CFIUS, Search, Thinking Fast & Slow

Cerebras, Liner, Sequoia, Relyance AI, Braintrust Data, Voyager AI, Distributional

Cerebras Pushes Roadshow Amid CFIUS Review

Reuters reports Cerebras Systems , the AI chip manufacturer positioned as an alternative to NVIDIA, is facing potential delays in its much anticipated initial public offering. The company is likely to postpone its roadshow, originally scheduled for early next week, due to an ongoing U.S. national security review by the Committee on Foreign Investment in the United States (CFIUS).

The review focuses on a minority investment from UAE-based tech conglomerate G42 , which is simultaneously Cerebras Systems 's largest investor and biggest client. In 2023, G42 accounted for 83% of Cerebras' total revenue, contributing $65.1 million. The company has also committed to purchasing $335 million worth of Cerebras stock by April, which would give it a stake exceeding 5%.

This scrutiny comes amidst concerns from the US Government that Middle Eastern companies may be serving as conduits for China to access advanced American AI technology. Cerebras and G42 have submitted a joint filing to CFIUS, later amending it to specify that G42 's shares would be non-voting securities.

Cerebras remains optimistic about receiving CFIUS approval later this year and plans to proceed with its IPO once clearance is obtained. The company, valued at approximately $4 billion in a 2021 funding round, reported total revenue of $136.4 million for the first half of 2024, a significant increase from $8.7 million in the same period the previous year.

Here's some additional context on the story via CNBC - https://www.cnbc.com/2024/10/11/cerebras-ipo-has-too-much-hair-as-chipmaker-tries-to-take-on-nvidia.html

South Korean Search Startup Liner Raises $29M

Liner , an AI search engine tailored for students and researchers, has successfully secured $29 million in a Series B funding round . The investment was led by InterVest capital partners , Atinum Investment , and Samsung Ventures , bringing Liner's total funding to $33 million.

Founded by Luke Jinu Kim and Brian(Chanmin) Woo , Liner has its roots in a browser extension developed over a decade ago to enhance web search results. The company has since evolved into a specialized AI-powered information retrieval platform, launched in 2023, that aims to provide more reliable and focused results than general-use AI services like ChatGPT or Perplexity .

Liner's technology narrows its search to credible sources such as academic papers and government databases, leveraging access to an extensive library of scientific journals and publications. The platform uses customized large language models to weigh the validity of information, allowing users to pose complex queries across various academic disciplines.

The Seoul-headquartered startup has seen significant traction in the US market, which accounts for its largest and fastest-growing user base. Liner boasts 10 million users across prestigious institutions like 美国加州大学伯克利分校 , 美国德克萨斯A&M大学 , and the 美国南加州大学 . Notably, about two-thirds of its paying users are based in the US, with the vast majority coming from higher education.

With this fresh injection of capital, Liner plans to expand its global team, particularly in San Francisco.

VC Funding Focus Increasingly on the Application Layer

A recent article by Sonya Huang , Pat Grady (and o1 ?) of Sequoia Capital makes a couple of interesting points. First, I remain firmly unconvinced that the latest series of AI models, exemplified by OpenAI's o1 are not "reasoning" in the traditional (i.e. human) sense of the term. The first half of the Sequoia article, I think, takes liberties with terms like "cognitive architecture ", and links to an article by LangChain co-founder Harrison Chase . Chase offers the following definition of "cognitive architectures"

What I mean by cognitive architecture is how your system thinks —?in other words, the flow of code/prompts/LLM calls that takes user input and performs actions or generates a response.

I will be the first to admit I know very little about computer science, and even less about how the human brain thinks. However, the definition above is clearly not how someone thinks. I'd argue we do not fully understand how humans think. I believe the definition above works well to describe the workflows, patterns and processes that help us arrive from a given problem to a desired solution. Designing a system with a series of "code/prompts/LLM calls" does NOT endow that system with the magical and elusive power of thinking.

Sonya and Pat also map pre- and post-o1 inferencing to System 1 and System 2 thinking respectively - a model proposed by the late Nobel laureate Daniel Kahneman to describe human (economic) decision making. I respectfully believe this mapping, while well-intentioned, is incorrect. LLMs are not reasoning in any human sense of the term. I strongly recommend watching the entirety of the following conversation on this topic -

The second half of the Sequoia piece is about how the transition to AI is going to drive a shift to service-as-a-software vs. the Cloud transition which brought about Software-as-a-Service. Software companies shifting to the Cloud were pursuing a $350B market opportunity according Sonya and Pat, the AI transition targets a market opportunity that is potentially in the trillions!

I don't know about the trillions yet, but I agree generative AI and the slew of technologies that enable better Enterprise AI applications that solve real-world problems at scale, with minimal human intervention, while being adaptable, observable, compliant and secure is undoubtedly a significant opportunity.

Which brings me to the following four announcements, all around making generative AI applications more reliable, usable and repeatable.

Relyance AI , a data governance and privacy compliance platform, has secured $32.1 million in a funding round led by Lightspeed and Bessemer Venture Partners . The company's AI-powered solution helps organizations manage data privacy, security, and compliance across their technology stack. Relyance AI plans to use the new funding to enhance its product capabilities and expand its market presence. With the increasing complexity of data regulations globally, Relyance AI 's platform aims to simplify compliance processes for enterprises dealing with large volumes of sensitive data.

Braintrust , a startup focused on AI model evaluation and testing, has raised $36 million in a round led by Andreessen Horowitz (a16z). The company provides tools and services to help organizations assess and improve the performance of their AI models. Braintrust 's platform is designed to address the need for reliable AI evaluation methods as businesses rely on AI to assist in critical decision-making. The fresh capital will be used to expand the company's team and further develop its AI testing capabilities.

Voyage AI , specializing in retrieval-augmented generation (RAG) technology, has received an undisclosed investment from Databricks Ventures. he company's AI-powered solution aims to enhance the accuracy of large language models by improving their ability to retrieve and incorporate relevant information. Voyage AI 's technology could potentially help enterprises build more reliable and context-aware AI applications.

Distributional , an enterprise AI testing platform, has announced a $19 million Series A funding round led by Redpoint Ventures , with participation from existing investors. The company's software helps organizations test and evaluate AI models, focusing on uncovering potential biases, errors, and vulnerabilities. Distributional plans to use the new funding to expand its team and enhance its product offerings.

And that's all folks. Happy Thanksgiving to my Canadian friends and a Happy Indigenous People's Day to all in the US.

If you found this useful, please share, comment and like the newsletter - it helps us grow!

To continue receiving updates, please click here to subscribe to InfraRead .

My day job is advising growing companies on corporate strategy, finance and M&A. If you are looking for an experienced partner with a track record of maximizing outcomes, get in touch .