Good morning. It’s Friday, June 28th.

Did you know: On this day in 2011, Google+ was introduced as an invite-only beta.

In today’s email:

• OpenAI Restricts Access to Adversarial Countries

• Rabbit R1 Security Breach

• Zuck: Closed Source AI Trying To Create God

• AI Avatar Video Calls

• 10 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Thank you for being an ad-free subscriber!

OpenAI To Restrict Access To China, Russia, North Korea

Starting July 9th, OpenAI will impose stricter API restrictions on unsupported countries, notably China, Russia, North Korea, and Iran. This action follows a warning email to developers about API usage from these regions. Affected developers face challenges in identifying traffic from unsupported areas, as OpenAI does not provide detailed data on API key usage by region.

Recent findings by OpenAI highlight the misuse of its AI models by state-backed actors from Russia, China, Iran, and Israel for propaganda and disinformation campaigns, though these efforts have shown minimal impact due to frequent human errors exposing AI-generated content. Previously, OpenAI and Microsoft identified and terminated accounts of state actors exploiting AI for malicious activities. The new API restrictions aim to curb such misuse, especially significant in an election year. Read more.

Jailbreakers Find Massive Security Flaw In Rabbit R1’s Code

Rabbit's R1 AI gadget has been implicated in a security breach uncovered by a group of developers, Rabbitude. They found hardcoded API keys in Rabbit's software, potentially exposing sensitive data to unauthorized access. These keys provided entry to Rabbit's accounts with services like ElevenLabs and SendGrid, used for email communications.

According to Rabbitude, this flaw allowed access to all responses from R1 devices. Despite being aware of the issue for over a month, Rabbit reportedly delayed securing the exposed data adequately. While some API keys have been revoked, access to the SendGrid key remained open as of the latest update. Rabbit responded by acknowledging the breach on its website and Discord channel, stating ongoing investigations into the safety of customer data. Read more.

Zuckerberg Disses Closed-Source AI Competitors As Trying To 'Create God'

Meta CEO Mark Zuckerberg recently took a not-so-subtle jab at competitors in the AI field for their closed-source approaches, likening their ambitions to "creating God." In a recent interview, Zuckerberg stressed the importance of open-source AI, arguing against the concept of a single, dominating AI entity. He emphasized the need for diverse AI developments that cater to various human interests, contrasting Meta's approach with those focused on monopolistic control.

This push for openness reflects Meta's overall strategy of building an open AI ecosystem where a multitude of AI solutions can flourish, as opposed to walled gardens. This philosophy aligns perfectly with their recent launch of AI Studio, a tool that lets creators whip up custom AI avatars for interactive messaging on Instagram. Read more.

Character.AI Allows Users To Talk With AI Avatars Over Calls

Character.AI has introduced a new feature allowing users to engage in calls with AI avatars, supporting multiple languages including English, Spanish, Portuguese, Russian, Korean, Japanese, and Chinese. Before its public launch, the feature underwent testing and has already facilitated over 20 million calls by 3 million users. It enables language practice, mock interviews, and integration into role-playing games. Users can easily initiate calls with user-generated AI characters with a single tap, ensuring smooth interactions with reduced latency.

The platform also offers seamless switching between calling and texting and includes an option to interrupt AI responses. Earlier, Character.AI launched Character Voices for one-on-one chats, featuring over 1 million user-created voices. This initiative coincides with user feedback noting changes in character behavior, attributed by the company to testing phases. Read more. \

Etcetera: Stories you may have missed

Figma announces big redesign with AI

SoftBank to invest in search startup Perplexity AI at $3 billion valuation

Hollywood workers union reaches pay, AI-use deal with top studios

China's Killer Robots Are Coming

Intel shows off first fully integrated optical compute interconnect, designed to scale up AI workloads

Persona’s founders are certain the world can use another humanoid robot

New tool detects AI-generated videos with 93.7% accuracy

Israel to build supercomputer to keep pace in global AI race

Anthropic Expands Claude AI Access for Government Agencies with AWS Partnership

Meta's new LLM Compiler could transform the way software is compiled and optimized

OpenAI is training a new model CriticGPT, to catch bugs in GPT-4’s code and to reduce hallucinations

OpenAI's ChatGPT and Microsoft's Copilot repeated a false claim about the presidential debate

OpenAI, Microsoft sued by Center for Investigative Reporting as news industry bolsters attack on AI  

Gemini 1.5 Pro 2M context window, code execution capabilities, and Gemma 2 are available today

Google Translate gets 110 new languages with AI's help, bringing the total to 243

Google's new cloud features aim to make GenAI more reliable and up-to-date

Google's Vertex AI to add Mistral Small, Large and Codestral models

Amazon Is Investigating Perplexity Over Claims of Scraping Abuse

Thisorthis.ai enables side-by-side AI model comparisons, facilitating informed decisions with shared results, community voting, and prompt analysis.

Aire quickly builds customizable business web apps from AI prompts, generating data models, workflows, user interfaces, and charts.

Dappier 2.0 is an online marketplace that protects and monetizes AI-accessed content, ensuring fair compensation for content creators.

ToucanFX, created by Leo Nagano, is an AI-powered platform for generating custom sound effects. It utilizes ElevenLabs' sound effects API.

Skyvern is an open source AI agent that automates browser-based workflows, replacing fragile scripts with a simple API for diverse tasks.

Question Base uses AI to automate responses to repetitive questions in Slack, capturing and organizing knowledge shared in conversations.

GPUDeploy is a marketplace offering low-cost, on-demand GPU rentals from reliable providers at wholesale prices, simplifying access to compute resources.

 AiTerm simplifies terminal usage by converting natural language into executable commands, providing real-time AI assistance and command suggestions.

Inductor Custom Playgrounds auto-generate shareable playgrounds for LLM apps using a single CLI command, facilitating collaborative development and rapid iteration.

Proxlight Designer 3 accelerates GUI development in Python using AI for code generation and Figma for designing Tkinter interfaces.

📄 Octo-planner: On-device Language Model for Planner-Action Agents

📄 CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs  ​​

📄 APIGen: Automated Pipeline for Generating Verifiable and Diverse Function-Calling Datasets ​​

📄 Segment Any Text: A Universal Approach for Robust, Efficient and Adaptable Sentence Segmentation ​​

📄 Aligning Diffusion Models with Noise-Conditioned Perception

📄 Simulating Classroom Education with LLM-Empowered Agents  

📄 Adam-mini: Use Fewer Learning Rates To Gain More

📄 OMG-LLaVA: Bridging Image-level, Object-level, Pixel-level Reasoning and Understanding

📄 Simulating Classroom Education with LLM-Empowered Agents

📄 SeaKR: Self-aware Knowledge Retrieval for Adaptive Retrieval Augmented Generation

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.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email