Cara grew from 40k to 650k users in a week because artists are fed up with Meta’s AI policies

Cara, an anti-AI social platform founded by photographer Jingna Zhang, has rapidly grown from 40,000 to 650,000 users in just a week due to widespread dissatisfaction among artists with Meta's AI policies. Meta's use of public posts to train its generative AI systems, especially without proper opt-out options for non-European users, has led artists to flock to Cara, where they can post without fear of their work being used without consent. Zhang, a vocal advocate for artists' rights, has also been involved in legal battles against companies like Google and Stability AI for using artists' work to train AI without permission.

Cara combines features of Instagram and microblogging sites, allowing artists to showcase their portfolios and post updates. The platform collaborates with the University of Chicago’s Glaze project, adding an extra layer of protection against unauthorized AI scraping. Despite its rapid success, Cara faces significant challenges, including a recent unexpected $96,280 bill from Vercel, a web hosting company, due to the surge in users. Zhang, who has not sought venture funding, relies on a small team of volunteers to manage the platform’s growth.

Zhang’s previous experiences, from being a national athlete to managing an esports team and attending Stanford’s Ignite program, have equipped her with the skills needed to lead Cara. Despite the obstacles, Zhang remains committed to creating a space where artists can share their work without the fear of AI exploitation. As Cara navigates its newfound popularity, Zhang’s dedication to protecting artists' rights continues to resonate with a growing community of creatives.

Eric Schmidt Is Secretly Testing AI Military Drones In A Wealthy Silicon Valley Suburb

Last year, billionaire Eric Schmidt, former Google CEO, secretly founded White Stork, a military drone company now testing AI-guided aircraft in Menlo Park and Ukraine. The venture, renamed Project Eagle, involves testing drones with Ukraine's Ministry of Defense, aiming to enhance battlefield targeting through artificial intelligence. Schmidt, who has a history of government security clearances, has visited Ukraine’s frontlines and is recognized by Ukrainian officials for his support and technological contributions.

White Stork has attracted top talent from Apple, SpaceX, and Google, along with experts in machine learning, aerospace, and supply chain management. Notable hires include Hendrik Dahlkamp, an AI expert formerly at Apple, and Damon Vander Lind, a senior advisor from Kitty Hawk. The project also involves partnerships with Silicon Valley luminaries like Sebastian Thrun and has expanded recruitment through universities and AI hackathons.

Despite Schmidt's public visits to Kyiv, the venture operates discreetly through a web of LLCs and associated companies like Merops and Volya Robotics. Schmidt has written op-eds on modern warfare, highlighting the transformative impact of AI-empowered kamikaze drones. His predictions are tempered by the harsh realities of war, emphasizing the psychological toll on both drone operators and ground troops, referencing Nietzsche’s philosophy about the abyss staring back.

GPT-4 can now hack autonomously

Agents based on large language models (LLMs) are increasingly capable of solving complex tasks, including exploiting real-world vulnerabilities. Traditional agents struggle with zero-day vulnerabilities where the exploit is unknown. However, our new work demonstrates that teams of AI agents can autonomously exploit these zero-day vulnerabilities using a hierarchical planning and task-specific agent (HPTSA) approach. This technique involves an exploration and planning agent, a team manager agent, and task-specific expert agents. In testing, HPTSA successfully hacked over half of the vulnerabilities in our benchmark, outperforming open-source scanners and previous models.

HPTSA's architecture addresses limitations in single-agent performance by enabling specialized focus on different types of vulnerabilities. The exploration/planning agent maps out potential exploits, which the team manager then allocates to task-specific agents. These agents are designed with prompts and access to relevant documents to concentrate on specific vulnerability types, such as XSS or CSRF. This structured approach allows for efficient and targeted exploitation, significantly improving success rates compared to traditional methods and single-agent setups.

Our benchmark included 15 real-world, web-based vulnerabilities published after GPT-4's knowledge cutoff date. The HPTSA approach demonstrated its capability by autonomously exploiting 8 out of these 15 vulnerabilities. Case studies showed the agents' ability to synthesize information across different execution traces, enhancing their problem-solving efficiency. These findings highlight the potential of multi-agent systems in cybersecurity and underscore the need for comprehensive safety evaluations beyond chatbot settings to fully understand LLM capabilities.

‘This Is Going to Be Painful’: How a Bold A.I. Device Flopped

Humane, a San Francisco-based AI hardware company, faced a significant setback with its $699 Ai Pin, a wearable AI device meant to replace smartphones. Despite a year of marketing, the product received poor reviews for issues like battery life, overheating, and design flaws, leading to only 10,000 orders instead of the hoped-for 100,000. The company is now in talks with HP and other potential buyers and has retained Tidal Partners to navigate these discussions while managing a new funding round.

Founders Bethany Bongiorno and Imran Chaudhri, both former Apple employees, aimed to revolutionize tech with a device that could send messages, search the web, and take photos, with a laser that projects text onto the user's palm. However, employees reported that the device's laser display consumed too much power, causing overheating, and that the battery life was insufficient. Humane attempted to mitigate these issues by providing a backup battery and charging case, which further increased the device's cost.

As sales lagged and technical issues persisted, Humane experienced employee departures and lay-offs. Critics cited a lack of readiness and marketing strategy, with a senior software engineer dismissed for raising concerns about the product's readiness. Following scathing reviews in major publications, Humane acknowledged the challenges and began working on updates to improve the device, including better voice navigation and integration of OpenAI’s newest chatbot system to enhance battery life and performance.

Despite the setbacks, Humane has seen some interest from businesses and wireless carriers in South Korea and Japan. The founders view the critical feedback as an opportunity to improve the Ai Pin. They continue to explore funding options to support their ambitions, while also dealing with a recent battery supplier issue that led to a recall of the charging case accessory.

Related: How to build a DOA product: Humane AI Pin founders banned internal criticism

Other stuff

• Sam Altman Was Bending the World to His Will Long Before OpenAI

• AI in software engineering at Google: Progress and the path ahead

• OpenAI wants to buy 'vast quantities' of nuclear fusion energy from Helion

• Can a technology called RAG keep AI models from making stuff up?

• Hello Qwen2

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.