Weekly AI Highlights Review: September 9–17

Recently, we've seen a welcome trend in which AI companies, especially those focused on the generative AI market, have faced increased scrutiny from governments, regulatory bodies, and the general public. Generally, the actions taken by legislation-enforcing bodies have been quite moderate. For instance, the Irish Data Protection Commission recently announced it would investigate whether Google should have performed a Data Protection Impact Assessment before collecting and processing European users' data to train PaLM 2.

PaLM 2 is an outdated LLM that has largely been replaced by the Gemini model family on Google's AI services, making it debatable whether this is the model DPC should be looking into. Many investigations into regulatory concerns committed by tech giants have remained moderate. Still, this says little about how things may shape up in the future and does show that tech giants are no longer completely off the hook regarding compliance. Indeed, in cases such as Meta's and OpenAI's, there seems to be an active effort to boost their outreach efforts, to increase the transparency of their approach to sensitive topics like safety and data privacy, and as an opportunity to reshape the prevalent narrative.

OpenAI recently shared a blog post with updates about its Safety and Security Committee. Most notably, the committee is becoming an “independent” board oversight group in which Sam Altman will no longer participate. Instead, the Safety and Security Committee is now chaired by Zico Kolter, Director of the Machine Learning Department at the School of Computer Science, Carnegie Mellon University. The committee includes OpenAI's board of directors members Adam D’Angelo, Paul Nakasone, and Nicole Seligman as participants. The newly reorganized committee reviewed the criteria OpenAI used to determine whether the new o1 was ready to launch. Moreover, following a 90-day review of the company's safety and security processes, the committee made some recommendations, which OpenAI has agreed to incorporate.

Meta, which became the subject of public criticism after edited photographs were labeled as "made with AI", announced further changes to its AI-generated or modified content labeling strategy. Facebook and Instagram posts containing content modified or edited with AI-powered tools will no longer prominently feature the "AI Info" label. Instead, AI-related information for these kinds of content will appear on the post's menu. Meta also published a blog post announcing it will resume training AI models on UK-based users' public data after the ICO supported its "legitimate interests" justification. The company will roll out in-app notifications for eligible users, explaining what it plans to do, and how users can opt out. Meta also stated it will make its Objection Form easier to find, and will honor all submitted and future users' objections.

Other notable highlights this week include:

The SAG-AFTRA Video Game Strike scored two major victories this week: Games studio Lightspeed L.A. and 80 video games have agreed to SAG-AFTRA's tiered and interim agreements including AI protections for talent, demonstrating that the union's proposed terms are reasonable and achievable.

Capsa AI secured £1.7M to transform private equity due diligence with AI: Capsa AI, a startup founded less than a year ago, has raised £1.7m in seed funding to expand its AI-powered platform that streamlines due diligence processes for private equity firms. The Capsa AI platform has demonstrated significant time savings for early adopters.

AI-powered finances and bookkeeping platform finally raised a $200M Series B: AI-powered bookkeeping startup finally has raised $200 million in funding to expand its all-in-one financial management platform for small and mid-sized businesses.

CardiaTec, a company building a heart tissue multi-nomics dataset, has raised $6.5M: CardiaTec, a company using computational methods to study cardiovascular disease, has raised $6.5 million in seed funding to build the world's first proprietary human heart tissue multi-omics dataset and advance its drug development pipeline.

Palantir and BP announce a 5-year strategic partnership: Palantir and BP have extended their decade-long collaboration with a new five-year agreement to enhance BP's oil and gas production operations with the safe deployment of large language models, to optimize performance and support decision-making across BP's global assets.

Smartcat raised a $43M Series C to expand its AI-powered translation services platform: Enterprise language AI platform Smartcat has raised $43 million in Series C funding to expand its AI-powered translation and multilingual content generation solutions, which allow global businesses to create ten times more content in multiple languages faster and more cost-effectively.

PLATMA raised $2M to accelerate its "Extreme No-Code" development platform: Ukraine-founded startup PLATMA has raised $2 million in pre-seed funding to expand its AI-powered no-code platform, which converts natural language into functional code, aiming to revolutionize digital transformation for businesses of all sizes.

A new NotebookLM feature lets users listen to conversations based on their sources: Google's NotebookLM, an AI research assistant powered by Gemini 1.5 Pro, has introduced a new Audio Overviews feature that converts uploaded documents into AI-generated audio discussions between two speakers.

Pixtral 12B is Mistral AI's first open-source multimodal model: Mistral AI has released Pixtral 12B, its first multimodal model. Pixtral 12B offers competitive performance in image and text understanding tasks, features a 128k context window, and is available under an Apache 2.0 license.

Tem secured $13.7M to disrupt the renewable energy market with AI: London-based startup Tem has raised $13.7 million in Series A funding to expand its AI-powered marketplace that directly connects businesses with renewable energy sources, potentially reducing costs by up to 25% and disrupting the traditional utility model.

Purple Transform raised £4.5 million to make railway lines safer with the help of AI: AI data analytics startup Purple Transform has raised £4.5 million to expand its AI-powered railway safety system. The system has shown significant results in incident reduction. As a result, Purple Transform will leverage the funds to expand its operations.

Osavul raised €2.78M ($3M) to bring its information security platform to the private sector: Osavul, a Kyiv-based media intelligence company founded in response to the Russian Invasion and using AI to counter information warfare, has secured €2.78M in funding from 42CAP, u.ventures, and SMRK to expand globally, develop its algorithms, and offer B2B solutions.

The White House secured AI-industry voluntary commitments against image-based sexual abuse: The Biden-Harris Administration has secured voluntary commitments from major AI companies to combat AI-generated image-based sexual abuse through responsible data sourcing, safeguards in development processes, and dataset modifications.

World Labs has emerged from stealth with $230 million to unlock spatial reasoning for AI: Fei-Fei Li's AI startup World Labs, backed by $230 million in funding and reportedly valued at over $1 billion, is developing large world models to enable AI systems to reason about the physical world, with applications in design, gaming, visual effects, and potentially robotics.