This Week in AI: May 27

For better or worse, OpenAI made sure it stole the week's spotlight in the aftermath of Scarlett Johansson's threat of legal action after the actress expressed concerns that one of ChatGPT's voices was eerily reminiscent of hers, recorded and showcased after Johansson declined to lend her voice to the assistant (more than once) and introduced to the world with a clear reference to the movie Her, famously claimed as Altman's favorite, and one in which Johansson voices an operating system designed to cater to its user's every need.

Somewhat tellingly, OpenAI agreed to pause the voice. Still, the damage had already been done: the incident has spurred a much wider conversation about the status of a person's likeness and voice within intellectual property and whether existing protections suffice to mitigate and prevent the abuse or misuse of people's appearances and voices. The incident was also a stark reminder that much of generative AI is based on stolen work and that still, AI firms seem to have little to no consideration for upholding copyright and intellectual rights protections. One of the few exceptions to this pattern appears to be the ever-growing collection of partnerships with media and news outlets; on that front, The Atlantic and Vox Media are OpenAI's pick of the week.

Another recent area of constant disregard seems to be security and alignment. After announcing a resignation he attributed to constant disagreements about OpenAI prioritizing profit over safety, Jan Leike announced that he would join rival firm Anthropic as an alignment team lead. From what little Leike shared, it seems the new team will be doing the kind of work originally expected from OpenAI's now-defunct Superalignment team.

Leike's announcement contrasted starkly with one of OpenAI's, where the company announced the creation of a Safety and Security Committee comprised entirely of insiders. It is hard not to conceive of this last announcement as (perhaps inadvertently) fanning the flames of the ongoing discussion about the Sky voice incident. At the very least, the commentary on the Safety and Security Committee complements the one about intellectual property nicely in that safety and rights protection are topics that tech giants seem to not care much for. As evidence, consider that the Center for Countering Digital Hate revealed how leading AI voice cloning tools can be easily manipulated to generate disinformation mimicking political leaders' voices, with an 80% effectiveness rate.

Oddly enough, OpenAI's shenanigans last week finished with mostly positive ChatGPT Edu and OpenAI for Nonprofits launches, two initiatives aimed at bringing OpenAI's offerings to non-profits and universities at an affordable price.

Other noteworthy developments include:

xAI has raised $6B to take on its competition: Musk's AI startup, xAI, recently announced the successful closing of a Series B at a $18 million pre-money valuation. The company plans to launch its first products to market, build on its infrastructure, and accelerate its R&D efforts.

Codestral is Mistral AI's first model for code-generation tasks: Mistral AI recently launched Codestral, a 22B-parameter model focused on helping developers with code-generation tasks, and shown to outperform models with more significant system requirements, including CodeLlama 70B, Llama 3 70B, and DeepSeek Coder 33B.

Microsoft has launched an official Copilot Telegram bot: Microsoft has launched its AI assistant Copilot, powered by GPT models and Bing Search, as a Telegram bot available under the username @CopilotOfficialBot, expanding the Copilot experience to third-party applications.

Exactly.ai raised $4.3M to empower visual artists with customized algorithms: Exactly.ai is a startup that allows artists to train AI models on their artwork, enabling them to retain copyright, generate new pieces in their distinct style, and monetize their work. The company recently raised $4.3 million to expand its artist-centric approach to AI image generation.

Perplexity Pages lets users unleash their creativity and share their research: Perplexity AI has introduced Perplexity Pages, a feature that generates visually appealing, shareable presentations on any topic based on user prompts. Users can also convert existing conversation threads into pages and have editing capabilities over the AI-generated content and visuals.

Function calling is now generally available for Claude 3 models: Anthropic has made its Claude 3 models even more capable by enabling tool use, which allows them to interact with external tools, APIs, and databases to perform dynamic tasks. Anthropic also launched features like streaming, forced tool selection, and debugging aid to support developers.

Amazon is celebrating the Fire TV device's 10th birthday with AI-powered voice search: To celebrate Fire TV's 10th birthday, Amazon published a recap on the AI Art feature, boasting over a million generations, and launched an AI-enhanced voice search providing personalized entertainment recommendations based on natural language queries.

Google addressed last week's AI Overviews oddities: Google has outlined the steps it took to improve the AI Overviews feature, including better detection of nonsensical queries, reducing satirical content, restricting advice from forums, adding safeguards for sensitive topics, and quickly addressing issues.

Faircado raised €3M to drive the circular economy with an AI-powered browser extension: Faircado, a Berlin-based startup with an AI-powered browser extension that helps consumers find second-hand alternatives while shopping online, has raised €3 million in pre-seed funding to scale its circular economy solution across Europe, grow its team, and enhance its technology.

Iris.ai secured €7.64M to optimize its AI engine for scientific text understanding: Iris.ai, the startup behind an AI engine that can provide high-quality insights from scientific research publications, raised €7.64 million to enhance its flagship RSpace platform's factual accuracy capabilities.

ElevenLabs launched a text-to-sound effects AI audio model: ElevenLabs has launched a groundbreaking new Text-to-Sound Effects feature that leverages AI to allow users to generate a wide variety of audio content like sound effects, instrumental tracks, and character voices simply from text prompts.

Intel, Google, Microsoft, and Meta are advocates for a new AI chip interconnect standard: Major tech companies like Intel, Google, Microsoft, Meta, and AMD have formed the Ultra Accelerator Link (UALink) Promoter Group to develop an open industry standard for connecting the AI accelerator chips used in data centers, aiming to reduce reliance on NVIDIA as demand for AI hardware soars.

EthonAI raised a $16.7M Series A to enhance its manufacturing analytics platform: EthonAI, a Swiss startup pioneering AI-powered manufacturing analytics, has raised $17 million in Series A funding to scale its breakthrough Manufacturing Analytics System.

You.com wants to put users in control with its Custom Assistants launch: You.com has launched a groundbreaking feature called Custom Assistants, which allows users to create personalized AI assistants tailored to their specific needs by selecting and fine-tuning leading large language models like GPT-4, Claude, and LLaMA through a simple interface.