Sarah Tavel: Will Foundation Models Be Commoditised? | E1149

Sarah Tavel: Will Foundation Models Be Commoditised? | E1149

Intro (00:00:00)

  • The speaker believes that if one needs an advanced model, it will likely be closed-source.
  • The speaker focuses on the application layer because that's where significant value is captured and created.
  • There is a tremendous amount of investment required to progress these models.
  • Each successive model is more expensive to train, suggesting an oligopoly in the future.
  • The speaker is excited to have Sarah Tavel on the show.

Background (00:00:51)

  • Sarah Tavel, a partner at Benchmark, initially declined an offer to join the firm but later accepted after getting to know the team and their unique approach to partnering with founders.
  • Peter Fenton, a general partner at Benchmark, is known for his relentless learning mindset, curiosity, and exceptional EQ in understanding people's motivations and helping founders reach their full potential.
  • When evaluating investment opportunities, Tavel emphasizes the importance of the "why now" factor, which requires a compelling and realistic explanation of why a particular technology or idea is poised for success at the present time.
  • A strong "why now" is crucial for a company's success, providing a driving force that helps founders overcome challenges and make progress. Without it, founders may struggle to create significant opportunities and surprise the market.
  • A sustaining "why now" is essential, as it needs to endure over time.
  • Balancing the desire for authenticity with the challenges posed by competing forces like TikTok and Instagram is crucial for success.

Value of Selling the Work in the Application Layer (00:09:11)

  • AI is currently being used by incumbents to improve productivity, while startups can use it disruptively by selling the work product or outcome.
  • As foundation models improve, more complex enterprise workflows will become automatable, but a human-in-the-loop approach is often used to bridge the gap between current AI capabilities and full automation.
  • Large firms may adopt AI work creation tools and charge clients the same hourly rates, becoming more efficient businesses, while new entrants may use AI to offer lower prices, faster services, and gain market share.
  • The application layer, which owns the end user, is believed to capture and create the most value, and companies that invest in this layer will benefit as underlying models improve.
  • The challenge for AI startups is that the distribution advantage of incumbents like Microsoft can hinder their adoption, and foundation models have the potential to be commoditized.
  • New companies are emerging that internalize AI's potential and create unique value propositions by owning more of the workflow and preparing experiences beyond what APIs offer.
  • The depth of usage and continued engagement with the product in early cohorts can indicate whether adoption is experimental or enduring.

The End State of the Model Landscape (00:23:37)

  • The AI customer service and sales tools market is highly competitive, and success depends on backing founders with competitive energy, high urgency, aggressiveness, and ambition.
  • AI startups selling work have a larger market potential and easier go-to-market compared to those selling productivity improvements or underlying technology.
  • The dilutive nature of AI companies can be rationalized by considering the potential for self-fulfilling prophecies, where significant capital investment enables talented teams to build moats through GPU investments and model training.
  • Sarah Tavel advises against FOMO (fear of missing out) and emphasizes the importance of disciplined investment decisions.
  • Benchmark, a venture capital firm, focuses on building independent and enduring companies by carefully selecting a few investment opportunities each year and committing the full partnership behind them.
  • AI startups targeting B2B use cases should focus on specific modalities rather than attempting to cover all modalities.
  • OpenAI's focus on its core Foundation model (GPT) means that other types of models, such as audio, image, and video, will progress but with less focus and ambition.
  • DeepL has built a strong foundation model and integrated it with various workflows, providing a seamless user experience.
  • There is significant work beyond the model itself, such as building infrastructure and ensuring efficient compute usage.
  • The progression of models becoming more expensive to train suggests a future oligopoly in the model landscape.

The Challenges & Future of AI Models (00:36:06)

  • Training AI models is becoming increasingly expensive due to the need for more specialized chips, power consumption, and research.
  • The cost to the end customer is expected to continue decreasing due to increasing competition.
  • The open-source vs. closed-source debate is ongoing, with the current trend favoring closed-source models due to the high cost of training and the uncertainty of recouping investment through open-sourcing.
  • Meta's recent open-sourcing of its large language model, Llama, may change the game if it proves to be economically viable in the long term.
  • For cutting-edge models, closed-source is currently the preferred approach, while open-source models may suffice for certain use cases where being on the frontier is not necessary.
  • The speaker did not explicitly state any worries about the space.

The Role of Competition in the AI Industry (00:39:15)

  • Benchmark's unique partnership model provides founders with a higher quality product by not delegating core work to internal consultants.
  • Benchmark's partners have become experts in recruiting through repetition, making it a key strength for the firm.
  • Sarah Tavel finds satisfaction in winning deals where Benchmark's partnership model and recruiting expertise were crucial in securing the investment.
  • A strong connection between founders and investors is fostered when there is a shared belief in the founder's vision and commitment.
  • Benchmark sometimes makes investment decisions as a team, demonstrating a collaborative approach.
  • The team-oriented approach allows all partners to benefit equally from successful investments and enables collective effort in supporting founders and realizing their potential.

Reflecting on Missed Investment Opportunities (00:46:02)

  • VCs should play an active role in supporting CEOs, asking critical questions, and providing assistance during challenges.
  • The quality of board membership is influenced by the founder's and VC's belief in the value of great partners.
  • Benchmark's investment model focuses on making a small number of focused investments and committed relationships with founders rather than scaling widely.
  • Selection bias can impact investment decisions, such as when founders decline board members or valuations exceed expectations.
  • Price is the most common reason for breaking the investment model, followed by board member status.
  • When considering a higher investment price, investors evaluate the company's potential to escape competition through factors like network effects or economies of scale.
  • Investors should avoid using price as a means to feel comfortable with a deal and should be cautious about the concept of "prata," which can create challenges for founders raising subsequent rounds.
  • Founders should prioritize trust and open communication with board members, especially during challenging times, and seek out board members with whom they can establish a trusted relationship.

Quick-Fire Round (00:58:03)

  • Sarah Tavel's biggest missed investment opportunity was Ethereum, despite recognizing its potential.
  • Sarah Tavel had a memorable meeting with a highly motivated founder who was determined to succeed against all odds.
  • Sarah Tavel's perspective on anti-Semitism has shifted, as she was previously unaware of its prevalence and is now shocked by its resurgence.
  • Sarah Tavel is concerned about the possibility of Donald Trump being re-elected and the potential commoditization of AI-powered language models, which could negatively impact democracy.
  • Becoming a parent has influenced Sarah Tavel's investment and operating decisions, as she now considers the opportunity cost of her time.
  • Sarah Tavel emphasizes the importance of a deep connection and alignment with founders when partnering for investments and highlights a recent, yet-to-be-announced investment that excites her due to her strong connection with the founder.
  • Sarah Tavel expresses gratitude for the insightful discussions and consistently high-quality shows with her conversation partner.

Overwhelmed by Endless Content?