Topics
late
AI
Amazon
Image Credits:Google
Apps
Biotech & Health
mood
Image Credits:Google
Cloud Computing
Commerce
Crypto
endeavor
EVs
Fintech
Fundraising
Gadgets
Gaming
Government & Policy
ironware
Layoffs
Media & Entertainment
Meta
Microsoft
privateness
Robotics
Security
Social
Space
Startups
TikTok
DoT
Venture
More from TechCrunch
Events
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
Google is embarrassed about its AI Overviews , too . After a deluge ofdunks and memes over the preceding hebdomad , which cracked on the poor quality and outright misinformation that arise from the tech giant ’s underbaked new AI - powered search feature , the companionship on Thursday issued a mea culpa of sort . Google — a companionship whose name is synonymous with searching the web — whose brand sharpen on “ organizing the world ’s information ” and putting it at user ’s fingertip — actuallywrote in a web log postthat “ some curious , inaccurate or unhelpful AI Overviews certainly did show up . ”
That ’s put it gently .
Theadmission of loser , penned by Google VP and Head of Search Liz Reid , seems a testimonial as to how the driveway to mash AI technology into everything has now somehow made Google Search worse .
In the post entitle “ About last week , ” ( this got past PR ? ) , Reid spells out the many ways its AI Overviews make mistakes . While they do n’t “ hallucinate ” or make thing up the direction that other large language models ( LLMs ) may , she tell , they can get affair haywire for “ other reasons , ” like “ misinterpreting queries , misinterpreting a nuance of terminology on the web , or not having a lot of great information available . ”
Reid also mention that some of the screenshots shared on societal media over the past week were faked , while others were for nonsensical enquiry , like “ How many sway should I corrode ? ” — something no one ever really searched for before . Since there ’s short actual information on this topic , Google ’s AI guided a drug user to satiric content . ( In the case of the rocks , the satirical subject matter had beenpublishedon a geologic software program supplier ’s website . )
It ’s deserving charge out that if you had Googled “ How many rocks should I eat ? ” and were gift with a set of unhelpful links , or even a jokey article , you would n’t be surprised . What people are oppose to is the self-assurance with which the AI jabber back that “ geologists recommend run through at least one small rock per day ” as if it ’s a factual answer . It may not be a “ delusion , ” in technical terminus , but the ending exploiter does n’t deal . It ’s insane .
What ’s unsettling , too , is that Reid claims Google “ tested the feature extensively before launch , ” including with “ racy red - team up sweat . ”
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
Does no one at Google have a sense of humor then ? No one sentiment of prompts that would generate short results ?
In improver , Google downplayed the AI feature ’s trust on Reddit user information as a seed of noesis and truth . Although people have regularly add “ Reddit ” to their search for so long that Googlefinally made it a built - in search filter , Reddit is not a consistency of factual knowledge . And yet the AI would point to Reddit forum military post to serve questions , without an apprehension of when first - hand Reddit knowledge is helpful and when it is not — or forged , when it is a trolling .
Reddit today ismaking bankby offer its information to companies likeGoogle , OpenAIandothersto train their models , but that does n’t mean users want Google ’s AI deciding when to search Reddit for an answer , or suggest that someone ’s opinion is a fact . There ’s nuance to learning when to explore Reddit and Google ’s AI does n’t translate that yet .
As Reid admits , “ forums are often a great origin of authentic , first - hand information , but in some cases can take to less - than - helpful advice , like using gum to get cheese to adhere to pizza pie , ” she allege , referencing one of the AI feature film ’s more striking failure over the retiring calendar week .
Google AI overview suggests adding gum to get cheese to stick to pizza pie , and it turn out the author is an 11 year old Reddit comment from substance abuser F*cksmith 😂 pic.twitter.com/uDPAbsAKeO
— Peter Yang ( @petergyang)May 23 , 2024
If last workweek was a catastrophe , though , at least Google is iterate speedily as a result — or so it says .
The company say it ’s face at instance from AI Overviews and identify shape where it could do better , include work up proficient signal detection mechanisms for preposterous question , limiting the user of user - generate content for responses that could offer deceptive advice , adding triggering confinement for queries where AI Overviews were not helpful , not showing AI Overviews for severe news theme , “ where freshness and factualness are important , ” and summate extra spark purification to its protections for health lookup .
With AI companies work up ever - ameliorate chatbots every day , the interrogation is not on whether they will ever outperform Google Search for help us sympathise the world ’s information , but whether Google Search will ever be able to get up to pep pill on AI to gainsay them in return .
As farcical as Google ’s mistakes may be , it ’s too soon to count it out of the race yet — especially yield the massive scale of Google ’s beta - testing work party , which is fundamentally anybody who uses hunting .
“ There ’s nothing quite like take millions of people using the feature with many new searches , ” says Reid .
