nous Research released Hermes 3, an open-source model available in three sizes, with the 405B parameter model achieving state-of-the-art performance. Hermes 3 excels in roleplaying, agentic tasks, and long-context coherence, rivaling Meta’s Llama-3.1 405B. The availability of such advanced open models increases transparency and competition in the AI landscape.

• Swiss startup FinalSpark launched a service allowing scientists to rent access to "biocomputers" made from human brain cells for $500 a month. These biocomputers, which use organoids and dopamine-based reinforcement for training AI, are claimed to be 100,000 times more energy-efficient than traditional systems. While promising for reducing AI's energy consumption, the use of brain organoids raises significant ethical concerns.

• California’s AI safety bill, SB 1047, was revised after pushback from AI companies like Anthropic and open-source developers. The revisions reduce the legal risks for AI companies by removing the ability to sue before a catastrophe and softening the requirements for safety practices. The bill is set for a final vote, reflecting California's attempt to balance AI innovation with safety through adaptive regulation.

• Researchers developed a new AI technique to find shorter solutions to Rubik’s Cubes by analyzing the puzzle's structure and identifying the best moves more efficiently. This method improves on existing Rubik’s Cube-solving AI, with potential applications in optimizing complex problems like supply chains and drug discovery.

• China-based robotics startup AGIBOT unveiled five advanced humanoid robots, including the flagship Yuanzheng A2, directly challenging Tesla’s Optimus bot. AGIBOT plans to ship 300 units by the end of 2024, emphasizing its focus on commercialization and cost control, marking an intensifying competition in humanoid robotics between the US and China.

• Victor Miller, a mayoral candidate in Wyoming, plans to use a customized ChatGPT GPT named Vic to help run local government if elected. Despite OpenAI shutting down his account twice, Miller has filed to run with the AI’s assistance, making him the first political candidate to openly advocate for AI in governance, highlighting the potential future role of AI in politics.

• Luma Labs released Dream Machine 1.5, an upgraded AI video generation model with improved text-to-video and image-to-video capabilities. The new model offers better prompt understanding, smoother motion, and higher-quality cinematography, marking a significant step forward in the rapidly advancing field of AI video generation.

• OpenAI launched free fine-tuning for GPT-4o, allowing developers to customize the model for better performance and accuracy. This offer is available through September 23, with free fine-tuning up to 1 million tokens per day. This move is expected to lead to a surge in more capable AI applications as developers take advantage of the opportunity.

• A group of authors filed a lawsuit against AI startup Anthropic, accusing the company of training its Claude chatbot on pirated copies of copyrighted books. This lawsuit highlights the ongoing legal challenges around AI training data, especially concerning the use of unauthorized materials, which remains a contentious issue in the courts.

• Microsoft released Phi-3.5-MoE, a compact AI model using a mixture-of-experts approach to rival the reasoning capabilities of much larger models. While it outperforms many popular models, it falls short against OpenAI’s GPT-4o mini. This development underscores the competition among tech giants to create more efficient AI that can run directly on mobile devices.

• Ideogram released version 2.0 of its text-to-image model, introducing five new image styles (General, Realistic, Design, 3D, Anime) and an iOS app. The Realistic style excels at rendering human features like hands and hair, and the Design style improves text rendering, making Ideogram 2.0 a new standard for high-quality image generation.

• xAI began rolling out early beta access for Grok 2, an AI model that leverages real-time data from X and uses Flux.1 to generate unfiltered AI images. Grok 2 offers a unique advantage with its ability to reference specific tweets and provide up-to-date information, positioning xAI as a serious contender in the AI chatbot space.

• Scientists from ETH Zurich and Disney Research developed an AI system that generates realistic, physics-based movements for virtual characters and robots from simple text or image inputs. The system, which adapts to physical constraints and can be applied to real robots, marks a significant advancement in making humanoid robots more natural and adaptable, potentially accelerating their integration into daily life.

Quickfire News

• Elon Musk revealed that xAI is developing an in-house image generation system: aims to replace the current Flux model in Grok 2 but is months away from release.

• Google expanded its AI-generated search summaries to six new countries: added hyperlinks and quality improvements after initial issues.

• U.S. Consumer Financial Protection Bureau highlighted AI risks in finance: emphasized that existing laws apply and innovation needs consistent regulatory treatment.

• Grammarly is reportedly rolling out a new AI content detector tool: can detect whether content was created by AI, a human, or a combination of both.

• CodeRabbit raised $16 million: aims to automate code reviews using AI, providing actionable, human-like feedback to developers.

• Apptronik, an automation company, reported it is preparing for a commercial launch of humanoid robots: expected by the end of 2025.

• Claude introduced a new screenshot capture button: allows users to easily include images from their screen in prompts.

• Midjourney released a unified web-based AI image editor: features advanced tools for seamlessly modifying and extending generated images.

• Rebellions and Sapeon, South Korean AI chip makers, signed a merger agreement: aims to challenge global leaders like Nvidia.

• Bzigo launched Iris: an AI-powered mosquito detector that tracks and marks mosquitoes with a laser pointer for easy swatting.

• Coinbase started a $15,000 accelerator grant program: targets projects combining AI with crypto wallets to enable economic participation.

• Microsoft unveiled PowerToys Workspaces: a feature to auto-arrange apps, plus an AI-powered copy-paste tool with OpenAI API integration.

• AMD acquired server maker ZT Systems for $4.9 billion: aims to strengthen its AI capabilities and compete with leaders like Nvidia.

• Berkeley Law launched a new Master of Laws program focused on AI law and governance: expected to start next summer.

• The United States invested $335 billion in AI over the past decade: triple China’s investment, with 71,000 AI job postings in 2024 alone.

• Stability AI appointed Hanno Basse as its new Chief Technology Officer: an entertainment industry veteran, to drive business growth.

• ElevenLabs released its AI-powered text-to-speech app Reader globally: supports over 30 languages and hundreds of new voices.

• Perplexity introduced code interpreter upgrades: enables library installation and chart rendering for AI-powered data visualization.

• IDC forecasted worldwide AI spending to reach $632 billion by 2028: generative AI expected to account for 32% of that.

• AI influencer Justin Fineberg and Cassidy AI raised $3.7 million: to build an intelligent automation platform for reliable AI workflows.

• CodeSignal launched Conversation Practice: an AI-powered tool for simulating workplace conversations and providing personalized feedback.

• LTX Studio opened to the public: launched five new features including character animation, dialogue, face motion capture, generation, and keyframe control.

• LVMH founder Bernard Arnault invested in five AI startups in 2024: done through his family office, Aglaé Ventures.

• Chinese scientists developed a brain-inspired AI network model: aims to address challenges like high resource consumption and interoperability.

• Nvidia unveiled advances in digital humans and avatar tech: including Nemotron-4 4B NIM, the first small AI language model for game characters.

• Midjourney opened its web-based AI image editor and new tools to everyone: offers free trials for new users to test it out.

• McAfee released AI-powered deepfake detection software: available for select Lenovo PCs to protect users from AI-generated scams.

• Best Buy introduced AI-powered delivery tracking: provides minute-by-minute updates to meet customer expectations for transparency.

• MIT CSAIL researchers developed an AI assistant: oversees teams to align roles and beliefs, aiming to improve collaboration.

• Groq launched a new API for a distilled version of OpenAI’s Whisper text-to-speech model: 240 times faster and significantly cheaper.

• California and Google drafted a $300 million, 5-year partnership: funds in-state newsrooms and AI initiatives, including a $40 million annual “AI Innovation Accelerator”.

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter : make sure to send the link below to your colleagues and friends :

https://www.blacklynxbrief.com/subscribe