• AI Breakfast
  • Posts
  • The Computer Chip With Built-In Brain Tissue

The Computer Chip With Built-In Brain Tissue

...and the military funding behind it

Author

AI Breakfast
July 24, 2023

Good morning. It’s Monday, June 24th.

Did you know: 10 years ago today, Google introduced the first Chromecast.

In today’s email:

  • Merging AI with the brain

  • Twitter rebrands as "X," focuses on payments, commerce

  • Perplexity AI adds image search, "Llama Chat" chatbot

  • OpenAI refines ChatGPT with custom instructions

  • Leading AI firms commit to enhancing security

  • Benesse launches AI service for kids' research

  • AI camera DreamGenerator transforms images

  • 5 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think of this edition by replying to this email, or DM us on Twitter.

Today’s edition is brought to you by:

MinervaAI: Empowering AML Compliance with AI-driven Solutions

Streamline and automate regulatory obligations, KYC, EDD, and ongoing monitoring, with auditable reports and reduced manual work.

Trusted by leading companies, MinervaAI offers a cloud-based risk assessment platform for real-time financial crime analysis and decreased risk.

Today’s trending AI news stories

Computer chip with built-in human brain tissue gets military funding: A team of scientists from Monash University and Cortical Labs has been granted $600,000 by Australia’s National Intelligence and Security Discovery Research Grants program for their research on merging human brain cells with AI. The team aims to create “DishBrain” computer chips capable of learning throughout lifetimes, potentially surpassing current silicon-based hardware. The technology could revolutionize machine learning for applications like self-driving cars, drones, and robots.

'X is the future state of unlimited interactivity,' says Twitter CEO Linda Yaccarino: The recently appointed CEO announced Twitter plans to undergo a significant transformation of the social media platform. They revealed their decision to abandon the iconic bird logo, rebranding the platform as “X,” and focusing on entering the realms of payments, banking, and commerce with urgency. The envisioned “X” app has the potential to serve as both a social media platform and incorporate features like messaging and mobile payments.

Perplexity AI Revolutionizes Chatbots with Image Search & Meta's Llama 2 Experiment: Perplexity AI, a pioneering generative AI chatbot startup, has introduced many new features, redefining the user experience. Among the major updates is the image search feature, allowing the AI to respond to queries with visual content, enriching interactions. Moreover, Perplexity has rolled out a novel chatbot, “Llama Chat,” based on Meta’s Llama 2 AI model, further enhancing the platform’s capabilities. With a focus on personalization through AI Profile creation, experimentation, and opportunities in Perplexity Labs, the company remains dedicated to offering advanced and interactive AI solutions.

OpenAI’s Trust and Safety Head Steps Down as Devs Pledge to Spend More Time Fixing ChatGPT: OpenAI’s trust and safety head, Dave Wilner is stepping down as developers pledge to spend more time fine-tuning ChatGPT. OpenAI announced that Plus users can now add “custom instructions” to the chatbot, seeking a more consistent response. However, the company faces challenges as its trust and safety head amid increased regulatory scrutiny and given the Federal Trade Commission’s demand for AI safeguards. OpenAI is extending support for GPT-3.5 and GPT-4 models, addressing concerns raised in a recent study.

Marc Andreessen says his A.I. policy conversations in D.C. 'go very differently' once China is brought up: Marc Andreessen, a Silicon Valley venture capitalist, finds that discussions with policymakers in Washington, D.C., about AI take a different turn when China is mentioned. While the first conversation revolves around the American government’s concerns about tech companies, the second conversation shifts focus to the importance of American AI success and the need to compete with China. Andreessen highlights China’s vision of using AI for population control and exporting such capabilities to other nations, prompting a broader consideration of global implications.

Anthropic CEO says jailbreaking AI systems could become a matter of "life and death": Anthropic, a competitor to OpenAI, has unveiled its new chatbot, Claude 2, which is on par with ChatGPT but with greater safety precautions. CEO Dario Amodei raises concerns about “jailbreaks,” wherein AI models generate content not intended by developers or against the law. While the current results may seem trivial, Amodei fears that the increasing dangers of AI could lead to serious dangers in science, engineering, and biology. Unlike other AI companies, Anthropic relies on fixed rules and AI evaluation instead of human feedback to ensure safety and ethical compliance.

Leading AI firms volunteer security commitments to Biden administration: Leading AI companies, including Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI, voluntarily committed to enhancing AI security during a meeting with President Joe Biden. The commitments include building AI systems with security as a top priority, developing watermarking systems to identify AI-generated content, and sharing best practices to prevent misuse and protect national security. Cybersecurity professionals underscore the need for ongoing monitoring and vulnerability assessments to address AI security risks,

Benesse to launch AI service to help kids with research projects: Benesse, a prominent Japanese education services company, is set to release an AI-powered service to aid elementary school students with their research projects. Leveraging generative AI, the service will be accessible for free on the company’s website for parents. By offering valuable suggestions and tips, the AI aims to assist students in identifying research themes and compiling their findings. Notably, the AI refrains from providing exact answers, fostering critical thinking and independent learning among young scholars.

‘Judeo-Christian’ roots will ensure U.S. military AI is used ethically, general says: Lt. Gen. Richard G. Moore Jr. of the U.S. military asserted that the country’s approach to using AI in warfare is more ethical due to its “Judeo-Christian society.” The general emphasized that the U.S. will be cautious in deploying AI technology, aligning it with the nation’s values. Experts, however, argue that AI ethics transcend religious perspectives and encompass broader concerns for human welfare and justice.

DreamGenerator is an AI camera with integrated Stable Diffusion prompts: DreamGenerator, an AI camera, utilizes Stable Diffusion prompts to transform captured images into new themes like sky, medieval, or space while preserving essential elements. Developed by Kyle Goodrich, the camera aims to simplify the complex prompting process in AI systems, offering unique images. It combines the open-source image AI Stable Diffusion with ControlNet, a fine-tuning method, enhancing image-to-image capabilities. The camera’s design and availability details are yet to be disclosed.

🎧 Did you know AI Breakfast has a podcast read by a human? Join AI Breakfast team member Luke (an actual AI researcher!) as he breaks down the week’s AI news, tools, and research: Listen here

5 new AI-powered tools from around the web

Cloozo is the ultimate no-code chatbot builder, powered by OpenAI. It redefines client engagement, offering unique bots with individual OpenAI and Pinecone keys. Train your chatbots on your documents for targeted responses, easily embed with JavaScript, and enjoy 50% off lifetime access.

BiozumAI enables businesses to create an AI assistant in seconds. With a personal touch, it guides visitors through the pre-purchase, purchase, and post-purchase stages on websites, apps, and social commerce platforms.

ChatGPT Master by Onzo is a comprehensive bundle of 4800+ prompts ideal for data analysts, engineers, scientists, and statisticians of all skill levels. Covering topics like Python, Data, & Analytics, Data Science, and more, it serves as a co-pilot for learning data-related skills.

BenchLLM is a robust evaluation tool for LLM-powered apps, designed by AI engineers for AI engineers. Test your code on the fly, build test suites, and generate quality reports using automated, interactive, or custom evaluation strategies. Supports OpenAI, Langchain, and other APIs, making evaluations easy and flexible.

SpeakPerfect is an AI-powered tool for creating high-quality audio content. Users can easily transform their ramblings or uploaded recordings into polished audio pieces. With simple and efficient processing, getting great audio content in just one shot is possible. Get optimal results with at least 20 seconds of audio input.

arXiv is a free online library where scientists share their research papers before they are published. Here are the top AI papers for today.

In this research paper, the persistent challenge of divergence artifacts in diffusion models during slow sampling takes center stage. To combat this vexing issue, two innovative techniques step into the limelight - the Heavy Ball (HB) momentum and the Generalized Heavy Ball (GHVB). Through their clever integration into numerical methods, the researchers expand the stability regions, effectively banishing artifacts and elevating image quality to new heights. Surpassing conventional diffusion solders, these cutting-edge methods usher in a realm of efficient and artifact-free low-step sampling. The study also delves into related approaches for bolstering diffusion models’ sampling speed, including model distillation, and high-order numerical methods.

The research paper introduces “Box Diff,” a novel approach to address the challenge of controlling object synthesis in text-to-image synthesis. Unlike traditional methods, BoxDiff seamlessly integrates spatial constraints into the denoising step of Stable Diffusion models, eliminating the need for additional training and paired layout-image data. This training-free approach allows for precise control over object and context synthesis, producing high-fidelity and diverse concept coverage. With BoxDiff, interactive cooperation with AI in image content creation becomes more feasible, promising exciting new possibilities in the field of image synthesis.

The research paper “CopyRNeRF: Protecting the CopyRight of Neural Radiance Fields” introduces a pioneering approach to safeguard the intellectual property of Neural Radiance Fields (NeRF). NeRF, a powerful representation of digital media, requires copyright protection due to the complexity of its training process. The paper proposes an innovative method that embeds copyright messages into watermarked color representations, ensuring both invisibility and robustness during rendering. By incorporating distortion-resistant rendering and a message extraction mechanism, the proposed approach achieves high-quality rendering while safeguarding the copyright of NeRF models. The results demonstrate its effectiveness in protecting the core models from unauthorized use.

The researchers introduce “Subject-Diffusion,” a revolutionary approach to personalized image generation without test-time fine-tuning. The researchers leverage a vast dataset comprising 76 million images, subject detection bounding boxes, segmentation masks, and text descriptions. With a unique unified framework combining text and image semantics, “Subject-Diffusion” exhibits exceptional performance in generating high-fidelity customized images from a single reference image. The model surpasses existing methods in both single and multiple-subject generation, promising significant advancements in open-domain personalized image synthesis.

FaceCLIPNeRF is a text-driven 3D face manipulation method using Deformable Neural Radiance Fields (NeRF). It allows detailed facial expression editing with just one text input, eliminating the need for manual annotations and attribute search. The approach uses a scene manipulator to control facial deformation with a latent code and introduces the Position-conditional Anchor Compositor (PAC) for spatially varying latent codes. This method achieves accurate reflection of visual attributes from descriptive and emotional texts while preserving the face’s identity and visual quality. Extensive experiments demonstrate its effectiveness in manipulating faces reconstructed with NeRF.

Thank you for reading today’s edition.

Your feedback is valuable.


Respond to this email and tell us how you think we could add more value to this newsletter.

Attending Ai4 this year? We will be!