Toyota Research Institute Uses Generative AI to Teach Robots Breakfast-Making Skills

Pivot 5: 5 stories. 5 minutes a day. 5 days a week.

1. Toyota Research Institute Uses Generative AI to Teach Robots Breakfast-Making Skills

Toyota Research Institute (TRI) is pushing the boundaries of robotics by using generative AI to teach robots how to make breakfast. Unlike traditional methods that require extensive coding, TRI's approach involves giving robots a sense of touch. This tactile capability allows the robots to "feel" their actions, making complex tasks easier to execute.

The institute is working on developing "Large Behavior Models" for robots, which learn by observing and then generalize new skills. According to Russ Tedrake, MIT robotics professor and VP of robotics research at TRI, these LBMs can perform skills they've never been explicitly taught. The researchers have already trained the robots in over 60 challenging skills, such as pouring liquids and using tools, with plans to reach 1,000 skills by the end of 2024.

Google and Tesla are also conducting similar research. Their robots use experience to infer how to perform tasks, much like TRI's approach. However, the process is generally slow and labor-intensive, requiring more than just feeding data to an AI model. Despite these challenges, TRI's research represents a significant advancement in the field of robotics, demonstrating the potential for AI-trained robots to carry out complex tasks with minimal instruction.

Read the full story here 

2. HiddenLayer secures $50M in Series A funding to fortify enterprise AI models

HiddenLayer, an Austin-based cybersecurity startup, has announced a $50 million Series A funding round to bolster the defenses of enterprise AI models. The round was led by M12, Microsoft’s Venture Fund, and Moore Strategic Ventures, with participation from Booz Allen Ventures, IBM Ventures, Capital One Ventures, and Ten Eleven Ventures.

The startup has developed its MLSec Platform, a suite of tools designed to safeguard enterprise machine learning and AI models. These tools passively monitor the performance and operations of enterprise ML/AI models in real-time, scanning for vulnerabilities and offering recommendations for hardening them.

Earlier this year, HiddenLayer struck a partnership with Databricks, allowing Databricks' enterprise customers to use HiddenLayer’s MLSec Platform directly on their models. With the new funding, HiddenLayer plans to hire another 40 personnel by the end of the year and continue expanding its client base, making it a cybersecurity startup to watch closely in the AI industry.

Read the full story here 

3. Microsoft's AI research leak exposes 38 terabytes of sensitive data

Microsoft AI researchers inadvertently leaked a staggering 38 terabytes of confidential company data on GitHub, according to a report from cloud security firm Wiz. The leaked files contained a full disc backup of two employees' workstations, including sensitive personal data, company secrets, and over 30,000 internal Microsoft Teams messages.

The leak occurred when Microsoft AI researchers were attempting to publish open-source training material and AI models for image recognition on GitHub. A misconfigured SAS token granted general access to the entire storage account, rather than limiting it to the intended AI material. This error granted "full control" access, allowing anyone to potentially inject malicious code into the AI models.

While Microsoft has since resolved the issue and stated that no customer data was exposed, the incident serves as a stark reminder of the security risks associated with handling large volumes of data for AI training. As companies rush to bring new AI products to market, the necessity for stringent security measures becomes increasingly evident.

Read the full story here

4. LimeWire Acquires BlueWillow to Expand Its Generative AI Capabilities in Content Creation

LimeWire, once infamous for music piracy, has pivoted into the world of content creation and recently announced its acquisition of BlueWillow, a popular generative AI image creation platform. BlueWillow has quickly gained traction, boasting a community of 2.5 million members on Discord and generating over 500 million images.

The acquisition aligns with LimeWire's ongoing efforts to diversify its business model. Initially relaunched as an NFT marketplace for music creators, LimeWire has since shifted its focus towards content creation. The company has raised about $17.5 million through token sales and was valued at around $60 million earlier this year.

While the financial terms of the deal remain undisclosed, BlueWillow's team will not be part of the acquisition. Ritankar Das, the founder of BlueWillow, mentioned that the team plans to leave and work on a new AI venture that is still in stealth mode. As LimeWire continues to evolve, the acquisition of BlueWillow marks a significant step in its journey to become a comprehensive platform for AI-driven content creation.

Read the full story here 

5. Google Bard AI assistant gets smarter with app integrations and fact-checking features

Google has rolled out significant updates to its AI assistant, Google Bard, enhancing its capabilities and addressing some of its limitations. One notable addition is the "double-check button," aimed at countering inaccuracies or "hallucinations" in Bard's responses. When users click on a small "G" logo below Bard's results, the assistant searches the web to validate its statements.

The updates also introduce Bard Extensions, opt-in features that allow the assistant to pull data from Google's other services like Gmail, Google Drive, and Google Maps. For instance, Bard can read through your emails to pull dates for planning a trip or summarize a resume stored in Google Drive.

Addressing privacy concerns, Google assures that content from these services will not be viewed by human reviewers and will not be used for ads. While experts and tests suggest that Google Bard may be less capable than OpenAI's ChatGPT-4, Google's ability to integrate its vast array of apps gives it a unique edge.

Read the full story here 

Listen to the daily Pivot 5 podcast on Spotify, Apple Podcasts, and more