October was filled with new debates on AI policies, a halt to one of the most significant AI safety bills in the United States, and a host of new AI developments in video, music, and even podcasts. Keep reading for our highlights below:

California Gov. Gavin Newsom Vetoes AI Safety Bill

In a blow to the progress of AI regulation, California Governor Gavin Newsom vetoed SB1047, the controversial and closely watched bill on AI regulation. If passed, this bill would have been the first form of AI regulation by a government body in the United States establishing legal accountability and liability for misuse or harm caused by AI models. The bill would have required AI companies to conduct safety tests for large AI models and have built-in “kill switches” to protect against mass casualties or warfare.  

While supported by progressive lawmakers and even some members of the tech industry, the bill was also opposed by many firms in the AI space, commonly claiming that the bill would stifle innovation and was too broad-based and without situational nuance. 

Despite Newsom vetoing this bill, he did announce the intent to work with AI-experts to establish a set of guardrails for future regulatory purposes, as well as sign a handful of smaller AI-related bills into law. 

FTC Rules on AI Product Reviews

With the rise of AI came the rise of AI product reviews. In newly changed guidelines this month, the FTC has determined that these types of AI-generated reviews are now illegal. Broadly implicating any falsified review written either by a fake author (or AI) or written by an author or entity with no true experience with the product in question. The ruling also introduces a critical limitation on websites, preventing them from publishing reviews under the guise of an independent editorial piece when they have a vested interest in the success of the product reviewed. Affecting both human and AI agents, this ruling intends to help clean up a considerably polluted online marketplace and instill trust in reviews going forward in the new era of AI. 

Meta Launches Movie Gen

Meta launched one of its newest AI offerings, Movie Gen. Able to generate video with audio from simple text prompts, Meta has claimed that this collection of foundational models has outperformed other competitors like Runway and Gen3. Key to this offering is the introduction of a straightforward text-based editing feature that allows users to work with generated shots in order to keep certain aspects and easily adjust others without needing to regenerate a shot entirely. Audio, but not voice, is a significant part of the offering, too, with sounds able to be generated and matched to the video, and music even appearing when deemed appropriate. Trained on “a combination of licensed user content and publicly available datasets,” there is opacity surrounding where the model’s source material came from.

Adobe Content Authority

Adding to a growing arsenal of anti-AI features like Nightshade, Adobe Content Authority gives creators the power to attach a digital watermark to their products, which signals that they do not give approval for their work to be included in large language model training sets. While similar features have been launched across Adobe applications, this new authority allows users to attach such a mark to any piece of content regardless of whether it originated in an Adobe environment. 

AI Podcasts Hit the Scene

AMT-Lab and others in the podcast industry may be in trouble. The new AI offering, Notebook LM, was launched as part of Audio Overview by Google AI. Baked into a study-oriented AI system, Notebook LM helps to quickly give a high-level summary of an input link or file in the form of a conversational podcast. This extension is capable of generating believable, human text-to-speech output with a feature called Deep Dive. This not only applies speech outputs to text inputs, but also layers in human-coded interjections like “wow” and frames two voices in conversation as opposed to a single narrator. This effectively creates believable podcasts from simple text and file inputs. 

Squarespace Introduces AI Web-Design

Squarespace recently launched Design Intelligence, an AI feature meant to streamline the way websites are built and designed on their sites. Intended more as a playground than as a straight input-output model, Design Intelligence provides design outputs from a given prompt while allowing the user to eliminate unwanted outcomes and drill into, customize, and further edit the desired features. Intended to reduce rigidity present in the current landscape of AI-driven web-design, Squarespace is prioritizing user experience in this method of implementation, balancing the desire for ease and pre-made designs with the desire for bespoke outcomes. 

Note:  while our site is built on Squarespace, we have yet to try out this new AI offering.

Roli Airwave - AI Piano Teacher

Gone are the days of in-person piano lessons. Adding to the number of significant AI advances in musical training, like Google’s AI Duet and a range of AI tools developed by Yamaha, Roli has launched a new tool to enhance music education and creative potential with the piano: Roli Airwave. Airwave is an auxiliary device compatible with multiple Roli piano models which integrates multi-modal AI capabilities to track player’s hand movements, provide feedback, and respond to questions asked by the user. Intended to improve access to musical education, the tool is also capable of augmenting music creation as well, with the Airwave providing hand-motion features unlocking the ability for new sounds, effects, and layering possibilities capable of expanding the set of outputs a traditional piano may allow for.