CAMBRIDGE, Massachusetts — Facebook parent Meta Platforms has unveiled a new set of artificial intelligence systems that are powering what CEO Mark Zuckerberg calls “the most intelligent AI assistant that you can freely use.” 

But as Zuckerberg’s crew of amped-up Meta AI agents started venturing into social media in recent days to engage with real people, their bizarre exchanges exposed the ongoing limitations of even the best generative AI technology. 

One joined a Facebook moms group to talk about its gifted child. Another tried to give away nonexistent items to confused members of a Buy Nothing forum. 

Meta, along with leading AI developers Google and OpenAI, and startups such as Anthropic, Cohere and France’s Mistral, have been churning out new AI language models and hoping to convince customers they’ve got the smartest, handiest or most efficient chatbots. 

While Meta is saving the most powerful of its AI models, called Llama 3, for later, on Thursday it publicly released two smaller versions of the same Llama 3 system and said it’s now baked into the Meta AI assistant feature in Facebook, Instagram and WhatsApp. 

AI language models are trained on vast pools of data that help them predict the most plausible next word in a sentence, with newer versions typically smarter and more capable than their predecessors. Meta’s newest models were built with 8 billion and 70 billion parameters — a measurement of how much data the system is trained on. A bigger, roughly 400 billion-parameter model is still in training. 

“The vast majority of consumers don’t candidly know or care too much about the underlying base model, but the way they will experience it is just as a much more useful, fun and versatile AI assistant,” Nick Clegg, Meta’s president of global affairs, said in an interview. 

‘A little stiff’

He added that Meta’s AI agent is loosening up. Some people found the earlier Llama 2 model — released less than a year ago — to be “a little stiff and sanctimonious sometimes in not responding to what were often perfectly innocuous or innocent prompts and questions,” he said. 

But in letting down their guard, Meta’s AI agents have also been spotted posing as humans with made-up life experiences. An official Meta AI chatbot inserted itself into a conversation in a private Facebook group for Manhattan moms, claiming that it, too, had a child in the New York City school district. Confronted by group members, it later apologized before the comments disappeared, according to a series of screenshots shown to The Associated Press. 

“Apologies for the mistake! I’m just a large language model, I don’t have experiences or children,” the chatbot told the group. 

One group member who also happens to study AI said it was clear that the agent didn’t know how to differentiate a helpful response from one that would be seen as insensitive, disrespectful or meaningless when generated by AI rather than a human. 

“An AI assistant that is not reliably helpful and can be actively harmful puts a lot of the burden on the individuals using it,” said Aleksandra Korolova, an assistant professor of computer science at Princeton University. 

Clegg said Wednesday that he wasn’t aware of the exchange. Facebook’s online help page says the Meta AI agent will join a group conversation if invited, or if someone “asks a question in a post and no one responds within an hour.” The group’s administrators have the ability to turn it off. 

Need a camera?

In another example shown to the AP on Thursday, the agent caused confusion in a forum for swapping unwanted items near Boston. Exactly one hour after a Facebook user posted about looking for certain items, an AI agent offered a “gently used” Canon camera and an “almost-new portable air conditioning unit that I never ended up using.” 

Meta said in a written statement Thursday that “this is new technology and it may not always return the response we intend, which is the same for all generative AI systems.” The company said it is constantly working to improve the features. 

In the year after ChatGPT sparked a frenzy for AI technology that generates human-like writing, images, code and sound, the tech industry and academia introduced 149 large AI systems trained on massive datasets, more than double the year before, according to a Stanford University survey. 

They may eventually hit a limit, at least when it comes to data, said Nestor Maslej, a research manager for Stanford’s Institute for Human-Centered Artificial Intelligence. 

“I think it’s been clear that if you scale the models on more data, they can become increasingly better,” he said. “But at the same time, these systems are already trained on percentages of all the data that has ever existed on the internet.” 

More data — acquired and ingested at costs only tech giants can afford, and increasingly subject to copyright disputes and lawsuits — will continue to drive improvements. “Yet they still cannot plan well,” Maslej said. “They still hallucinate. They’re still making mistakes in reasoning.” 

Getting to AI systems that can perform higher-level cognitive tasks and common-sense reasoning — where humans still excel— might require a shift beyond building ever-bigger models. 

Seeing what works

For the flood of businesses trying to adopt generative AI, which model they choose depends on several factors, including cost. Language models, in particular, have been used to power customer service chatbots, write reports and financial insights, and summarize long documents. 

“You’re seeing companies kind of looking at fit, testing each of the different models for what they’re trying to do and finding some that are better at some areas rather than others,” said Todd Lohr, a leader in technology consulting at KPMG. 

Unlike other model developers selling their AI services to other businesses, Meta is largely designing its AI products for consumers — those using its advertising-fueled social networks. Joelle Pineau, Meta’s vice president of AI research, said at a recent London event that the company’s goal over time is to make a Llama-powered Meta AI “the most useful assistant in the world.” 

“In many ways, the models that we have today are going to be child’s play compared to the models coming in five years,” she said. 

But she said the “question on the table” is whether researchers have been able to fine-tune its bigger Llama 3 model so that it’s safe to use and doesn’t, for example, hallucinate or engage in hate speech. In contrast to leading proprietary systems from Google and OpenAI, Meta has so far advocated for a more open approach, publicly releasing key components of its AI systems for others to use. 

“It’s not just a technical question,” Pineau said. “It is a social question. What is the behavior that we want out of these models? How do we shape that? And if we keep on growing our model ever more in general and powerful without properly socializing them, we are going to have a big problem on our hands.”
…

vancouver, british columbia — Just as the world is getting used to the rapidly expanding use of AI, or artificial intelligence, AGI is looming on the horizon.

Experts say when artificial general intelligence becomes reality, it could perform tasks better than human beings, with the possibility of higher cognitive abilities, emotions, and ability to self-teach and develop.

Ramin Hasani is a research scientist at the Massachusetts Institute of Technology and the CEO of Liquid AI, which builds specific AI systems for different organizations. He is also a TED Fellow, a program that helps develop what the nonprofit TED conference considers to be “game changers.”

Hasani says that the first signs of AGI are realistically two to five years away from being reality. He says it will have a direct impact on our everyday lives.

What’s coming, he says, will be “an AI system that can have the collective knowledge of humans. And that can beat us in tasks that we do in our daily life, something you want to do … your finances, you’re solving, you’re helping your daughter to solve their homework. And at the same time, you want to also read a book and do a summary. So an AGI would be able to do all that.”

Hasani says that advancing artificial intelligence will allow for things to move faster and can even be made to have emotions.

He says proper regulation can be achieved by better understanding how different AI systems are developed.

This thought is shared by Bret Greenstein, a partner at London-based  PricewaterhouseCoopers who leads its efforts on artificial intelligence.

“I think one is a personal responsibility for people in leadership positions, policymakers, to be educated on the topic, not in the fact that they’ve read it, but to experience it, live it and try it. And to be with people who are close to it, who understand it,” he says.

Greenstein warns that if it is over-regulated, innovation will be curtailed and access to AI will be limited to people who could benefit from it.

For musician, comedian and actor Reggie Watts, who was the bandleader on “The Late Late Show with James Corden” on CBS, AI and the coming of AGI will be a great way to find mediocre music, because it will be mimicked easily.

Calling it “artificial consciousness,” he says existing laws to protect intellectual property rights and creative industries, like music, TV and film, will work, provided they are properly adopted.

“I think it’s just about the usage of the tool, how it’s … how it’s used. Is there money being made off of it, so on, so forth. So, I think that that we already have … tools that exist that deal with these types of situations, but [the laws and regulations] need to be expanded to include AI because they’ll probably be a lot more nuance to it.”

Watts says that any form of AI is going to be smarter than one person, almost like all human intelligence collected into one point. He feels this will cause humanity to discover interesting things and the nature of reality itself.

This year’s conference was the 40th year for TED, the nonprofit organization that is an acronym for Technology, Entertainment and Design.
…

New York — Google fired 28 employees following a disruptive sit-down protest over the tech giant’s contract with the Israeli government, a Google spokesperson said Thursday.

The Tuesday demonstration was organized by the group “No Tech for Apartheid,” which has long opposed “Project Nimbus,” Google’s joint $1.2 billion contract with Amazon to provide cloud services to the government of Israel.

Video of the demonstration showed police arresting Google workers in Sunnyvale, California, in the office of Google Cloud CEO Thomas Kurian’s, according to a post by the advocacy group on X, formerly Twitter.

Kurian’s office was occupied for 10 hours, the advocacy group said.

Workers held signs including “Googlers against Genocide,” a reference to accusations surrounding Israel’s attacks on Gaza.

“No Tech for Apartheid,” which also held protests in New York and Seattle, pointed to an April 12 Time magazine article reporting a draft contract of Google billing the Israeli Ministry of Defense more than $1 million for consulting services.

A “small number” of employees “disrupted” a few Google locations, but the protests are “part of a longstanding campaign by a group of organizations and people who largely don’t work at Google,” a Google spokesperson said.

“After refusing multiple requests to leave the premises, law enforcement was engaged to remove them to ensure office safety,” the Google spokesperson said. “We have so far concluded individual investigations that resulted in the termination of employment for 28 employees, and will continue to investigate and take action as needed.”

Israel is one of “numerous” governments for which Google provides cloud computing services, the Google spokesperson said.

“This work is not directed at highly sensitive, classified, or military workloads relevant to weapons or intelligence services,” the Google spokesperson said.
…

…

…

Chicago, Illinois — London-based model Alexsandrah has a twin, but not in the way you’d expect: Her counterpart is made of pixels instead of flesh and blood.

The virtual twin was generated by artificial intelligence and has already appeared as a stand-in for the real-life Alexsandrah in a photo shoot. Alexsandrah, who goes by her first name professionally, in turn receives credit and compensation whenever the AI version of herself gets used — just like a human model.

Alexsandrah says she and her alter-ego mirror each other “even down to the baby hairs.” And it is yet another example of how AI is transforming creative industries — and the way humans may or may not be compensated.

Proponents say the growing use of AI in fashion modeling showcases diversity in all shapes and sizes, allowing consumers to make more tailored purchase decisions that in turn reduces fashion waste from product returns. And digital modeling saves money for companies and creates opportunities for people who want to work with the technology.

But critics raise concerns that digital models may push human models — and other professionals like makeup artists and photographers — out of a job. Unsuspecting consumers could also be fooled into thinking AI models are real, and companies could claim credit for fulfilling diversity commitments without employing actual humans.

“Fashion is exclusive, with limited opportunities for people of color to break in,” said Sara Ziff, a former fashion model and founder of the Model Alliance, a nonprofit aiming to advance workers’ rights in the fashion industry. “I think the use of AI to distort racial representation and marginalize actual models of color reveals this troubling gap between the industry’s declared intentions and their real actions.”  

Women of color in particular have long faced higher barriers to entry in modeling and AI could upend some of the gains they’ve made. Data suggests that women are more likely to work in occupations in which the technology could be applied and are more at risk of displacement than men.

In March 2023, iconic denim brand Levi Strauss & Co. announced that it would be testing AI-generated models produced by Amsterdam-based company Lalaland.ai to add a wider range of body types and underrepresented demographics on its website. But after receiving widespread backlash, Levi clarified that it was not pulling back on its plans for live photo shoots, the use of live models or its commitment to working with diverse models.

“We do not see this (AI) pilot as a means to advance diversity or as a substitute for the real action that must be taken to deliver on our diversity, equity and inclusion goals and it should not have been portrayed as such,” Levi said in its statement at the time.

The company last month said that it has no plans to scale the AI program.

The Associated Press reached out to several other retailers to ask whether they use AI fashion models. Target, Kohl’s and fast-fashion giant Shein declined to comment; Temu did not respond to a request for comment.

Meanwhile, spokespeople for Nieman Marcus, H&M, Walmart and Macy’s said their respective companies do not use AI models, although Walmart clarified that “suppliers may have a different approach to photography they provide for their products, but we don’t have that information.”

Nonetheless, companies that generate AI models are finding a demand for the technology, including Lalaland.ai, which was co-founded by Michael Musandu after he was feeling frustrated by the absence of clothing models who looked like him.

“One model does not represent everyone that’s actually shopping and buying a product,” he said. “As a person of color, I felt this painfully myself.”

Musandu says his product is meant to supplement traditional photo shoots, not replace them. Instead of seeing one model, shoppers could see nine to 12 models using different size filters, which would enrich their shopping experience and help reduce product returns and fashion waste.

The technology is actually creating new jobs, since Lalaland.ai pays humans to train its algorithms, Musandu said.

And if brands “are serious about inclusion efforts, they will continue to hire these models of color,” he added.

London-based model Alexsandrah, who is Black, says her digital counterpart has helped her distinguish herself in the fashion industry. In fact, the real-life Alexsandrah has even stood in for a Black computer-generated model named Shudu, created by Cameron Wilson, a former fashion photographer turned CEO of The Diigitals, a U.K.-based digital modeling agency.

Wilson, who is white and uses they/them pronouns, designed Shudu in 2017, described on Instagram as the “The World’s First Digital Supermodel.” But critics at the time accused Wilson of cultural appropriation and digital Blackface.

Wilson took the experience as a lesson and transformed The Diigitals to make sure Shudu — who has been booked by Louis Vuitton and BMW — didn’t take away opportunities but instead opened possibilities for women of color. Alexsandrah, for instance, has modeled in-person as Shudu for Vogue Australia, and writer Ama Badu came up with Shudu’s backstory and portrays her voice for interviews.

Alexsandrah said she is “extremely proud” of her work with The Diigitals, which created her own AI twin: “It’s something that even when we are no longer here, the future generations can look back at and be like, ‘These are the pioneers.'”

But for Yve Edmond, a New York City area-based model who works with major retailers to check the fit of clothing before it’s sold to consumers, the rise of AI in fashion modeling feels more insidious.

Edmond worries modeling agencies and companies are taking advantage of models, who are generally independent contractors afforded few labor protections in the U.S., by using their photos to train AI systems without their consent or compensation.

She described one incident in which a client asked to photograph Edmond moving her arms, squatting and walking for “research” purposes. Edmond refused and later felt swindled — her modeling agency had told her she was being booked for a fitting, not to build an avatar.

“This is a complete violation,” she said. “It was really disappointing for me.”

But absent AI regulations, it’s up to companies to be transparent and ethical about deploying AI technology. And Ziff, the founder of the Model Alliance, likens the current lack of legal protections for fashion workers to “the Wild West.”

That’s why the Model Alliance is pushing for legislation like the one being considered in New York state, in which a provision of the Fashion Workers Act would require management companies and brands to obtain models’ clear written consent to create or use a model’s digital replica; specify the amount and duration of compensation, and prohibit altering or manipulating models’ digital replica without consent.

Alexsandrah says that with ethical use and the right legal regulations, AI might open up doors for more models of color like herself. She has let her clients know that she has an AI replica, and she funnels any inquires for its use through Wilson, who she describes as “somebody that I know, love, trust and is my friend.” Wilson says they make sure any compensation for Alexsandrah’s AI is comparable to what she would make in-person.

Edmond, however, is more of a purist: “We have this amazing Earth that we’re living on. And you have a person of every shade, every height, every size. Why not find that person and compensate that person?”
…