This text belongs to 'Artificial', the AI newsletter that Delia Rodríguez sends every Friday.
Dear readers, everything is speeding up again, and it hasn't even been a year since ChatGPT 3.5 blew our minds. This has been an important week, dense with news.
On Monday OpenAI announced in a post that ChatGPT is now multimodal, that is, “it can now see, hear and speak.” Some of these functions are already available in the app, others will take a while, but this means that the chatbot is capable of interpreting images, chatting or giving oral instructions. At the moment it will use five types of voices. The possibilities become clearer with examples. Can:
- Find Wally in a picture
- Program a website based on the typical behavior diagram drawn on a blackboard
- Act as a therapist to have a talk with
- Do your math homework just by uploading a photo of the problems
- Tell you how to fix your bike by uploading its image
- Explain what you can do with the little you have in the fridge if you send him a photo
Actually, I have the feeling that this week we have been warned that we are going to have AI everywhere, embedded in hundreds of small uses, gestures or objects. Will be:
- On social networks and instant messaging. Meta has announced many new features about its integration into Facebook, Instagram or Whatsapp, for example to create images or stickers on the fly to liven up our conversations with friends or to chat with one of the avatars of celebrities with whom they have reached an agreement, such as Kendall Jenner, Mr Beast, Paris Hilton or Snoop Dogg.
- In glasses, both prescription and sunglasses. Meta has also released a second version of their Ray Bans, this time quite nice. They can play music, take calls and photos or start a live streaming. It has five microphones, two cameras and 32GB to store audiovisual material. When recording, a light comes on.
- In stock photos. Without stealing anyone's intellectual property, Getty has created a tool to generate artificial images from the warehouse of millions of stock photos he owns. His clients will only have to describe in text what they need. In addition to the usual license to use, Getty will provide coverage against possible legal problems, a decision similar to what Microsoft has made with some of its products. Getty has sued Stable Diffusion for using its material without permission.
- On a possible new mysterious device. SoftBank CEO Masayoshi Son, OpenAI founder Sam Altman and former Apple designer Jony Ive are in talks to create OpenAI AI hardware, according to The Information. There aren't many details about the project, or what type of device they are thinking about (a cell phone? an “Alexa”? something else?), but it doesn't sound strange. Ive and Altman are friends, and a designer from Ive's school already designed The Orb, Worldcoin's controversial retina scanner.
- In podcasts. Spotify has reached an agreement with the authors of some great podcasts and is going to start dubbing them into other languages by imitating the original voices. The technology is from ChatGPT.
- In mobile photographs. It's not official yet, but there are rumors that the new Pixel will incorporate AI tools capable of inventing photos that never existed. We are not talking about using a lot of imagination to improve photos of night skies, as Google phones already do now, but rather about recreating a group photo making everyone look good at the same time, when that moment never happened.
- In robots. Optimus, Tesla's robot, already knows how to organize objects and do yoga.
- In internet searches. ChatGPT once again has a connection to the network through Bing, so it will not use the excuse that it only has information until September 2021. OpenAI withdrew access a few months ago because it was bypassing the paywall of some media outlets. Now link to your sources and respect robot.txt (the file where websites indicate whether you have permission to pass through or not). Currently available for paying users, who are finding that it still fails quite a bit and that, for the moment, it does not replace Google.
Very important because it is the first great precedent for a collective agreement in an industry affected by AI: the Hollywood screenwriters' strike has ended after 148 days. They have managed to limit its use, it cannot “write or rewrite literary material” or be used to harm your rights. It is up to each individual to use ChatGPT-type tools in their work, but they cannot be forced or do so secretly. Studies must warn if any material they provide to the authors is of artificial origin. The union will decide whether or not to give up the rights to the scripts to feed AIs.
José Antonio Bayona, on AI: "The dangerous thing is not the tool but the use it is made of. In the creative field there are risks, but films really work by whether they have a soul or not, not by the algorithm, and that It is something that AI cannot replicate. In Hollywood we have been watching movies written according to a pattern for many years. But when you break the mold and create something really new and with soul, the box office explodes. You have to separate data and knowledge a lot, which "It is the processing of that data and that is where the soul comes in. And that will not be able to be replicated with a machine." By Astrid Meseguer.
By the way, a very interesting read in Wired (in English) where an author explains how for years he has been testing artificial intelligence for writing. In his experience, Chat GPT version 3 achieved exceptional, surprising results, getting him some of his best phrases. From that version on, the “adjustment” of the chatbot to be corporate and secure filled the results with clichés.
Amazon has finally made a move in artificial climbing and has made an investment of, attention, 4 billion in Anthropic, the “ethical” AI company founded by former OpenAI employees and creator of Claude. Google had already made a much smaller investment in the company in February. The agreement is reminiscent of Microsoft's with OpenAI. Anthropic will use the Amazon Web Services (AWS) cloud platform and its chips, and Amazon is allying itself with one of the leading companies in the sector.
A privacy hole in Bard, Google's chatbot, allowed private conversations to appear in search results. These were issues that had previously been shared with another person (and a link had been generated for it). They found out about it because an Australian SEO discovered it and posted it on X.
The viral application of the week is krea.ai, which generates attractive optical illusions by “hiding” texts, logos or other images in a background. Surely you have seen many tests in X. In Xataka.
Something a little strange for a simple troll. Sam Altman posted a comment on Reddit - something he hadn't done in years - saying that they had achieved artificial general intelligence (AGI) in-house, which is what we should really call artificial intelligence, not today's generative artificial intelligence. ). He soon edited the post saying that if they had gotten it they wouldn't announce it on Reddit.
Open AI is seeking a valuation of €90 billion, three times more than at the beginning of the year.
The Atlantic has published under its paywall a tool that searches if a work or author is among the 183,000 titles in the “Book3” pirated book database, used according to some claims to feed Meta's AI.
IAnxiety level this week: elevated by two details hidden in this newsletter, perhaps not the most important. The use of ChatGPT as a private therapist and his skill in writing fiction in an earlier version.
Photo: The United States Copyright Office has ruled that the famous work “Théâtre d'Opéra Spatial”, created by AI Midjourney, could not be copyrighted because it was not created by humans.
