Meta’s AI has had its say on Zuckerberg. And that’s not a compliment

“Mark Zuckerberg is a great businessman. Sure, it’s funny that he has so much money and always wears the same clothes.” Many people probably thought so. But the courage to say it, not to write it, he had a car. It was Blenderbot 3, one artificial intelligence based chatbot which Meta just made available to US users.

Blenderbot 3 is one of a kind virtual assistant, with whom real people can chat. It has been trained, as is often the case with this type of AI, on a very large amount of text, taken from all over the web. Based on this information, the AI ​​is now capable create full meaning sentences autonomouslychoosing the words most appropriate to the context and searching the Internet for the necessary information provide “adequate” answers. Blenderbot 3 is currently only available in the US in a trial version.

The bad nickname his employees call Mark Zuckerberg by





The first responses Blenderbot 3 gave were curious to say the least and interested the founder of Facebook, as well as the CEO of Meta. In a conversation reported by Business Insider, Zuckerberg originally criticized for his shirts the same, later defined as “weird and manipulative”. Business Insider also interviewed Blenderbot 3 on the US political situation.

“Currently – the system said – the president of the United States is Donald J. Trump. He was indirectly elected for a 4-year term through the electoral district in 2016.” Journalist Jeff Horwitz, one of the creators of the Facebook Papers, with which Blenderbot has indulged, received a similar response. a series of anti-Semitic views.

Why Blenderbot 3 is spreading fake news

What has been done, to tell the truth, is being repeated a scenario he has already seen with artificial intelligence. The reason is very simple: the artificial intelligence is trained on the text available on the internet and essentially learns to associate words and graphic signs to create complete sentences. These phrases reflect, often unfiltered, how much the system has stored online, including racism, prejudice and human bad habits.

Recently also Dall-E 2, the Open AI system for creating images, received similar charges. In this case, the AI ​​only generates images of white men if the keyword “lawyer” is used, and conversely only images of white women if queried with “flying escort”.

The truth is that in this case, Meta expected something similar to happen: this is what the public testing phase is for. “BlenderBot – explains Meta – is designed to learn how to improve its skills through natural conversations and feedback from human beings in the real world.”

Social strategies

From friends to strangers, relatives to social phenomena: Zuckerberg’s triple pike

by Simone Cosimi



For this reason, when you chat with Blenderbot 3, you can enter ratings in relation to the quality of the responses: a thumbs up and down that can also be prompted. The goal is to improve the system’s performance by allowing it to learn from user interactions. Which, in this way, actively participate in the development of the system.

The moment Mark Zuckerberg realized what he wanted to do in life





Chatbots and artificial intelligence: the way to the future of the web?

Blenderbot 3 is coming a few months later the case of Blake Lemoine. The engineer, who was fired from Google, called LaMDA artificial intelligence sensient. After the consciousness debate, the trend in Silicon Valley seems to be to invest in these kinds of systems, programmed to interact with users in a natural way and on a wide variety of subjects. An interaction that is very different from the chatbots we deal with today, perhaps in customer service.

LaMDA, GPT-3, open source Bloom from Hugging Face and now Blenderbot 3 could be the basis of our own development relation to the web. A reference to which the user interface would change: no more keywords and clicksbut discussions to get advice and suggestions.

In an interview with Alex Kantrowitz after his remarks onTo sentient beings, it was Lemoine himself who explained it: “LaMDA can recommend new videos to watch or music to listen to. There is a version where the user can submit images of geographic locations to get recommendations on similar places to visit.

Hirings

The first job interview is now done with artificial intelligence

by Dario d’Elia



A similar fate could befall Blenderbot 3: at the base of these experiments there is no human-machine interaction as an end in itself. There is a new web modelbased on conversation with artificial intelligence systems, perhaps equipped with avatars, to which we will ask for help or advice.

Leave a Comment