Enlarge / An artist's impression of a robot hand using a desktop calculator. (credit: Aurich Lawson | Getty Images )

Language models like ChatGPT have revolutionized the field of natural language processing, but they still struggle with some basic tasks such as arithmetic and fact-checking. Last Thursday, researchers from Meta revealed Toolformer , an AI language model that can teach itself to use external tools such as search engines, calculators, and calendars without sacrificing its core language modeling abilities.

The key to Toolformer is that it can use APIs (application programming interfaces), which are a set of protocols that allow different applications to communicate with one another, often in a seamless and automated manner. During training, researchers gave Toolformer a small set of human-written examples demonstrating how each API is used and then allowed it to annotate a large language modeling dataset with potential API calls. It did this in a "self-supervised" way, meaning that it could learn without needing explicit human guidance.

The model learned to predict each text-based API call as if they were any other form of text. When in operation—generating text as the result of a human input—it can insert the calls when needed. Moreover, Toolformer can "decide" for itself which tool to use for the proper context and how to use it.

Enlarge / AI image synthesis advances in 2022 have made images like this one possible, which was created using Stable Diffusion, enhanced with GFPGAN, expanded with DALL-E, and then manually composited together. (credit: Benj Edwards / Ars Technica)

More than once this year, AI experts have repeated a familiar refrain: "Please slow down." AI news in 2022 has been rapid-fire and relentless; the moment you knew where things currently stood in AI, a new paper or discovery would make that understanding obsolete.

In 2022, we arguably hit the knee of the curve when it came to generative AI that can produce creative works made up of text, images, audio, and video. This year, deep-learning AI emerged from a decade of research and began making its way into commercial applications, allowing millions of people to try out the tech for the first time. AI creations inspired wonder, created controversies, prompted existential crises, and turned heads.

Here's a look back at the seven biggest AI news stories of the year. It was hard to choose only seven, but if we didn't cut it off somewhere, we'd still be writing about this year's events well into 2023 and beyond.

Enlarge / A screenshot of an online game of Diplomacy , including a running chat dialog, provided by a Cicero researcher. (credit: Meta AI )

On Tuesday, Meta AI announced the development of Cicero, which it clams is the first AI to achieve human-level performance in the strategic board game Diplomacy . It's a notable achievement because the game requires deep interpersonal negotiation skills, which implies that Cicero has obtained a certain mastery of language necessary to win the game.

Even before Deep Blue beat Garry Kasparov at chess in 1997 , board games were a useful measure of AI achievement. In 2015, another barrier fell when AlphaGo defeated Go master Lee Sedol. Both of those games follow a relatively clear set of analytical rules (although Go's rules are typically simplified for computer AI).

But with Diplomacy, a large portion of the gameplay involves social skills. Players must show empathy, use natural language, and build relationships to win—a difficult task for a computer player. With this in mind, Meta asked, "Can we build more effective and flexible agents that can use language to negotiate, persuade, and work with people to achieve strategic goals similar to the way humans do?"

Enlarge / An AI-generated illustration of robots making science. (credit: Ars Technica)

On Tuesday, Meta AI unveiled a demo of Galactica, a large language model designed to "store, combine and reason about scientific knowledge." While intended to accelerate writing scientific literature, adversarial users running tests found it could also generate realistic nonsense . After several days of ethical criticism , Meta took the demo offline, reports MIT Technology Review.

Large language models (LLMs), such as OpenAI's GPT-3 , learn to write text by studying millions of examples and understanding the statistical relationships between words. As a result, they can author convincing-sounding documents, but those works can also be riddled with falsehoods and potentially harmful stereotypes. Some critics call LLMs " stochastic parrots " for their ability to convincingly spit out text without understanding its meaning.

Enter Galactica, an LLM aimed at writing scientific literature. Its authors trained Galactica on "a large and curated corpus of humanity’s scientific knowledge," including over 48 million papers, textbooks and lecture notes, scientific websites, and encyclopedias. According to Galactica's paper , Meta AI researchers believed this purported high-quality data would lead to high-quality output.

Enlarge / An illustrated depiction of data in an audio wave. (credit: Meta AI)

Last week, Meta announced an AI-powered audio compression method called "EnCodec" that can reportedly compress audio 10 times smaller than the MP3 format at 64kbps with no loss in quality. Meta says this technique could dramatically improve the sound quality of speech on low-bandwidth connections, such as phone calls in areas with spotty service. The technique also works for music.

Meta debuted the technology on October 25 in a paper titled " High Fidelity Neural Audio Compression ," authored by Meta AI researchers Alexandre Défossez , Jade Copet, Gabriel Synnaeve, and Yossi Adi. Meta also summarized the research on its blog devoted to EnCodec.

Meta claims its new audio encoder/decoder can compress audio 10x smaller than MP3. (credit: Meta AI)

Meta describes its method as a three-part system trained to compress audio to a desired target size. First, the encoder transforms uncompressed data into a lower frame rate "latent space" representation. The "quantizer" then compresses the representation to the target size while keeping track of the most important information that will later be used to rebuild the original signal. (This compressed signal is what gets sent through a network or saved to disk.) Finally, the decoder turns the compressed data back into audio in real time using a neural network on a single CPU.

Enlarge / Still image from an AI-generated video of a teddy bear painting a portrait. (credit: Meta )

Today, Meta announced Make-A-Video , an AI-powered video generator that can create novel video content from text or image prompts, similar to existing image synthesis tools like DALL-E and Stable Diffusion . It can also make variations of existing videos, though it's not yet available for public use.

On Make-A-Video's announcement page, Meta shows example videos generated from text, including "a young couple walking in heavy rain" and "a teddy bear painting a portrait." It also showcases Make-A-Video's ability to take a static source image and animate it. For example, a still photo of a sea turtle, once processed through the AI model, can appear to be swimming.

The key technology behind Make-A-Video—and why it has arrived sooner than some experts anticipated—is that it builds off existing work with text-to-image synthesis used with image generators like OpenAI's DALL-E. In July, Meta announced its own text-to-image AI model called Make-A-Scene .