February 2024
This report was published by 451 Research on January 30, 2024.
Introduction
It has been a quick start to the year for generative AI, with a flurry of product and funding announcements to start 2024. Technology event CES 2024 provided an opportunity to showcase how developments in generative AI have started to filter into consumer technology products. Several funding announcements have been made, and while M&A is still nascent in this space, deals have taken place involving Qlik and Typeface. Meanwhile, legal and regulatory issues were prominent, with The New York Times filing a lawsuit against Microsoft Corp. and OpenAI for copyright infringement. Potential indemnity issues around third-party and fine-tuned models are likely to be a challenge this year, as are experiments with pricing models. Additionally, AI was a dominant theme at this year's week-long wintery World Economic Forum Annual Meeting at Davos in Switzerland.
The Take
The stage is set for a rapid progression of generative AI into robotics and virtual assistants over 2024. Google DeepMind opened the year with white papers on three new capabilities, and startup Figure AI demonstrated the ability for its humanoid robot to learn how to make coffee after training on video. The conversational performance of large language models in the past year has been compared favorably to the experience with early voice activated assistants, such as Alexa. It is perhaps unsurprising then that these assistants are being imbued with generative AI, with announcements surrounding not just Alexa but also in-car assistance over the past few weeks. These assistants may prove a valuable installed base for generative AI features in homes or offices. We are starting to see early indications of just how much more consumers and enterprises are prepared to pay for generative AI features on top of existing software licenses.
Product releases and updates
OpenAI formally rolled out its GPT store in early January. The GPT store enables users that took advantage of November 2023's new controls for GPT tailoring, to monetize the models they have developed. Full details on the payments around the marketplace have still not been disclosed, but OpenAI says initial payments in the US would be based on user engagement.
There were several generative AI announcements made at CES, a prominent consumer technology event. These were not limited to traditional technology providers, with Volkswagen AG following Mercedes-Benz Group AG in integrating ChatGPT into its vehicles, as an upgrade to its IDA voice assistant. In another update, Amazon.com Inc.'s Alexa announced a number of generative AI "skills," including a skill designed by Character.AI that allows a user to converse with chatbots imbued by a character — for example, a historical figure or a fitness coach.
Significant attention was paid to Rabbit R1, a handheld device hosting a proprietary AI-powered operating system, at CES. The company claims it has developed a new type of foundation model trained on videos and screenshots of applications. Known as a large action model, the strategy is to establish personalized LAMs that learn to replicate user actions across interfaces. Designed as an additional device alongside a customer's phone, rather than a replacement, the design concept was to remove the need to shift between multiple apps and allow a user to issue instructions in natural language. Rabbit announced a $10 million series A round in December 2023, following a $20 million seed round in October 2023.
Humanoid robotics startup Figure AI says that, using video footage of people making coffee, it has successfully trained a robot to use a Keurig coffee machine. These capabilities are part of a wider drive in robotics to use neural networks, in addition to large volumes of video and image data, to enable machines to engage in net new actions. Figure AI also announced a commercial agreement with BMW Group in January, with its humanoid robots to be deployed in a manufacturing facility in South Carolina.
Google DeepMind publicly released papers on three new robotics systems — AutoRT, SARA-RT and RT-Trajectory. AutoRT combines robotics control models (such as RT-2) and foundation models, more specifically LLM and visual language models. AutoRT uses a VLM to analyze its surroundings and an LLM to suggest the tasks the robot should carry out. In tests the system successfully orchestrated actions of 20 robots simultaneously. Self-Adaptive Robust Attention for Robotics Transformers is designed to improve the performance and efficiency of robotics control models, to boost the speed with which robots can make decisions. RT-Trajectory can be applied to videos, and overlays them with 2D trajectory sketches to improve training performance.
GitLab says it will be charging $19 per user per month for its GitLab Duo Pro add-on, which includes previously announced Code Suggestions for code completion and generation, with additional assignment controls to ensure only approved users can enable AI within their workflows. Duo Pro will be available through all GitLab deployment options in February.
Mark Zuckerberg said Meta Platforms Inc. will merge two AI research groups — FAIR and GenAI — and will pursue artificial general intelligence and open source it "to make it as widely available as we responsibly can." He expects that Meta will have about 350,000 NVIDIA H100 GPUs by year-end and overall 600,000 equivalents when other types of GPU are included from NVIDIA and other suppliers.
Funding and M&A
The assets of Kyndi, which develops a generative AI answer engine, were acquired by Qlik. Qlik is positioning the acquisition of patents and technology as providing a way for its customers to better interpret unstructured data, as a complement to its existing structured data querying capabilities. Kyndi went through a few evolutions of its strategy in its 10 years — from explainable AI to natural language processing engines to use-case-specific implementations of NLP. It had raised $47 million in three rounds and counted Intel Capital and PivotNorth Capital among its investors. Qlik has been owned by Thoma Bravo since 2016.
Typeface, which positions itself as a "personalized enterprise generative AI platform," announced the acquisition of TensorTour alongside the general availability of Typeface Hub. Few details have been released about the transaction, but TensorTour co-founder and CEO Gaurav Sharma appears to have taken a position as head of AI research at Typeface. TensorTour had positioned itself as a provider of text, audio and visual AI technologies, and Typeface cited the company's "domain-specific models and deep expertise in multimedia AI content" in its announcement.
Perplexity AI has raised $73.6 million in a series B round of funding. Notable investors included NVIDIA Corp.; Amazon founder Jeff Bezos, who invested through Bezos Expeditions; and returning investors Databricks Ventures. The generative AI question and answer tool provider has achieved $155.5 million in funding since its August 2022 incorporation and disclosed a post-money valuation of $520 million.
Korean startup Upstage, which has developed a range of LLMs, has completed a $19 million round. Investor SK Networks suggests it will help Upstage expand internationally. Upstage has largely focused on the application of LLMs for search and document processing use cases.
Synthetic voice startup ElevenLabs has received $80 million in a series B round. Co-led by returning investor Andreessen Horowitz, the round brought it to $101 million in funding, with a $1.1 billion post-money valuation.
Robin AI, a UK-headquartered startup offering generative AI for legal use cases, received $26 million in series B funding. The round was led by Temasek Holdings, with Quantum Light Capital, AFG Partners and returning investor Plural Platform participating.
Luma AI, developers of multimodal AI including a text-to-3D generation model, announced $42 million in funding, raising at a post-money valuation between $200 million and $300 million. Alongside the round, the company announced a new investor, Andreessen Horowitz. The company has raised $72 million in total.
Generative AI design tool Recraft announced a $12 million series A round led by Khosla Ventures and former GitHub CEO Nat Friedman. Recraft offers what it terms as an "AI toolbox" with capabilities such as text-to-image, image-to-image and image enhancement.
The developers of Miko, a family of companion bots for children, RN Chidakashi Technologies announced a $973,000 mature round in January. This brings total funding to just under $66 million since its 2015 incorporation, with the latest transaction including participation from new investor Alteria Capital Advisors. The company claims its conversational robot is powered by "deep learning," with its Miko Mini robot released in September 2023 incorporating ChatGPT.
Politics and regulations
The New York Times filed a lawsuit against OpenAI and Microsoft, alleging that the companies violated its intellectual property rights by using its journalistic content to train AI language models, specifically ChatGPT. OpenAI has responded, stating that it disagrees with the lawsuit and that it considers the training of AI models using copyrighted material as fair use. The company also mentioned that it is working to minimize the occurrence of "regurgitation," which it describes as a rare issue. OpenAI's position is that it collaborates with news organizations and is creating new opportunities, but it provides an opt-out for content usage. The lawsuit is ongoing, and both parties are expected to present their arguments in court. The case is significant as the two companies were in the process of negotiating, and it raises questions about the fair use of copyrighted material in training AI models and the potential impact of AI-generated content on traditional media organizations.
An analysis of hyperscaler indemnity clauses surrounding generative AI suggests the scope of the protections are more limited than is commonly assumed by customers or prospects. AWS, Google and Microsoft offer protections around copyright lawsuits for customers of their generative AI services, but an analysis by the Financial Times reveals these protections have a number of limitations. According to the report, hyperscaler indemnifications do not extend to third-party models available through their AI development platforms, excluding Microsoft's protections around partner OpenAI. In addition, the report suggests that Google's indemnity does not extend to fine-tuned models, and that Amazon voids protections if copyright infringement could be traced back to fine-tuning.
The International Monetary Fund (IMF) released a new report indicating that AI is projected to impact nearly 40% of all jobs globally. The report suggests that in most scenarios, AI is likely to exacerbate overall inequality, with the potential to affect up to 60% of jobs in advanced economies. While about half of the impacted jobs in these economies may benefit from AI integration, leading to increased productivity, the other half could face reduced labor demand, potentially resulting in lower wages and decreased hiring. The IMF also highlighted that AI's impact on jobs is expected to be less pronounced in low-income countries, where it may affect approximately 26% of jobs.
SAP SE announced a restructuring that it says will cost $2.2 billion and impact about 8,000 employees as it focuses on AI-driven opportunities. The money will be spent on retraining employees with AI skills or voluntary redundancy programs, but it expects to end 2024 with a headcount similar to the start of the year. As illustrated below, respondents to 451 Research's VotE: AI & Machine Learning, Use Cases 2024 suggest that explicit headcount reduction metrics were uncommon at their organizations. However, employee productivity improvements, which was measured by 44% of organizations, is commonly used to assess AI project performance.
Figure 1: Headcount reduction rarely an explicit AI project KPI
Figure 1: Headcount reduction rarely an explicit AI project KPI
Source: 451 Research's Voice of the Enterprise: AI & Machine Learning, Use Cases 2024.
Q. Which of the following key performance indicators does your organization track for AI initiatives? Please select all that apply.
Base: Does or will track KPIs (n=662).
© 2024 S&P Global.
The World Economic Forum's annual winter meeting at Davos had four themes, but "Artificial Intelligence as a Driving Force for the Economy and Society" seemed to dominate proceedings. The discussions — albeit the public ones we know about, as the more important ones are done in private — highlighted the increasing role of generative AI in different industries and the need to focus on accuracy and regulation. There was a focus on the need for higher accuracy in generative AI to enable its application in high-stakes industries. Additionally, the conversation at Davos also revolved around the regulation of generative AI, with concerns raised about its potential impact on human rights, personal privacy and societies.