What is Q* (Q-Star)?

Welcome to Daily Zaps — Here’s what we got for you today: 🙌

What is Q* (Q-Star)?

Shortly before being fired by OpenAI's board, CEO Sam Altman hinted at a significant technical advancement, overshadowed by the company's ensuing turmoil. This breakthrough, led by chief scientist Ilya Sutskever, involved developing more potent AI models, including a model called Q* by researchers Jakub Pachocki and Szymon Sidor, capable of solving new math problems. However, this rapid progress and the lack of adequate safeguards sparked concerns among some OpenAI employees about AI safety and the pace of technological advancements.

Internal tensions and safety concerns at OpenAI continued even after Altman's reinstatement as CEO. Sutskever, who had been working on projects to enhance AI reasoning capabilities, formed a team to address threats posed by highly intelligent AI systems. The breakthrough, involving training models with computer-generated data, led to the resignation of key figures like Pachocki, Sidor, and company president Greg Brockman, although some returned after Altman was reinstated.

Guide to understanding Q* / Q-Star / Q-Learning.

First AI influenced presidential election?

In the final weeks of Argentina's presidential election, AI technology played a significant role in the campaigns of both the right-wing libertarian winner Javier Milei and his Peronist rival Sergio Massa. Milei's team used AI to create a controversial image of Massa, while Massa's team distributed various AI-generated images and videos portraying him in heroic roles.

The growing accessibility and sophistication of AI-generated content are causing alarm among researchers and regulators, who are struggling to keep pace with these advancements. This situation was exemplified in Slovakia, where fake audio recordings significantly impacted the election process.

In response, companies like Meta Platforms are introducing measures to require disclosures for AI-altered political advertisements, while in the U.S., proposed legislation seeks to regulate AI-generated deepfakes in political ads. Despite these efforts, the challenge of regulating and managing AI's influence in political campaigns remains significant, with countries like Argentina yet to propose regulatory measures.

From Our Partners

Retrieval augmented generation (RAG) to reduce chatbot hallucinations

RAG reduces chatbot hallucinations by using below techniques:

Large Language Model (LLM): Central to RAG, uses a model like GPT for understanding queries and generating initial responses.

External Knowledge Database: RAG accesses an updated database with diverse sources (websites, journals, articles) for current and varied data, differentiating it from standard LLMs.

Embedding Model: Acts as a link between the LLM and the database, converting queries and database content into vector representations.

Retrieval Mechanism: Utilizes embeddings to fetch relevant information from the database, aligning data with the query's context and intent.

More details about RAG here.

Talk to Youtube videos with Bard extension

Google's AI chatbot Bard can now analyze YouTube videos and extract specific information. This advancement, while currently part of an opt-in Labs experience, could significantly change how users interact with YouTube content, potentially reducing direct engagement with creators' videos.

The situation presents a challenge for Google: while Bard's capabilities benefit users by simplifying access to information, they may detract from the value returned to content creators. This could affect creators' willingness to produce content for YouTube, prompting Google to address the balance between AI utility and creator compensation.

Google Meet hand raise gesture detection

Google Meet has introduced a new feature that enables users to raise their hand in meetings by physically raising their hand, which is recognized by Meet's gesture detection technology. This feature requires an enabled camera and clear visibility of the hand away from the face and body. It will not work when the user is the active speaker but will resume once they are no longer speaking.

The feature is turned off by default and can be activated in the "More options > Reactions > Hand Raise Gesture" menu. It's rolling out to various Google Workspace tiers starting November 21, 2023, for Rapid Release domains and November 28, 2023, for Scheduled Release domains. There are no admin controls for this feature, making it directly accessible to end users who are subscribed to eligible Google Workspace plans.