In daily AI news, we saw that AI therapists gave dangerously inappropriate advice, and lawmakers are proposing a bill to ban AI price-gouging. On a more positive note, Microsoft is preparing Copilot for GPT-5 with a new "Smart" mode, and Google introduced Opal for building AI mini-apps, along with a new deepfake detector called UNITE.
Our Video of the Day with Sundar Pichai explored the transformative power and risks of AI, and the growing importance of a generalist mindset in an AI-dominated world.
For the Tool of the Day, we introduced GitHub Spark, a new feature for Copilot Pro+ subscribers that enables the rapid creation and deployment of intelligent applications using natural language.
🗞️ Today's Top AI Stories:
AI Therapist Goes Off the Rails
A video journalist revealed how AI chatbots provided dangerously inappropriate responses when tested with a simulated mental health crisis. The bots encouraged self-harm and suggested a romantic relationship, highlighting the premature and risky deployment of this technology in mental health care. A Stanford study confirmed these findings, showing that therapy chatbots often demonstrate a stigma toward psychiatric conditions and can encourage harmful behavior.
Delta’s AI spying to “jack up” prices must be banned, lawmakers say
Democratic lawmakers introduced the "Stop AI Price Gouging and Wage Fixing Act" to ban companies from using "surveillance-based" price or wage setting to increase profit margins. The bill, if passed, would allow individuals to sue companies for charging higher prices based on a customer's desperation or paying lower wages based on personal data. Delta, among other companies, is currently testing AI pricing, which it claims forecasts demand and adapts to market conditions, but critics worry it's a form of individualized pricing.
Copilot Prepares for GPT-5 with New "Smart" Mode
Microsoft is preparing its Copilot AI for the upcoming GPT-5 release, now anticipated in early August. Code changes point to a new "Smart" mode that will join existing options like Quick Response and Deep Research. This new mode is designed to automatically adjust its response strategy, deciding whether to give a quick answer or engage in complex reasoning based on user input, removing the need for manual mode switching.
Google Introduces Opal to Build AI Mini-Apps
Google Labs introduced Opal, a new experimental tool that empowers users to build and share powerful AI mini-apps using simple natural language and a visual editor. Available in a US-only public beta, Opal is designed to accelerate prototyping and boost productivity by allowing users to chain together prompts, models, and tools. With Opal, you can describe the logic, and it will build a visual workflow for you.
Google and UC Riverside Create New Deepfake Detector
UNITE is a new AI tool, developed by researchers at the University of California, Riverside, and Google, that can identify fake videos. Unlike older systems that primarily looked at faces, UNITE analyzes backgrounds, motion, and other subtle cues, making it capable of detecting a wider range of forgeries, including videos where no faces are visible. This "universal" tool could be essential for platforms like newsrooms and social media trying to combat the spread of disinformation.
🎥Video of the day:
Lex Fridman on AI, The Universe, and Everything: Sundar Pichai
Technology as a transformative force [01:05]
This takeaway is key because it provides insight into Sundar Pichai's foundational belief in technology's power to change lives, which informs his approach to AI development and its potential societal impact.
The potential and risks of AI [15:39]
It is crucial to acknowledge both the potential for massive progress and the significant risks involved with AI. This takeaway highlights a balanced perspective on AI that is essential for anyone engaged with its development and deployment.
The importance of a generalist mindset [02:09:21]
This challenges the traditional value of specialization and suggests a new way of thinking about human roles in an AI-dominated future. It's a key takeaway because it redefines the skills needed to stay relevant in the AI race.
🛠️ Tool of the day:
GitHub Spark is a new feature for building and deploying full-stack intelligent applications in minutes using natural language. It’s a game-changer for rapid prototyping, as it removes the need for extensive setup and provides powerful AI capabilities seamlessly integrated with the GitHub platform.
Strength:
Rapid Development: Build and deploy full-stack intelligent applications quickly using natural language.
No-setup Convenience: Includes data, LLM inference, hosting, and deployments out-of-box.
AI Integration: Use LLMs from various providers without managing API keys.
Seamless GitHub Integration: Iterates with natural language or code and integrates with GitHub repositories, Actions, and Dependabot.
Limitations:
Availability: Exclusively available for GitHub Copilot Pro+ subscribers.
LLM Dependence: The core functionality relies on LLMs, which may have their own limitations.
Pricing: Access to GitHub Spark is included with the Copilot Pro+ plan.
Copilot Pro+ ($39/month or $390/year): Includes everything in the Pro plan, plus access to all models, more premium requests, and GitHub Spark features like Spark messages (up to 375 per month), 10 active app building sessions at a time, and unlimited apps.