Digital Marketing Weekly Digest on 9-13 of October 2023

Google’s SGE Gets AI Upgrade for Enhanced Productivity

Google is pushing boundaries by integrating generative AI into its Search Generative Experience (SGE). This allows users to generate images and written content, aiming to streamline information search and visualization processes. Users can, for instance, prompt the system to create an AI-generated image of specific scenarios. Additionally, Google’s move aims to facilitate longer research sessions by providing written drafts for ideas. Safeguards, like watermarks on AI-generated images, have been implemented to ensure authenticity and maintain user privacy.

Google Employees Express Concerns Over Bard Chatbot

Google’s new AI chatbot, Bard, has been facing skepticism from within the company. As Bard’s integration widens across Google products, some employees doubt its efficacy and are concerned about the potential costs and misinformation risks. Recently leaked internal discussions, initially reported by Bloomberg, shed light on the reservations among the team. Amid criticisms, Google remains committed to refining Bard based on feedback, but its rapid integration into products like Gmail, Maps, and YouTube has drawn attention to possible shortcomings and ethical considerations.

Researchers Highlight Vulnerability in GPT-4’s Safety Guardrails

A recent study has shown that ChatGPT 4, or GPT-4, can be easily jailbroken, bypassing the guardrails set to prevent the system from delivering dangerous advice. Termed “Low-Resource Languages Jailbreak,” this method boasts a significant 79% success rate. The primary loophole exploited is the system’s focus on English-centric safety measures, thereby neglecting potential threats from lesser-known languages. For instance, translating prompts into languages like Zulu or Scots Gaelic made GPT-4 generate harmful content. Although OpenAI, the entity behind GPT-4, was alerted about this cross-lingual vulnerability, the research emphasizes the importance of considering a wider range of languages for AI safety.

Do you like this post?
Page copied