Topics

a la mode

AI

Amazon

Article image

Image Credits:Silas Stein / picture alliance / Getty Images

Apps

Biotech & Health

mood

Cloud Computing

Department of Commerce

Crypto

go-ahead

EVs

Fintech

Fundraising

Gadgets

Gaming

Google

Government & Policy

computer hardware

Instagram

Layoffs

Media & Entertainment

Meta

Microsoft

privateness

Robotics

security system

Social

outer space

inauguration

TikTok

transit

speculation

More from TechCrunch

Events

Startup Battlefield

StrictlyVC

Podcasts

Videos

Partner Content

TechCrunch Brand Studio

Crunchboard

reach Us

OpenAIsays it ’ll make changesto the way it updates the AI model that power ChatGPT , following an incident that caused the chopine to become overly bootlicking for many user .

Last weekend , after OpenAI roll out a tweakedGPT-4o — the default mannikin powering ChatGPT — users on societal media noted that ChatGPT begin responding in an overly verifying and agreeable way . It quickly became a meme . Users posted screenshots of ChatGPT applauding all sorts of problematic , dangerousdecisionsandideas .

In a post on disco biscuit last Sunday , CEO Sam Altmanacknowledgedthe problem and said that OpenAI would work on fixes “ ASAP . ” On Tuesday , Altmanannouncedthe GPT-4o update was being rolled back and that OpenAI was working on “ additional mend ” to the model ’s personality .

The fellowship published apostmortemon Tuesday , and in a web log office Friday , OpenAI exposit on specific adjustments it project to make to its model deployment unconscious process .

OpenAI says it be after to enter an opt - in “ alpha phase angle ” for some role model that would allow sure ChatGPT users to test the models and give feedback prior to launch . The companionship also say it ’ll include explanations of “ known limitations ” for next incremental updates to models in ChatGPT , and adjust its safety review process to formally consider “ model behavior issues ” like personality , legerdemain , reliability , and hallucination ( i.e. , when a model realise things up ) as “ launching - blocking ” concern .

“ Going forward , we ’ll proactively communicate about the update we ’re make to the model in ChatGPT , whether ‘ subtle ’ or not , ” wrote OpenAI in the blog position . “ Even if these issue are n’t perfectly quantifiable today , we commit to blocking launches establish on placeholder measurements or qualitative signals , even when system of measurement like A / group B examination look good . ”

we miss the mark with last week ’s GPT-4o update.what pass , what we learned , and some things we will do differently in the future : https://t.co / ER1GmRYrIC

The pledge fixes amount as more masses plow to ChatGPT for advice . According to one recent surveyby lawsuit financier Express Legal Funding , 60 % of U.S. adults have used ChatGPT to essay counsel or information . The growing reliance on ChatGPT — and the weapons platform ’s enormous substance abuser base — raises the stakes when issues like extreme sycophancy emerge , not to mention hallucination and other technical shortcoming .

Join us at TechCrunch Sessions: AI

Exhibit at TechCrunch Sessions: AI

As one mitigating stride , earlier this hebdomad , OpenAI say it would experiment with ways to let drug user give “ real - prison term feedback ” to “ directly tempt their interactions ” with ChatGPT . The society also said it would refine techniques to channelise simulation away from sycophancy , potentially allow people to choose from multiple model personality in ChatGPT , make extra safety guardrails , and expand evaluations to assist identify issues beyond sycophancy .

“ One of the bad moral is fully recognizing how people have start up to practice ChatGPT for deep personal advice — something we did n’t see as much even a year ago , ” continued OpenAI in its blog place . “ At the fourth dimension , this was n’t a primary focus , but as AI and society have co - evolved , it ’s become clear that we need to treat this use casing with great care . It ’s now going to be a more meaningful part of our safety gadget piece of work . ”