20VC Why The Future of AI Is Open Not Closed, Why We Are Years Away From AI Being Autonomous, Why AI Founders Do Not Need to Move to the Valley & Why Founders Should Not Meet Investors in Between Rounds with Clem Delangue @ Hugging Face

In a comprehensive discussion with Harry Stebbings on "20 VC," Clem Delang, co-founder and CEO of Hugging Face, delves into the AI landscape, emphasizing the company's success and growth, having raised over $160 million from notable investors like Sequoia and Lux Capital. Delang underscores Hugging Face's commitment to the open-source AI community and its pivot from an AI friend platform to an AI technology provider. He challenges the notion that AI is nearing human-like autonomy, instead portraying it as a new paradigm for technology development. Delang also shares insights on the venture capital ecosystem, advocating for a focus on financial support and strategic fundraising over continuous investor engagement. He stresses the importance of startups enjoying the building process rather than fixating on growth milestones.

• Hugging Face was named after the hugging face emoji, which the founders favored and wanted to use as a company name.
• The company was initially intended to have an emoji as its stock ticker symbol, reflecting a desire for innovation and modernity in branding.
• The name was meant to be temporary but gained popularity and brand recognition, making it a permanent choice.
• The founding of Hugging Face was driven by the founders' professional camaraderie and shared excitement for AI's potential.
• The company began with an AI friend concept, akin to a Tamagotchi, focused on entertainment rather than productivity.
• After sharing their underlying technology, they saw traction from the community and pivoted to an AI platform model.

"Yeah, when we started hugging face, we joked with my co-founders Julianne Thomas, that we wanted to be the first company to go public with an emoji rather than the three-letter."

This quote explains the unconventional and playful origin of the company's name, reflecting the founders' desire to stand out and innovate within the tech industry.

"The reality is that the company was formed because of some sort of professional crush between me and my co-founders, where we were like, we absolutely want to work together."

Clem Delang emphasizes the importance of the strong professional relationship among the co-founders as a foundational element of Hugging Face.

• Hugging Face raised initial funding rounds based on their AI friend concept, which facilitated billions of messages between users and the AI.
• The pivot to an AI platform was influenced by the positive reception and traction from the open-source community and companies.
• The shift illustrates the flexibility and responsiveness of the company to market feedback and community engagement.

"And so that basically made us pivot from this AI to this AI platform that we are now."

This quote highlights the strategic decision to pivot from a product-focused company to a platform-oriented business model due to external interest and potential for broader impact.

• Clem Delang views the current interest in AI from VCs and the public as a catch-up to the reality of AI's pervasive use in technology.
• AI is already massively used in products and services by companies like Google, Facebook, and Zoom.
• The excitement around AI is seen not as hype but as a realization of its integral role in modern technology.

"My understanding of the current situation is that the VC and the mainstream interest is like a catch up on the reality."

This quote conveys Delang's perspective that the surge of interest in AI is a belated acknowledgment of its already significant role in technology, rather than a speculative bubble.

• Delang credits open science and open source for the rapid advancements in AI, as they facilitate a collaborative environment for innovation.
• He believes without the open sharing of research and software, AI development would be decades behind where it is today.
• Recent mainstream breakthroughs, such as GPT models and improved hardware, have allowed AI to become more accessible to the public.

"Most of the progress that we've seen in AI is based on open science and open source."

This quote emphasizes the critical contribution of open science and open-source communities to the advancement of AI technology.

• Delang acknowledges the energy and activity in Silicon Valley regarding AI but also recognizes the global distribution of AI talent.
• He cites the example of Llama, an open-source model from Meta, which had a majority of its authors based in Paris.
• Delang suggests that while Silicon Valley has a lot of excitement around AI, it is not necessary for AI founders to be physically located there.

"I don't think you do. I think you have to be in Silicon Valley."

This quote challenges the notion that AI startups must be based in Silicon Valley to succeed, highlighting the distributed nature of AI expertise and the global talent pool.## Location Independence for Founders

• Clem Delang emphasizes the importance of founder happiness and the ability to build a company from anywhere.
• Being happy in one's location can lead to the successful building of a great company.
• Founders should prioritize finding where they are happiest, irrespective of the perceived need to be in a specific location like Silicon Valley.

"At the end of the day I think you can build a company from anywhere. The most important thing, and I'm sometimes like calling bullshit on founders, saying, oh, I need to be there. I've taken a very strict decision to move there for my company. At the end of the day, I think it's important for founders to be happy. And if they're happy they can build a great company."

The quote highlights the argument against the conventional wisdom that founders must relocate to places like Silicon Valley to succeed. Clem Delang suggests that founder happiness is more crucial for company success than location.

• There are two distinct models for AI development: a centralized model (one model to rule them all) and a decentralized open-source model (many models).
• The centralized model focuses on a few large, generalist AI models developed by a limited number of organizations.
• The decentralized model involves a distributed approach where many companies build and train specialized AI models for specific use cases.

"They differ a lot in where do you allocate AI builders one model to rule them all? You bet on kind of like models getting bigger and bigger with more and more generalist capabilities, and the builders of these models being concentrated in one or few organizations."

Clem Delang explains the difference between centralized and decentralized AI development models, highlighting that the centralized approach concentrates on fewer, larger models with broad capabilities.

• Companies must decide between using a single large AI model or multiple specialized models for their products.
• The choice involves considering short-term ease versus long-term differentiation and capability building.
• Relying on a single AI model API is quicker initially but may lead to higher costs and less differentiation in the long run.
• Training and optimizing custom AI models are likened to writing unique code for a technology product tailored to specific use cases.

"It's a tough question, especially because it involves a lot of short term versus long term. The reality that sometimes today using one model behind an API is faster and easier at the beginning. But the challenge in the long run, you have more risk because then you don't really internally build the capabilities to actually do AI yourself."

Clem Delang discusses the trade-offs between using readily available AI models and developing in-house AI capabilities, emphasizing the risks and benefits associated with each approach.

• Large enterprises may initially prefer bundled AI solutions due to ease of use and perceived security.
• This presents an opportunity for AI-native startups to innovate and disrupt the market with tailored AI approaches.
• Over time, companies may shift from easy bundled solutions to more customized AI models as they recognize the need for differentiation and internal AI capabilities.

"And that's why there's a huge opportunity for new companies to disrupt the incumbents, because they're going to go for the easy solution, whereas other companies that are more like AI native are going to be going for the more disruptive approaches."

Clem Delang sees the initial preference of enterprises for bundled AI services as an opportunity for startups to stand out by offering more specialized and innovative AI solutions.

• The legal status of training data is a significant challenge for both open-source and proprietary AI models.
• Legal clarity regarding fair use and regulatory expectations is anticipated to improve within the year, which will benefit the AI field.

"Yeah, he has a point. I would argue that it's a challenge for the proprietary approaches too, because they're also going to get challenged by that."

Clem Delang acknowledges the legal challenges associated with the use of training data in AI, noting that it affects all types of AI development models.

• The relationship between AI companies and content creators needs to evolve to benefit both parties.
• Initiatives like training on opt-out datasets show progress towards a fairer model for content use in AI.
• Hugging Face has trained on a fully opt-out data set, allowing developers to remove their data from training, which is a step towards addressing content access and monetization issues.

"I hope we get into a model that works better for everyone, for content creators to keep incentivizing them to create good content and AI companies alike."

Clem Delang expresses hope for a more equitable model that fairly compensates content creators while enabling AI companies to innovate.

• Hugging Face follows a freemium model where most companies use the platform for free, and a subset pays for premium features.
• Paying customers, which include major companies, opt for premium support, enterprise features, and faster GPUs.
• The pricing model is not yet optimized for maximum revenue as the focus is on adoption and usage, which are seen as more critical KPIs.

"Our model is simpler than what people think as a platform with a lot of usage. We kind of follow a kind of classic freemium model."

Clem Delang describes Hugging Face's business model, emphasizing the simplicity and classic structure of their freemium approach.

• Monetization is not the most critical question for platforms with network effects, where adoption and usage are the primary focus.
• Revenue generation is seen as delayed revenue, with the expectation that widespread adoption will lead to profitability.
• Hugging Face views monetization as a learning process, with gradual steps from six-figure to higher revenues, adapting as the AI technology evolves.

"It's not the most important question because as a platform with network effects, the adoption and the usage is like the number one KPI for us especially."

Clem Delang downplays the immediate importance of monetization, focusing instead on the long-term benefits of widespread adoption and usage of Hugging Face's platform.

• The debate centers on whether startups or fast-moving incumbents will benefit most from AI advancements.
• Incumbents with existing distribution networks and rapid adoption of AI, like Microsoft and Adobe, are positioned to capture significant market gains.
• However, there remains an opportunity for startups to achieve distribution and impact the market with innovative AI applications.

"And you mentioned adoption there. And it leads me to think about who gains from this next wave most predominantly."

Harry Stebbings reflects on the potential beneficiaries of AI advancements, considering the impact of adoption and distribution in determining whether startups or incumbents will prevail.## AI Startups and Incumbents

• AI startups are distinguished by their focus on training models, creating new architectures, and optimizing models.
• Incumbents find it challenging to adapt to AI due to the need for a different technology-building paradigm.
• AI requires scientific research and development, which can be a slower process not suited to larger, established companies.

"And if you think about an AI startup as a company that is actually training models, creating new architectures, optimizing models themselves, I think it's a different story because this is really hard to do for the incumbents." "It's a completely different way to build technology."

These quotes explain that AI startups have an advantage in innovation over incumbents because they are designed to focus on the core aspects of AI development, which requires a fundamentally different approach compared to traditional technology development.

• Hiring is a significant challenge due to the need for a hybrid of science and engineering skills.
• Competition for talent is fierce, and the cost of hiring skilled individuals is high due to the increased funding in the sector.

"I would say hiring probably right now, like getting the best people and getting this hybrid profile because it's science plus engineering, hiring and getting the right set of co-founders, early team members is the harder thing, especially because there's a lot of competition with others."

This quote highlights the difficulty AI startups face in attracting the right talent, emphasizing the unique combination of skills required and the competitive job market driven by well-funded companies.

• AI startups may require more funding due to higher costs associated with computing and specialized personnel.
• There's uncertainty about whether the trend of raising large amounts of funding is necessary or effective.
• The diminishing returns on investment for larger AI models question the need for excessive funding.

"It does cost more money to build an AI first startup than a regular kind of like software startup." "The return on investment, on training larger and larger models is starting to go down."

These quotes discuss the financial aspects of building an AI startup, indicating that while they are more expensive to establish than traditional software startups, the strategy of seeking large funding rounds for more compute power may not always yield proportional benefits.

• Regulation of AI should focus on current challenges such as biases and misinformation rather than hypothetical, science fiction scenarios.
• There is disagreement on whether AI should be regulated preemptively or reactively.

"Regulation is necessary because it's a new way of building technology, and it's going to create some challenges." "These challenges are not so much AI running wild autonomously and taking over the world."

The speaker disagrees with the idea that AI should be regulated before it becomes a problem, arguing that the real issues requiring regulation are already present and do not involve the extreme scenarios often depicted in science fiction.

• The anthropomorphization of AI and the exaggeration of its capabilities are misleading and unhelpful.
• AI should be understood as a new method for building technology, not as an autonomous entity.

"The biggest thing is all this talk about AGI and anthropomorphization of AI, right? Considering and characterizing AI as human."

This quote expresses frustration with the misrepresentation of AI in public discourse, emphasizing that AI is far from being a semi-human entity and should be viewed realistically.

• The speaker has a policy of not engaging with external investors between funding rounds to maintain focus on company building.
• A humorous anecdote is shared about receiving a term sheet via email from an investor who wanted to participate after the formal fundraising process had concluded.

"One of these rules is that I don't talk to any external investors in between rounds." "Here is a term sheet before even talking to me, just as a reply on an email."

These quotes detail the speaker's approach to managing relationships with investors, including a personal rule to avoid discussions outside of designated fundraising periods and a unique experience of receiving a term sheet unexpectedly.

• There's debate over the best approach to building relationships with investors.
• The speaker prefers intense, focused interactions during fundraising rather than ongoing engagement.
• The counterargument suggests that relationships built over time are more likely to lead to trust and alignment.

"I spend like a shit ton of time with them. At least three days full time, which is a lot of time." "People invest in lines, not dots."

These quotes capture the differing opinions on investor relations, with the speaker advocating for short, concentrated interactions and the interlocutor suggesting a more gradual relationship-building approach.## Relationship Dynamics in Early Stage Fundraising

• The initial phase of investor-founder relationships can be artificial due to the imbalance and hierarchy.
• Authentic relationships are important, but these can be compromised by the pressure to sell or be sold to.
• Clem Delang suggests that the early days of investor-founder interactions lack purity due to the underlying intent to sell or be convinced.

"Intensity of relationship within three days, that is a completely manufactured relationship. I will tell you anything you want to hear, baby."

This quote highlights the artificial nature of relationships formed under the pressure of fundraising, where honest connections are secondary to the goal of securing an investment.

• Investors' primary role is to provide financial support and help with fundraising efforts for subsequent rounds.
• Clem Delang believes investors have deviated from their main role, sometimes acting as operators, which can lead to issues due to their limited time and understanding.
• A misalignment between investor and entrepreneur roles can negatively impact companies.

"Something I believe in is that investors are first and foremost investors, meaning that their main value adds is to do rounds to help you on financial matters."

This quote summarizes Delang's view that investors should focus on financial support and assistance with fundraising rather than operating roles within the company.

• Entrepreneurs should not expect the journey to become easier as their company grows.
• Each stage of a company's development presents unique challenges.
• Entrepreneurs should focus on enjoying the journey and building a company they are passionate about, rather than solely aiming for financial milestones.

"One thing that I wish I knew earlier is that it doesn't get easier."

Delang reflects on the misconception that company growth leads to fewer challenges, emphasizing the importance of enjoying the entrepreneurial process.

• Fundraising experiences can vary greatly, regardless of the company's stage.
• The ease or difficulty of raising funds is more dependent on traction, momentum, and achievements than on the stage of the company.
• The fundraising process is ultimately a negotiation between two parties, which can be simple or complex based on the dynamics of the relationship.

"Something I didn't expect is that the way you race isn't so much dictated by your stage and your rounds, but more dictated by the situation that you're in in terms of traction, in terms of momentum, in terms of achievements."

Delang shares his insight that the conditions of a company, such as its traction and momentum, are more influential in fundraising than the stage of investment.

• All companies will eventually develop their own AI models, similar to ChatGPT or GPT-4.
• This technological integration will become a standard across various industries.

"In my opinion, all companies will have their own AI models."

Delang predicts the widespread adoption of AI models by companies, indicating a future where AI is an integral part of business operations.

• The success of Hugging Face is closely tied to the overall success and advancement of AI technology.
• The company's open-source, community-driven approach is designed to contribute to and support the broader AI ecosystem.

"The biggest kind of like market risk for us is that if AI fails to deliver, it's not going to work for hugging face no matter what."

Delang identifies the dependency of Hugging Face on the progression and reliability of AI technology as a whole, which underscores the company's commitment to the AI community.

• Richard Socher is highlighted as an impactful angel investor due to his extensive background in science, business, and entrepreneurship.
• Angel investors with diverse experiences can provide valuable insights and support across different aspects of a company.

"I would go with Richard Socher, who's one of the most prominent scientists in NLP."

Delang praises Richard Socher as an angel investor who has contributed significantly to Hugging Face, emphasizing the importance of having knowledgeable and multifaceted investors.

• Realizing that business challenges do not diminish with growth can lead to a healthier focus on the process of building a company.
• This mindset shift encourages entrepreneurs to build a company they are passionate about and to derive joy from the journey itself.

"The fact that nothing gets easier because it changed my mindset."

This quote describes Delang's personal revelation that challenges persist regardless of company size, which led him to prioritize the intrinsic value of the entrepreneurial process.

• Hiring machine learning engineers who can build new AI architectures and train state-of-the-art models is extremely difficult.
• There is a limited number of people with the necessary experience and skills in this field.

"Machine learning engineer. And by machine learning engineer, I mean someone who's really building new architecture for AI models and able to train state of the art models."

Delang identifies the role of machine learning engineer as particularly challenging to fill, due to the scarcity of experienced professionals in the AI industry.

• Hugging Face aims to be impactful and useful in the AI space, rather than just being the biggest company.
• The company focuses on making a significant contribution to AI, with size and growth seen as potential byproducts of their success.

"So hopefully in ten years, hugging face would be the most impactful organization and company in AI."

Delang expresses his hope for Hugging Face to become a leading and influential entity in AI, emphasizing impact over sheer size.