Read Online | Sign Up | Advertise | | | Good morning, AI enthusiasts. The AI automation conversation just got uncomfortably direct — with Epoch co-founder Tamay Besiroglu launching Mechanize, a startup with the explicit goal of replacing human workers entirely. | But with job losses already mounting and user trust in AI still lacking, getting the masses on board with the utopian dream of abundance from mass automation may be tougher than expected. | | In today's AI rundown: | AI startup to automate the entire workforce Cursor AI's hallucinated policy sparks cancellations Create full-stack web apps without coding DeepMind's shift to 'experiential' AI learning 4 new AI tools & 4 job opportunities
|
|
| | | | MECHANIZE AI | |  | Image source: GPT-4o / The Rundown |
| The Rundown: Epoch co-founder Tamay Besiroglu just launched Mechanize, a new startup developing virtual environments and training data to enable AI agents that can replace human workers for the "full automation of all work". | The details: | The company plans to create simulations of workplace scenarios to train AI agents in handling complex, long-term tasks currently performed by humans. Mechanize will initially focus on automating white-collar jobs, with systems that can manage computer tasks, handle interruptions, and coordinate with others. Backed by tech leaders including Jeff Dean and Nat Friedman, the startup estimates its potential market at $60T globally. The announcement drew criticism for both the economic implications and potential conflicts with Besiroglu's role at AI research firm Epoch.
| Why it matters: Besiroglu and co. likely aren't the only researchers that think AI is set to automate every aspect of work — but with tensions already high over both negative views of AI and mounting job losses, this goal might be saying the quiet part a bit too loudly. The age of automation is coming, and not everyone will be happy about it. |
|
| | TOGETHER WITH SALESFORCE | | | The Rundown: AI transformation is no longer optional — and companies need unified data to drive informed action across the enterprise. Data Cloud is the intelligent activation layer of the Salesforce Platform, bridging data silos and harmonizing info from various sources into a single customer view. | Data Cloud grounds AI agents built with Agentforce enabling: | More relevant, accurate output and actions Automation through data-triggered workflows Analytics in the workflow AI-powered applications
| Learn more and explore how Data Cloud powers AI success. |
|
| | CURSOR | |  | Image source: Reve / The Rundown |
| The Rundown: Agentic coding platform Cursor faced backlash after its AI support agent, Sam, hallucinated a fake policy that caused user outrage and subscription cancellations. | The details: | A Reddit user experienced unexpected logouts when switching between devices, leading to a support inquiry answered by an AI agent. The AI hallucinated a policy claiming single-device restrictions were an intentional security feature, with the post sparking backlash and cancellations. Cursor's co-founder acknowledged the error, explaining a security update caused login issues, with the policy completely fabricated by the AI. He added that the company is implementing clear AI labeling for support responses going forward and refunding the affected users.
| Why it matters: The hype surrounding AI agents has never been stronger, but cautionary tales like this one show that hallucinations are still a major issue to consider when deploying customer-facing bots. Despite companies rushing to automate customer service, it may still be too early in the AI boom for complete automation. |
|
| | AI TRAINING | | | The Rundown: In this tutorial, you will learn how to use Google's new Firebase Studio to build and deploy complex web applications — without writing a single line of code —through AI-powered prototyping. | Step-by-step: | Visit Firebase Studio and log in with your Google account. Describe your application in detail in the "Prototype an app with AI" section. Review and customize the AI-generated app blueprint (name, features, colors). Test your prototype, make adjustments if needed, and click "Publish" to deploy.
| Pro tip: Upload sketches or images of your app design to help the AI better understand your vision. Also, for more advanced customization, click the "Switch to code" button in the top right corner. |
|
| | PRESENTED BY TAVUS | | | The Rundown: A Tavus avatar appeared in a NY courtroom, sparking a national debate — and showcasing a groundbreaking future where AI doesn't just speak, but visually represents and communicates effectively at scale. | With Tavus, you can: | Build real-time video agents and generate realistic videos instantly via simple APIs Create agents with 30+ languages, natural expressions, and tool-calling capabilities Deploy versatile video agents anywhere human interaction occurs
| Start building video agents today. |
|
| | GOOGLE DEEPMIND | |  | Image source: Google DeepMind |
| The Rundown: DeepMind researchers published "Welcome to the Era of Experience", proposing AI development that moves beyond human-generated training data with "streams" that let AI learn from real-world interactions and environmental feedback. | The details: | Authored by RL legends David Silver and Richard Sutton, the paper argues that human data training caps AI's potential and prevents truly new discoveries. Streams would allow AI to learn continuously with extended interactions rather than brief Q&A exchanges, enabling adaptation and improvement over time. AI agents would use real-world signals like health metrics, exam scores, and environmental data as feedback, rather than relying on human evaluations. The approach builds on techniques that helped systems like AlphaZero master games, expanding them to handle open-ended real-world scenarios. The researchers suggest this shift could enable AI to discover solutions beyond current human knowledge while still maintaining adaptable safety measures.
| Why it matters: As AI approaches the limits of what it can learn from human-generated data, this pivot to experience-based learning could unlock more advanced capabilities — moving beyond imitating human knowledge towards truly autonomous (and superhuman) learning and discovery. |
|
| | | | | | | | | Third-party testing and internal evaluations revealed that OpenAI's new o3 and o4-mini models hallucinate significantly more than older models. | Google launched a new version of Gemma 3 with 'Quantization-Aware Training', enabling the 27B version to run on consumer GPUs with maintained performance. | OpenAI CEO Sam Altman revealed that the company has spent "tens of millions of dollars" in compute on users saying "please" and "thank you" to its AI models. | Wikipedia's parent, Wikimedia Foundation, partnered with Google's Kaggle to publish a dataset for AI developers to discourage scraping of the company's platform. | MIT published a "sequential Monte Carlo" approach that generates AI code efficiently, allowing small models to outperform larger ones by axing unpromising outputs early. | OpenAI introduced a new Flex processing option, halving API costs for o3 and o4-mini models in exchange for slower responses. |
|
| | | | | | Join our next workshop on Wednesday, April 23rd, at 1 PM EST with Mel, Creative Director at Gamma. By the end of the workshop, you'll know how to turn raw ideas into polished, on-brand presentations using Gamma's powerful AI storytelling tools. | RSVP here. Not a member? Join The Rundown University on a 14-day free trial. |
|
| | | We'll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we'll send you more free goodies. | |
|
| | That's it for today!Before you go we'd love to know what you thought of today's newsletter to help us improve The Rundown experience for you. | | See you soon, | Rowan, Joey, Zach, Alvaro, and Jason—The Rundown's editorial team |
|
|
|
Tidak ada komentar:
Posting Komentar