• One Cerebral
  • Posts
  • Fine Tuning vs Pre Training Models - What is best?

Fine Tuning vs Pre Training Models - What is best?

Co-Founder/CEO at Sierra (ex-Facebook CTO): Bret Taylor

Credit and Thanks: 
Based on insights from 20VC with Harry Stebbings.

Key Learnings

  • The AI landscape resembles a bubble, similar to the dot-com era, with potential for long-term value despite short-term excesses.

  • Entrepreneurship is largely a learnable skill; many leaders develop their capabilities over time.

  • Companies should adopt ready-made AI applications instead of building from scratch.

  • Conversational AI is emerging as a key interface for customer interactions due to its low friction and ease of use.

  • Balancing AI agency and control is essential; too much freedom can lead to unpredictability, while too little results in robotic interactions.

Today’s Podcast Host: Harry Stebbings

Title

Why Pre-Training is for Morons & Companies Will Build Their Own Software 

Guests

Bret Taylor

Guest Credentials

Bret Taylor is a prominent tech entrepreneur and executive with an estimated net worth of $258 million as of November 20241. His impressive career includes co-creating Google Maps and serving as CTO at Facebook, where he was promoted less than a year after joining through the acquisition of FriendFeed, a startup he co-founded. Taylor later co-founded Quip, which was acquired by Salesforce for $750 million, and went on to become Salesforce's Co-CEO. Most recently, Taylor co-founded Sierra, an AI startup that raised $175 million in funding at a $4.5 billion valuation, and serves as the Chairman of the Board at OpenAI.

Podcast Duration

1:16:09

This Newsletter Read Time

Approx. 5 mins

Deep Dive

In the evolving landscape of artificial intelligence, the question of whether we are at "Peak AI" looms large. The current surge in investment and innovation has led many to speculate about the sustainability of this growth. A notable perspective suggests that while we may be experiencing a bubble, it is essential to recognize that bubbles can yield significant advancements. Just as the dot-com bubble birthed enduring giants like Amazon and Google, the current AI boom may similarly lead to transformative companies that redefine industries. The excitement surrounding AI is palpable, yet it is accompanied by a cautionary note about the potential for excess and the need for a discerning approach to investment.

As AI models become increasingly sophisticated, there is a growing concern about their ability to replace traditional software solutions. The argument posits that while these models can automate tasks and enhance productivity, they may not fully supplant the need for dedicated software applications. For instance, the transition from building bespoke software to utilizing AI-driven solutions mirrors the earlier shift from on-premises software to cloud-based services. Companies are now seeking out AI solutions that can seamlessly integrate into their existing workflows, rather than attempting to replace entire systems. This evolution highlights the importance of understanding the unique value that AI can bring to specific business contexts.

AI services companies are poised to play a pivotal role in the next generation of applications. As organizations grapple with the complexities of implementing AI, these service providers offer essential expertise in navigating the transition. The demand for tailored solutions that address specific business needs is on the rise, and companies like Sierra are at the forefront of this movement. By developing branded customer-facing AI agents, Sierra exemplifies how AI services can enhance customer experiences while alleviating the burden of implementation for businesses. This shift towards AI services underscores the necessity for companies to adapt to the changing technological landscape while ensuring that they remain competitive.

Balancing the pursuit of artificial general intelligence (AGI) with the development of practical products presents a unique challenge for organizations. The aspiration to create AGI, which could fundamentally alter the fabric of society, must be tempered with the need to deliver tangible value through existing applications. This dual focus requires a strategic approach that prioritizes both long-term vision and immediate market needs. Companies must navigate the complexities of developing cutting-edge technology while ensuring that their products resonate with consumers and address real-world problems.

To thrive, companies must focus on creating value that justifies their investment in AI technology. This may involve leveraging existing models and fine-tuning them for specific applications, rather than incurring the substantial costs associated with pre-training new models. The emphasis on finding product-market fit before scaling operations is essential for ensuring long-term viability in a competitive environment.

As companies pour resources into building advanced models, the financial implications become increasingly significant. The challenge lies in determining whether these investments will yield sufficient returns to justify the expenditure. While some argue that the current trajectory of investment is unsustainable, others believe that the potential for groundbreaking advancements in AI will continue to attract funding. The key will be to strike a balance between ambitious goals and practical financial considerations.

The decision to build Sierra, a platform for creating branded AI agents, stemmed from a recognition of the transformative potential of conversational AI. The founders understood that the future of customer interaction would increasingly rely on AI-driven solutions that offer seamless, low-friction experiences. By focusing on developing a platform that enables companies to create their own AI agents, Sierra aims to redefine how businesses engage with their customers. This vision is rooted in the belief that conversational interfaces will become the dominant mode of interaction in the digital landscape.

However, the journey to build Sierra was not without its challenges. One of the most significant hurdles encountered was the inherent non-determinism of generative AI. Unlike traditional software, which operates on a set of predefined rules, AI models can produce unpredictable outputs. This unpredictability necessitated a shift in how businesses approach the design of AI agents. Instead of relying solely on rigid rules, companies must establish guardrails that allow for creativity while ensuring that the AI remains aligned with their brand values and operational goals.

In an era marked by rampant misinformation, the need for content verification and trust has never been more pressing. As AI-generated content becomes increasingly prevalent, concerns about authenticity and accuracy arise. The challenge lies in developing mechanisms that enable users to discern credible information from falsehoods.

Taylor also expressed the belief that most skills, including entrepreneurship, can be learned and improved with focus and dedication. He noted that while certain personality traits may predispose individuals to entrepreneurship, many successful leaders develop their capabilities through experience and confidence gained over time.

Actionable Insights

  • Embrace Iterative Development: Focus on responsible iterative deployment of AI technologies to learn from real-world applications and improve continuously.

  • Invest in Change Management: Prioritize training and support for employees to adapt to AI-driven changes in operations, ensuring smooth transitions and effective utilization.

  • Leverage Existing Models: Instead of pre-training new models, utilize high-quality existing models and fine-tune them to meet specific business needs, reducing capital expenditure.

  • Create Customer-Centric AI Solutions: Develop AI agents that provide tailored customer experiences, enhancing engagement and satisfaction through conversational interfaces.

  • Foster a Long-Term Vision: Maintain a relentless focus on long-term goals, communicating a clear vision to motivate teams and stakeholders to navigate short-term challenges effectively.

Why it’s Important

Understanding the dynamics of the AI market is crucial for entrepreneurs and business leaders as it informs strategic decision-making and investment priorities. The insights shared by Taylor highlight the necessity of balancing innovation with practical application, ensuring that businesses can harness AI's potential while mitigating risks associated with overvaluation and speculative investments.

What it Means for Thought Leaders

For thought leaders, the discussion emphasizes the importance of fostering a forward-thinking mindset that embraces both technological advancements and the human elements of leadership. It suggests that as AI continues to evolve, leaders must remain adaptable and open to new ideas, ensuring that their organizations can thrive in an increasingly complex landscape.

Mind Map

Key Quote

"I think we are in a bubble, but I think bubbles have different shapes… it would be dangerous to miss a bubble as strictly excess."

As AI technologies continue to mature, we can expect a shift towards more specialized applications that cater to specific industries and use cases. The trend of commoditization will likely lead to increased competition, driving down costs and making AI solutions more accessible to smaller businesses. Additionally, the emphasis on change management and user experience will become paramount as organizations seek to integrate AI seamlessly into their operations, ultimately reshaping the way businesses interact with technology and their customers.

Check out the podcast here:

Latest in AI

1. OpenAI's chairman, Bret Taylor, has seen his AI startup, Sierra, achieve a valuation of $4.5 billion following a recent funding round that raised $175 million, led by Greenoaks Capital. Sierra, which focuses on AI-powered customer service chatbots, has reported over $20 million in annualized revenue and aims to minimize "hallucinations" in its AI products. This significant valuation increase reflects strong investor interest in AI ventures, which have accounted for a third of all venture capital investments this year. The rapid growth of Sierra highlights the ongoing enthusiasm and confidence in the AI sector.

2. ChatGPT's web application has introduced a new feature that allows users to search through their old chat histories. This functionality, previously available only on the mobile app, enables users to locate past conversations more efficiently without the need to scroll through lengthy chat logs. The addition of this search option reflects OpenAI's commitment to enhancing user experience by making it easier for users to access and reference their previous interactions with the AI.

3. In Google's Q3 earnings call, Sundar Pichai announced that AI is now responsible for generating 25% of new code at the company. Pichai's remarks underscore the significant role AI technologies are playing in enhancing productivity and innovation within the organization.

Useful AI Tools

1. Browser Use: AI-powered web interaction tool for seamless agent automation.

2. Lorikeet: Advanced AI for complex customer support ticket resolution.

3. Raphael AI: Instant, login-free AI image generation platform.

Startup World

1. ElevenLabs, a leader in AI voice cloning and dubbing, raised $250 million in a Series C funding round led by ICONIQ Growth, with participation from previous investor Andreessen Horowitz. The company's valuation has now reached between $3 billion and $3.3 billion, reflecting the high demand for its API-based platform that offers a wide range of voice tools including text-to-speech, voice cloning, and voice translation.

2. OpenAI has partnered with Retro Biosciences, a longevity science startup backed by Sam Altman, to develop GPT-4b micro, an AI model designed to re-engineer proteins with the aim of extending human lifespan by 10 years. The collaboration focuses on the Yamanaka factors, a set of proteins that can convert adult human cells into pluripotent stem cells, potentially leading to the creation of replacement cells and lab-grown human organs.

3. LG Electronics has acquired a majority stake in Bear Robotics, a Silicon Valley-based startup specializing in AI-driven autonomous service robots, by exercising a call option to acquire an additional 30% stake, bringing its total ownership to 51. This strategic acquisition, valuing Bear Robotics at approximately $600 million, aligns with LG's ambition to strengthen its presence in the robotics sector.

Analogy

The current AI boom is like the early days of the internet—hyped, volatile, and full of speculation. While bubbles may burst, they also leave behind lasting innovations, much like Amazon and Google emerged from the dot-com era. AI is shifting from an experimental novelty to a practical tool, just as cloud computing once did for software. Companies that adapt AI to solve real-world problems, rather than chasing AGI dreams, will be the ones that endure. The key isn’t just building the most advanced AI—it’s making AI useful, reliable, and seamlessly integrated into everyday workflows.

What did you think of today's email?

Your feedback helps me create better emails for you!

Loved it

It was ok

Terrible

Thanks for reading, have a lovely day!

Jiten-One Cerebral

All summaries are based on publicly available content from podcasts. One Cerebral provides complementary insights and encourages readers to support the original creators by engaging directly with their work; by listening, liking, commenting or subscribing.

Reply

or to participate.