Month: February 2023

Microsoft is adding support for Bing Chat and the other “new Bing” features to the Windows taskbar as part of 2023’s first major Windows 11 feature update. Microsoft Chief Product Officer Panos Panay announced the updates in a blog post released today.

The Windows update doesn’t open the new Bing preview to anyone who hasn’t already signed up for it, and there’s currently a waitlist for new users who want to try the feature. But if and when Microsoft expands the Bing preview to more of its users, millions of PCs that automatically install today’s update will already have built-in support for it.

You can read about the other changes in the new Windows 11 updates here. Anyone running the Windows 11 2022 updates can download them manually via Windows Update starting today, and all of the new changes will roll out to those PCs automatically in March.

New Bing’s ups and downs

If you subscribe to the maxim that all press is good press, then Bing has probably had its best month of publicity in its entire 13-year history.

Bing Chat, based on OpenAI’s underlying ChatGPT technology, has only been available to the public for a few weeks, and the results have oscillated between “entertaining” and “concerning.” Although the bot can convey a surprising and useful range of responses to user queries, Bing Chat has also been shown to spread misinformation; spill secrets about its inner workings and then insist that it didn’t; get weirdly morose and/or belligerent during extended chat sessions; and threaten Ars Technica reporters.

Rather than ending the public preview—as happened when the Internet broke Tay, a previous attempt at an AI chatbot—Microsoft is attempting to address Bing Chat’s problems in public before opening it to a wider audience. Recent fixes include limiting the bot to five or fewer responses in a single chat instance and making it stop talking about itself and its simulated “feelings.”

In the meantime, the company is working to integrate the new Bing and other AI features into more of its products. That started with the Microsoft Edge browser earlier this month and was later expanded to Skype and now Windows 11. In the near future, Microsoft is also reportedly planning to add ChatGPT-powered features to apps like Word and Outlook to aid with document and email creation (imagine a supercharged Clippy that will write that cover letter for you).

Microsoft will likely continue to research and implement new AI-based features thanks to a “multi-year, multi-billion dollar investment” in OpenAI, which in addition to ChatGPT is also responsible for the DALL-E image generator, the Whisper speech recognition system, and other projects. The moves have reportedly caused “panic” inside Google, which sees AI-powered chat as a threat to its core search business. Google’s own AI chatbot, codenamed “Bard,” was unveiled earlier this month via an unimpressive onstage demo, plus a since-pulled advertisement where Bard offered an untrue answer to the only question that was asked of it.

Listing image by Microsoft

On Tuesday, the US Copyright Office declared that images created using the AI-powered Midjourney image generator for the comic book Zarya of the Dawn should not have been granted copyright protection, and the images’ copyright protection will be revoked.

In a letter addressed to the attorney of author Kris Kashtanova obtained by Ars Technica, the office cites “incomplete information” in the original copyright registration as the reason it plans to cancel the original registration and issue a new one excluding protection for the AI-generated images. Instead, the new registration will cover only the text of the work and the arrangement of images and text. Originally, Kashtanova did not disclose that the images were created by an AI model.

“We conclude that Ms. Kashtanova is the author of the Work’s text as well as the selection, coordination, and arrangement of the Work’s written and visual elements,” reads the copyright letter. “That authorship is protected by copyright. However, as discussed below, the images in the Work that were generated by the Midjourney technology are not the product of human authorship.”

Last September, in a story that first appeared on Ars Technica, Kashtanova publicly announced that Zarya of the Dawn, which includes comic-style illustrations generated from prompts using the latent diffusion AI process, had been granted copyright registration. At the time, we considered it a precedent-setting case for registering artwork created by latent diffusion.

However, as the letter explains, after the Copyright Office learned that the work included AI-generated images through Kashtanova’s social media posts, it issued a notice to Kashtanova in October stating that it intended to cancel the registration unless she provided additional information showing why the registration should not be canceled. Kashtanova’s attorney responded to the letter in November with an argument that Kashtanova authored every aspect of the work, with Midjourney serving merely as an assistive tool.

That argument wasn’t good enough for the Copyright Office, which describes in detail why it believes AI-generated artwork should not be granted copyright protection. In a key excerpt provided below, the Office emphasizes the images’ machine-generated origins:

Based on the record before it, the Office concludes that the images generated by Midjourney contained within the Work are not original works of authorship protected by copyright. See COMPENDIUM (THIRD ) § 313.2 (explaining that “the Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author”). Though she claims to have “guided” the structure and content of each image, the process described in the Kashtanova Letter makes clear that it was Midjourney—not Kashtanova—that originated the “traditional elements of authorship” in the images.

The letter provides additional analogies for understanding why the Copyright Office thinks Kashtanova is not the creator of the images, including the idea of hiring a human to create images using descriptions and performing a text-based image search on the Internet. The overall argument in the letter may serve as an important legal precedent for future attempts to copyright AI-generated images.

In an Instagram post, Kashtanova reacted to the letter by framing it as an overall win for AI-augmented artists. She says the ruling is “great news” in the sense that it protects the comic book’s story and the image arrangement, which “covers a lot of uses for the people in the AI art community.”

But on the issue of losing copyright protection for the individual images, Kashtanova says she is not giving up the fight:

I was disappointed in one aspect of the decision. The Copyright Office didn’t agree to recognize my copyright of the individual images. I think that they didn’t understand some of the technology so it led to a wrong decision. It is fundamental to understand that the output of a Generative AI model depends directly on the creative input of the artist and is not random. My lawyers are looking at our options to further explain to the Copyright Office how individual images produced by Midjourney are [a] direct expression of my creativity and therefore copyrightable.

Despite precedents for earlier algorithmically generated artwork receiving copyright protection, this ruling means that AI-generated imagery, without human-authored elements, cannot currently be copyrighted in the United States. The Copyright Office’s ruling on the matter will likely hold unless it’s challenged in court, revised by law, or re-examined in the future.

It’s possible that the ruling may eventually be reconsidered as the result of a cultural shift in how society perceives AI-generated art—one that may allow for a new interpretation by different members of the US Copyright Office in the decade ahead. For now, AI-powered artwork is still a novel and poorly understood technology, but it may eventually become the standard way visual arts emerge. Not allowing for copyright protection would potentially preclude its use by large and powerful media conglomerates in the future. So the story of AI and copyrights is not over yet.

On Friday, Meta announced a new AI-powered large language model (LLM) called LLaMA-13B that it claims can outperform OpenAI’s GPT-3 model despite being “10x smaller.” Smaller-sized AI models could lead to running ChatGPT-style language assistants locally on devices such as PCs and smartphones. It’s part of a new family of language models called “Large Language Model Meta AI,” or LLAMA for short.

The LLaMA collection of language models range from 7 billion to 65 billion parameters in size. By comparison, OpenAI’s GPT-3 model—the foundational model behind ChatGPT—has 175 billion parameters.

Meta trained its LLaMA models using publicly available datasets, such as Common Crawl, Wikipedia, and C4, which means the firm can potentially release the model and the weights open source. That’s a dramatic new development in an industry where, up until now, the Big Tech players in the AI race have kept their most powerful AI technology to themselves.

“Unlike Chinchilla, PaLM, or GPT-3, we only use datasets publicly available, making our work compatible with open-sourcing and reproducible, while most existing models rely on data which is either not publicly available or undocumented,” tweeted project member Guillaume Lample.

Meta calls its LLaMA models “foundational models,” which means the firm intends the models to form the basis of future, more-refined AI models built off the technology, similar to how OpenAI built ChatGPT from a foundation of GPT-3. The company hopes that LLaMA will be useful in natural language research and potentially power applications such as “question answering, natural language understanding or reading comprehension, understanding capabilities and limitations of current language models.”

While the top-of-the-line LLaMA model (LLaMA-65B, with 65 billion parameters) goes toe-to-toe with similar offerings from competing AI labs DeepMind, Google, and OpenAI, arguably the most interesting development comes from the LLaMA-13B model, which, as previously mentioned, can reportedly outperform GPT-3 while running on a single GPU when measured across eight standard “common sense reasoning” benchmarks such as BoolQ, PIQA, SIQA, HellaSwag, WinoGrande, ARC, and OpenBookQA. Unlike the data center requirements for GPT-3 derivatives, LLaMA-13B opens the door for ChatGPT-like performance on consumer-level hardware in the near future.

Parameter size is a big deal in AI. A parameter is a variable that a machine-learning model uses to make predictions or classifications based on input data. The number of parameters in a language model is a key factor in its performance, with larger models generally capable of handling more complex tasks and producing more coherent output. More parameters take up more space, however, and require more computing resources to run. So if a model can achieve the same results as another model with fewer parameters, it represents a significant gain in efficiency.

“I’m now thinking that we will be running language models with a sizable portion of the capabilities of ChatGPT on our own (top of the range) mobile phones and laptops within a year or two,” wrote independent AI researcher Simon Willison in a Mastodon thread analyzing the impact of Meta’s new AI models.

Currently, a stripped-down version of LLaMA is available on GitHub. To receive the full code and weights (the “learned” training data in a neural network), Meta provides a form where interested researchers can request access. Meta has not announced plans for a wider release of the model and weights at this time.

Update (February 26, 2023): We have added the names of the standard academic benchmarks that Meta used to measure the performance of LLaMA with.