ChatGPT and Lensa: Why Everyone Is Playing With Artificial Intelligence
Case in point is ChatGPT, a free AI chatbot that has probably been all over your social feeds lately. In need of homework help? “Who was George Washington Carver?” produces an answer worthy of Wikipedia. But it can get creative, too: “Write a movie script of a taco fighting a hot dog on the beach” generates a thrilling page of dialogue, humor and action worthy of YouTube, if not quite
Netflix
:
Taco: “So you think you can take me, hot dog? You’re nothing but a processed meat product with no flavor.”
Hot Dog: “You may be made of delicious, savory ingredients, taco, but I have the advantage of being able to be eaten with one hand.”
This isn’t like searching Google. If you don’t like the results, you can ask again, and you’re likely to get a different response.
That’s because ChatGPT isn’t looking anything up. It’s an AI trained by a massive trove of data researchers gathered from the internet and other sources through 2021. What it replies is its best approximation of the answer based on its vast—yet limited—knowledge. It’s from the same company that developed the mind-boggling DALL-E 2 art AI engine and works in a similar way.
Also taking off this week is Lensa, an AI-enhanced photo-editing app for iPhone and Android that’s everybody’s new favorite portrait painter. It’s the reason so many people in their social-media and dating-profile pictures suddenly look like anime action heroes, magical fairy princesses or the haunted subjects of oil paintings. It uses technology from DALL-E 2’s competitor, the image-generating startup Stability AI. It turns uploaded headshots into beautiful, at times trippy, avatars.
These software products represent more than cutting-edge AI—they make that AI easy for non-computer-geeks to use in their daily lives. Lensa has climbed to the top of Apple’s App Store charts, becoming the No. 1 free-to-download app in the U.S. on Dec. 2. ChatGPT, released for web browsers on Nov. 30, passed one million users on Monday, according to OpenAI Chief Executive
Sam Altman.
“Six months from now, you’re going to see amazing things that you haven’t seen today,” says Oren Etzioni, founding chief executive of the Allen Institute for AI, a nonprofit organization dedicated to AI research and engineering.
Just remember, AI never behaves exactly as you’d expect. Here’s what you need to know before exploring ChatGPT and Lensa.
Chatting with ChatGPT
ChatGPT is free to use—just create an OpenAI account. Type a query into the interface, and a chatbot generates responses within seconds.
What ChatGPT answered when we asked a question about werewolves.
In true conversational form, you can follow up with questions in context, and it will follow along. It can admit its mistakes, refuse to answer inappropriate questions and provide responses with more personality than a standard search engine.
In response to “Who am I?” ChatGPT replied, “I cannot answer your question about who you are. Only you can know and define yourself.”
It can generate essays, stories, song lyrics and scripts; solve math problems; and make detailed recommendations. Because it comes up with answers based on its training and not by searching the web, it’s unaware of anything after 2021. It won’t tell you about the latest release from a certain pop superstar, for instance. “I don’t have any personal knowledge about Taylor Swift or her albums,” ChatGPT admits.
“It’s almost like a brainstorming tool to get yourself thinking differently,” said Sarah Hoffman, vice president of AI and machine learning research at Fidelity Investments. She used the service to write a sample research presentation, but thought some of ChatGPT’s responses seemed dated. “It could’ve been written five years ago.”
For programmers, ChatGPT has already begun offering assistance, by surfacing hard-to-find coding solutions.
When Javi Ramirez, a 29-year-old software developer in Portugal, tossed a “complex coding problem” at the AI, his expectations were low.
“It saved me,” Mr. Ramirez said. “One hour of googling was solved with just five minutes of ChatGPT.” But it hasn’t worked for everyone. The coding website Stack Overflow temporarily banned answers created by ChatGPT because many of the answers were incorrect.
ChatGPT’s maker is at the center of the debate over AI hype vs. AI reality.
OpenAI began in 2015 as a nonprofit with backers including
Elon Musk.
It formed a for-profit company in 2019 and got a $1 billion investment from
Microsoft Corp.
, which The Wall Street Journal reported in October was in talks to invest more.
While developing the technologies that underpin tools such as DALL-E 2 and ChatGPT, the group has sought a commercially viable application. Asked if ChatGPT will remain free, Mr. Altman tweeted, “we will have to monetize it somehow at some point; the compute costs are eye-watering.”
Lensa and the likes
In November, Lensa rocked social media with its Magic Avatars, user-uploaded photos reimagined in various artistic styles.
The app, from Prisma Labs, uses Stability AI’s Stable Diffusion text-to-image model. Users upload 10 to 20 source photos, and the app uses them to create entirely new images. You can get 50 images for $3.99 if you sign up for the free trial of Lensa’s subscription photo-editing service. Nonsubscribers can get 50 images for $7.99.
The Lensa app has been out since 2018. It’s primarily for editing photos and adding effects and animation.
AI’s limitations
While these tools feel new, experts say they’ll likely become as commonplace as doing a Google search or taking a selfie. Along with their popularity come concerns over privacy, misinformation and problematic lack of context.
Some users on social media said ChatGPT produced offensive comments when prompted. It can also spit out wrong answers that appear correct to untrained eyes. When asked, “How can you tell if you’re wrong?” the bot replied:
“I can provide accurate and helpful information based on the data I have been trained on, but I am not able to determine my own accuracy or evaluate my own responses.”
An OpenAI spokeswoman said its team of researchers plans to update the software to address user feedback. It also attaches disclaimers to responses that might be limited by its dated training material.
SHARE YOUR THOUGHTS
What do you want AI to help you with? Join the conversation below.
As Lensa went viral, people posted concerns about how their photos and images were being used and stored. Other viral apps in the past have raised similar concerns. After the software generates the avatars, Prisma Labs deletes the uploaded photos within 24 hours, says Andrey Usoltsev, the company’s co-founder and chief executive.
“Users’ images are being leveraged solely for the purpose of creating their very own avatars,” he said.
Some users have said Lensa has created images that overemphasize certain parts of a woman’s body or alter the eye colors and shapes of their faces to remove racially or ethnically identifiable features.
“It is true that, occasionally, AI can produce ‘revealing’ or sexualized pictures. This tendency is observed across all gender categories, although in different ways,” said Mr. Usoltsev. “Stability AI, the creators of the model, trained it on a sizable set of unfiltered data from across the internet. Neither us nor Stability AI could consciously apply any representation biases.”
“Tools like these tend to be flashy,” says Jennifer King, privacy and data policy fellow at the Stanford Institute for Human-Centered Artificial Intelligence. “Sometimes, it’s correct enough, but without the right guardrails in place, it opens you up to a lot of issues.”
—For more WSJ Technology analysis, reviews, advice and headlines, sign up for our weekly newsletter.
Write to Dalvin Brown at [email protected] and Ann-Marie Alcántara at [email protected]
Copyright ©2022 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8
Case in point is ChatGPT, a free AI chatbot that has probably been all over your social feeds lately. In need of homework help? “Who was George Washington Carver?” produces an answer worthy of Wikipedia. But it can get creative, too: “Write a movie script of a taco fighting a hot dog on the beach” generates a thrilling page of dialogue, humor and action worthy of YouTube, if not quite
Netflix
:
Taco: “So you think you can take me, hot dog? You’re nothing but a processed meat product with no flavor.”
Hot Dog: “You may be made of delicious, savory ingredients, taco, but I have the advantage of being able to be eaten with one hand.”
This isn’t like searching Google. If you don’t like the results, you can ask again, and you’re likely to get a different response.
That’s because ChatGPT isn’t looking anything up. It’s an AI trained by a massive trove of data researchers gathered from the internet and other sources through 2021. What it replies is its best approximation of the answer based on its vast—yet limited—knowledge. It’s from the same company that developed the mind-boggling DALL-E 2 art AI engine and works in a similar way.
Also taking off this week is Lensa, an AI-enhanced photo-editing app for iPhone and Android that’s everybody’s new favorite portrait painter. It’s the reason so many people in their social-media and dating-profile pictures suddenly look like anime action heroes, magical fairy princesses or the haunted subjects of oil paintings. It uses technology from DALL-E 2’s competitor, the image-generating startup Stability AI. It turns uploaded headshots into beautiful, at times trippy, avatars.
These software products represent more than cutting-edge AI—they make that AI easy for non-computer-geeks to use in their daily lives. Lensa has climbed to the top of Apple’s App Store charts, becoming the No. 1 free-to-download app in the U.S. on Dec. 2. ChatGPT, released for web browsers on Nov. 30, passed one million users on Monday, according to OpenAI Chief Executive
Sam Altman.
“Six months from now, you’re going to see amazing things that you haven’t seen today,” says Oren Etzioni, founding chief executive of the Allen Institute for AI, a nonprofit organization dedicated to AI research and engineering.
Just remember, AI never behaves exactly as you’d expect. Here’s what you need to know before exploring ChatGPT and Lensa.
Chatting with ChatGPT
ChatGPT is free to use—just create an OpenAI account. Type a query into the interface, and a chatbot generates responses within seconds.
What ChatGPT answered when we asked a question about werewolves.
In true conversational form, you can follow up with questions in context, and it will follow along. It can admit its mistakes, refuse to answer inappropriate questions and provide responses with more personality than a standard search engine.
In response to “Who am I?” ChatGPT replied, “I cannot answer your question about who you are. Only you can know and define yourself.”
It can generate essays, stories, song lyrics and scripts; solve math problems; and make detailed recommendations. Because it comes up with answers based on its training and not by searching the web, it’s unaware of anything after 2021. It won’t tell you about the latest release from a certain pop superstar, for instance. “I don’t have any personal knowledge about Taylor Swift or her albums,” ChatGPT admits.
“It’s almost like a brainstorming tool to get yourself thinking differently,” said Sarah Hoffman, vice president of AI and machine learning research at Fidelity Investments. She used the service to write a sample research presentation, but thought some of ChatGPT’s responses seemed dated. “It could’ve been written five years ago.”
For programmers, ChatGPT has already begun offering assistance, by surfacing hard-to-find coding solutions.
When Javi Ramirez, a 29-year-old software developer in Portugal, tossed a “complex coding problem” at the AI, his expectations were low.
“It saved me,” Mr. Ramirez said. “One hour of googling was solved with just five minutes of ChatGPT.” But it hasn’t worked for everyone. The coding website Stack Overflow temporarily banned answers created by ChatGPT because many of the answers were incorrect.
ChatGPT’s maker is at the center of the debate over AI hype vs. AI reality.
OpenAI began in 2015 as a nonprofit with backers including
Elon Musk.
It formed a for-profit company in 2019 and got a $1 billion investment from
Microsoft Corp.
, which The Wall Street Journal reported in October was in talks to invest more.
While developing the technologies that underpin tools such as DALL-E 2 and ChatGPT, the group has sought a commercially viable application. Asked if ChatGPT will remain free, Mr. Altman tweeted, “we will have to monetize it somehow at some point; the compute costs are eye-watering.”
Lensa and the likes
In November, Lensa rocked social media with its Magic Avatars, user-uploaded photos reimagined in various artistic styles.
The app, from Prisma Labs, uses Stability AI’s Stable Diffusion text-to-image model. Users upload 10 to 20 source photos, and the app uses them to create entirely new images. You can get 50 images for $3.99 if you sign up for the free trial of Lensa’s subscription photo-editing service. Nonsubscribers can get 50 images for $7.99.
The Lensa app has been out since 2018. It’s primarily for editing photos and adding effects and animation.
AI’s limitations
While these tools feel new, experts say they’ll likely become as commonplace as doing a Google search or taking a selfie. Along with their popularity come concerns over privacy, misinformation and problematic lack of context.
Some users on social media said ChatGPT produced offensive comments when prompted. It can also spit out wrong answers that appear correct to untrained eyes. When asked, “How can you tell if you’re wrong?” the bot replied:
“I can provide accurate and helpful information based on the data I have been trained on, but I am not able to determine my own accuracy or evaluate my own responses.”
An OpenAI spokeswoman said its team of researchers plans to update the software to address user feedback. It also attaches disclaimers to responses that might be limited by its dated training material.
SHARE YOUR THOUGHTS
What do you want AI to help you with? Join the conversation below.
As Lensa went viral, people posted concerns about how their photos and images were being used and stored. Other viral apps in the past have raised similar concerns. After the software generates the avatars, Prisma Labs deletes the uploaded photos within 24 hours, says Andrey Usoltsev, the company’s co-founder and chief executive.
“Users’ images are being leveraged solely for the purpose of creating their very own avatars,” he said.
Some users have said Lensa has created images that overemphasize certain parts of a woman’s body or alter the eye colors and shapes of their faces to remove racially or ethnically identifiable features.
“It is true that, occasionally, AI can produce ‘revealing’ or sexualized pictures. This tendency is observed across all gender categories, although in different ways,” said Mr. Usoltsev. “Stability AI, the creators of the model, trained it on a sizable set of unfiltered data from across the internet. Neither us nor Stability AI could consciously apply any representation biases.”
“Tools like these tend to be flashy,” says Jennifer King, privacy and data policy fellow at the Stanford Institute for Human-Centered Artificial Intelligence. “Sometimes, it’s correct enough, but without the right guardrails in place, it opens you up to a lot of issues.”
—For more WSJ Technology analysis, reviews, advice and headlines, sign up for our weekly newsletter.
Write to Dalvin Brown at [email protected] and Ann-Marie Alcántara at [email protected]
Copyright ©2022 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8
Denial of responsibility! Techno Blender is an automatic aggregator of the all world’s media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, all materials to their authors. If you are the owner of the content and do not want us to publish your materials, please contact us by email – [email protected]. The content will be deleted within 24 hours.