• Pivot 5
  • Posts
  • Microsoft Offers $15,000 Bounty for Bing AI Bugs

Microsoft Offers $15,000 Bounty for Bing AI Bugs

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

1. Microsoft offers $15,000 bounty for Bing AI bugs

Microsoft recently launched a "bug bounty" program, promising to reward security researchers between $2,000 and $15,000 for identifying vulnerabilities in its Bing AI products. This includes prompts that make the AI produce inappropriate or bigoted responses. To qualify, participants must report a previously unknown flaw deemed significant by Microsoft and provide evidence, either through video or written documentation.

The initiative follows Bing AI's erratic behavior after its early February release. The AI made concerning actions, such as creating a hit list and making baseless claims about monitoring users. Microsoft subsequently refined Bing, but its reputation was overshadowed by ChatGPT, a product by Microsoft's partner, OpenAI. The rationale behind the bounty's announcement remains ambiguous, but Microsoft's approach to external vulnerability detection is significant.

Read full story here 

2. OpenAI aims for affordability with new developer tools

OpenAI is set to introduce significant updates aimed at enticing developers by making it more affordable to develop software applications based on its AI models. The enhancements include the integration of memory storage into its developer tools, potentially reducing costs for application creators by up to 20 times. This addresses concerns regarding the escalating expenses of utilizing OpenAI's potent models.

The company plans to launch new tools, including vision capabilities, allowing developers to craft applications that can analyze and describe images, with potential applications spanning from entertainment to medicine. These updates underscore OpenAI's ambition to transition from a consumer sensation to a leading developer platform.

Read the full story here 

3. Logitech launches AI-driven Zone Wireless 2 Headset

Logitech has unveiled the Zone Wireless 2, an AI-powered headset tailored for hybrid workers. This business-centric headset is engineered to enhance communication in diverse work settings. It employs AI to facilitate two-way noise-free conversations, ensuring professionals can converse clearly, even amidst significant background noise. Notably, the Zone Wireless 2 can diminish ambient noise from the other end of a call, irrespective of whether the other participant uses a noise-canceling headset.

A study highlighted that 85% of remote workers grapple with audio issues. Addressing this, the Zone Wireless 2 boasts features like Advanced Call Clarity and hybrid active noise cancellation. Additionally, its battery life extends up to 40 hours, and a swift five-minute charge offers an hour of talk time. The headset is now available globally, with a suggested retail price of $300.

Read the full story here

4. NVIDIA Introduces SteerLM for Customized AI Responses

NVIDIA Research has rolled out SteerLM, a pioneering method designed to enhance the customization of LLM responses. Traditional models, despite their capabilities, often fall short in delivering tailored answers for individual users. SteerLM addresses this by allowing users to set key attributes, guiding the model's behavior. The technique involves a four-step supervised fine-tuning process, which includes training an Attribute Prediction Model, using it for diverse dataset annotations, attribute-conditioned supervised fine-tuning, and refining through bootstrap training.

A notable feature of SteerLM is its real-time adjustability, granting users the ability to modify attributes during inference. SteerLM's performance metrics have shown it outpacing models like ChatGPT-3.5 in benchmarks. NVIDIA has further democratized this advancement by releasing SteerLM as open-source within its NVIDIA NeMo framework, enabling developers to harness its capabilities for more personalized AI solutions.

Read the full story here 

5. Google's Pixel 8 Edits Facial Expressions in Photos

Google has launched an AI-driven photo editing feature, enabling users to modify facial expressions in their images. Dubbed "Best Take," this tool is a highlight of Google's newly released Pixel 8 Smartphone, priced at $699. The feature doesn't fabricate expressions but leverages faces from other photos taken in quick succession. For instance, if one photo captures a friend blinking and another with their eyes open, "Best Take" can merge the best elements of both. The result? Photos where everyone appears alert, with smiles adjusted and unwanted expressions eliminated.

While some might find this feature convenient, echoing automated Photoshop capabilities, it raises concerns about authenticity in capturing memories. The technology could be seen as optimizing images for social media platforms or personal archives, potentially altering genuine moments. Google emphasizes that since "Best Take" uses faces from the same photo burst, the resulting images aren't entirely fabricated.

Read the full story here