Wednesday, 30 April 2025
32.3 C
Singapore
36.1 C
Thailand
27.2 C
Indonesia
28.6 C
Philippines

Google struggles to correct weird AI answers in search

Google grapples with strange AI responses, racing to fix glitches and restore quality. Critics raise concerns over the company's output shift.

Google is facing a peculiar challenge as its new AI Overview feature is generating bizarre responses, such as advising users to apply glue to their pizza or suggesting they consume rocks. This has led to a flurry of memes on social media, prompting Google to swiftly disable AI overviews for specific searches. Users have noticed that these strange responses disappear shortly after being shared on social networks.

Challenges with AI overview

Despite being in beta since May 2023 under the name Search Generative Experience, the rollout of AI Overviews has been plagued with issues. CEO Sundar Pichai previously stated that Google had processed over a billion queries during the testing phase, highlighting the company’s extensive preparation. However, it seems that efforts to optimise the cost of delivering AI answers, reportedly reduced by 80% due to hardware and technical enhancements, may have preceded the technology’s readiness.

Critics have remarked on Google’s shift from a company known for high-quality output to one facing scrutiny for low-quality results. An anonymous AI founder expressed this sentiment to The Verge, highlighting a perceived decline in Google’s reputation.

Maintaining quality in AI outputs

In response to these concerns, Google spokesperson Meghann Farnsworth emphasised that the majority of AI Overview outputs offer “high-quality information.” Farnsworth noted that many of the odd examples cited were uncommon queries or doctored content. She confirmed that Google is actively removing AI overviews for certain queries and is using these instances to improve its systems, with some updates already in progress.

AI expert Gary Marcus, an emeritus professor at New York University, explained the challenge facing AI companies. While achieving 80% accuracy by approximating human data is relatively straightforward, the final 20% is significantly more challenging. Marcus believes this last portion might require artificial general intelligence (AGI), which current large language models, like Google’s Gemini and OpenAI’s GPT-4, are not equipped to deliver.

Google’s struggles with AI Overview come at a time of intense competition in the AI space. Bing made significant AI investments before Google, OpenAI is reportedly developing its own search engine, and a new AI search startup has already reached a US$1 billion valuation. Moreover, a younger demographic is increasingly favouring TikTok for its user experience, adding to Google’s competitive pressure.

Despite the setbacks, Google remains optimistic about AI Overview’s future. The current feature is just a small part of the larger vision, which includes multistep reasoning for complex queries, AI-organised result pages, and video search in Google Lens. However, the company recognises that its immediate focus must be on improving the basics to maintain its reputation.

Gary Marcus warned that current AI models lack the ability to perform sanity checks on their outputs, highlighting a significant challenge facing the industry. Google and other AI developers must address these issues to ensure their technologies meet user expectations and maintain trust in their capabilities.

Hot this week

Commvault expands cyber recovery services through CrowdStrike partnership

Commvault and CrowdStrike expand partnership to offer integrated cyber recovery and incident response services for stronger cyber resilience.

Grouphug brings AI to WhatsApp groups to turn private chats into memes

Grouphug wants to turn your WhatsApp group chats into memes using AI—and that’s only the beginning of this clever new app.

Vulnerability exploitation spikes as Tenable joins Verizon to highlight patching delays

Tenable reveals critical CVEs remain unpatched for over 200 days, risking exploitation, as highlighted in Verizon’s 2025 DBIR.

Veeam report reveals nearly 70% of organisations still targeted by ransomware

Nearly 70% of organisations were hit by ransomware last year, says Veeam, urging stronger recovery strategies and proactive resilience.

Gitex Asia x Ai Everything Singapore highlights robotics, AI and next-gen tech at inaugural event

Gitex Asia x Ai Everything Singapore highlights robotics, AI, startups, and tech innovations, shaping Southeast Asia’s digital future.

Google Play loses nearly half its apps since early 2024

Due to stricter rules and quality control changes, Google Play lost nearly half its apps in 2024, dropping from 3.4M to 1.8M.

Snapchat drops plans for simplified app, tests new five-tab layout instead

Snapchat has dropped its simplified app redesign and is testing a new five-tab layout to improve user experience and content discovery.

Startups fight back against Cluely’s AI cheating tool with detection software

Startups fight back against AI cheating tool Cluely with new detection software, while Cluely hints at future smart glasses and AI hardware.

Meta introduces new AI tools at LlamaCon

Meta's first LlamaCon event launches open AI tools to challenge OpenAI and promote accessible, developer-friendly AI solutions.

Related Articles

Popular Categories