This report was published by 451 Research on July 31, 2023
Introduction
July continued the pace of generative AI releases we saw in the first half of the year. China has been a news hotspot in July, with a number of Chinese-headquartered enterprises showcasing new products and a major regulatory update. One dynamic that may start to challenge the direction of the global market is consumer attitudes toward AI, which our data shows is trending negative across almost all demographics. Media attention paid to existential threats from AI, as well as more practical concerns around bias and intellectual property, appears to have contributed to the general public seeing a greater likelihood of AI having a significant impact on their career, personal life and society, and being more likely to predict that the impact will be negative.
The Take
Big funding rounds continue to roll in for generative AI startups, and we saw the first high-value M&A deal, with Databricks paying $1.3 billion for MosaicML. The deal effectively valued each MosaicML employee at $62 million, such is the race for talent at the top end of generative AI. We expect more such deals, as well as consolidation in the crowded text generation space, in particular. As our data in Figure 1 shows, sentiment among the general public isn't nearly as positive as it is among the technology industry, which is something politicians are acutely aware of as they continue to inch toward regulating generative AI.
Product releases and updates
Meta Platforms Inc. released Llama 2, an open-source large language model available for free for research and commercial use. Microsoft Corp. was announced as the "preferred partner" for the release, and will distribute the LLM through its Azure cloud service.
Microsoft announced Bing Chat Enterprise and pricing for Microsoft 365 Copilot. Bing Chat Enterprise is based on the consumer Bing Chat, which has been available for a few months and is based on OpenAI's GPT-4 LLM. The key difference is that the enterprise version adds commercial data protection in the form of no chat data being saved or used to train the models, and Microsoft cannot view the data either. Bing Chat Enterprise is currently in preview and comes at no additional cost in Microsoft 365 E3, E5, Business Standard and Business Premium. Microsoft plans a future stand-alone offering for $5 per user, per month. Speaking of pricing, Microsoft made Wall Street investors very happy when it revealed Microsoft 365 Copilot, which leverages LLMs and Microsoft Graph to give users a tool to ask questions and do analysis across all the Microsoft 365 applications (what used to be called Office 365). It will cost $30 per user, per month for Microsoft 365 E3, E5, Business Standard and Business Premium customers, once it is widely available. Microsoft's share price rose to an all-time high after that pricing was announced, indicating that investors believe a lot of organizations will be willing to pay that premium for Copilot.
Alibaba Cloud presented an image-generation tool at World Artificial Intelligence Conference 2023, hosted in Shanghai. The tool, known as Tongyi Wanxiang, can generate images in response to prompts in English and Mandarin Chinese. The tool is available in beta for enterprise customers in China.
Huawei Cloud presented Pangu Models 3.0 at its developer conference, HDC Cloud 2023. The third iteration of its AI models includes new generative capability sets — with its natural language processing capability, one of "five foundational models" underpinning the industry-tailored set, announced to now support copywriting, code generation and knowledge-based Q&A tasks.
Elon Musk launched xAI, a new company focused on somewhat vague concepts such as "good artificial general intelligence" and "understanding the universe," according to an audio chat held on Twitter by Musk during its July 12 launch. He also said xAI's models would use Twitter as training data and may also collaborate with Tesla Inc. XAI's all-male founding team counts companies such as Google DeepMind, Microsoft Research and Tesla, among others, in their backgrounds.
LlamaIndex, which had announced an $8.5 million seed round in June, released 0.7.0 of its eponymous project in early July — designed as a data framework for LLMs. With a query interface, as well as data ingestion and indexing tools, LlamaIndex is designed to connect proprietary data and LLMs. This latest release enhances metadata management capabilities and shows some disaggregation of the offering, making its response-synthesis modules and LLM abstractions stand-alone.
Funding
Inflection AI received $1.3 billion in funding at the end of June, led by returning investor Microsoft. The machine-learning startup announced "Personal AI," known as Pi, in May. The focus of the technology appears to be emulating emotional intelligence, and Inflection posits AI safety as a major differentiator. The transaction was raised at a post-money valuation of $4 billion.
SAP SE's Sapphire Ventures investment arm invested undisclosed amounts in generative AI foundation model firms Aleph Alpha, Anthropic and Cohere. The funding is part of Sapphire Ventures' commitment to invest more than $1 billion in what it calls "next-generation AI-powered enterprise technology startups." The company already has investments in companies including DataRobot and ThoughtSpot. Totals raised by the three companies prior to SAP's investment, according to S&P Global Inc.'s Capital IQ Pro, were $33.6 million for Aleph Alpha, $1.6 billion for Anthropic and $434.6 million for Cohere
AI video-creation platform Synthesia announced a $90 million series C round, which has given the company a $1 billion post-money valuation. The round was led by Accel Partners, with participation from NVentures and NVIDIA Corp.'s venture capital arm, alongside a number of existing investors.
Snowflake Ventures was one of a number of investors that contributed to a $50 million series A round for Reka, an AI model startup. Reka emerged from stealth in June, and was founded by a team of researchers stressing their backgrounds within DeepMind, Google, Baidu Inc. and Meta. The company claims its Yasa assistant, which is in closed beta, is well suited to tailored multimodal use cases.
Light Year, a Chinese company targeting artificial general intelligence, said in July that it had achieved $50 million in funding earlier this year. The transaction closed a few months before the company was acquired by Tianjin Sankuai Technology and Meituan at the end of June for 2 billion yuan ($280 million).
Blackbaud Inc.'s Tech Accelerator Program announced a July cohort focused on generative AI startups targeting social impact. Startups within the Accelerator are targeting a wide array of use cases, including a generative grant-writing capability for nonprofits, a fundraising marketing tool and impact-measurement applications. Program participants gain access to grant funding, resources and an annual showcase.
Politics and regulations
The US Federal Trade Commission (FTC) reportedly launched an investigation into OpenAI, to assess whether it had run afoul of consumer protection laws by causing "reputational harm" to consumers by essentially generating false statements about people. It sent a 20-page document demanding details of how OpenAI assesses the risks of its foundation models.
China made a number of changes to its AI guidelines, a document we highlighted in our May digest. Areas in the finalized version (which will go into effect on August 15) that appear to have been watered down include fines — with language removed detailing the minimum size for violations — and training data. The initial draft declared companies would have to be able to "guarantee" the quality of training data — around authenticity, accuracy and bias — whereas the latest merely asks for "effective measures" to be taken. Perhaps most interestingly, and a reflection of the country's desire to compete, AI technology that is designed in China but used overseas, and not by the Chinese public, will be exempt from these guidelines.
US comedian Sarah Silverman announced that she was suing OpenAI and Meta for copyright infringement, filing the suits alongside authors Christian Golden and Richard Kadrey. The lawsuit against Meta claims that content produced by the three appears in the dataset used to train Meta's Llama model series. The OpenAI suit suggests that, considering ChatGPT can summarize books by the three authors, it must use material taken from the books without the permission of the creatives. The suits join a growing wave of litigation from artists and actors around image-generation tools, and a number from other US authors against OpenAI's LLMs.
One trend that may start to influence politics and regulation is worsening attitudes toward AI. Consumer attitudes appear to have shifted from Q4 2022 to Q2 2023. AI was seen as increasingly transformative by respondents to our VoCUL: Connected Customer, but that transformation was seen more negatively. From Q2 2019 to Q4 2022, on a bi-annual basis, respondents predicting a positive societal impact from AI within the next two years significantly outweighed those with a negative perception, on average 46% against 25%. The second quarter of 2023 represented a stark shift, with a much closer 40% against 37%. Attitudes were trending negatively across almost every demographic assessed within the survey, with major shifts for Generation X and baby boomer respondents, as well as higher-educated respondents. The shift for higher-educated respondents, where concerns around career have accelerated markedly, indicates that generative AI — and the reframing of AI's automating potential — may be a driver of more negative predictions.
