Inside OpenAI | Logan Kilpatrick (head of developer relations)

Logan’s background (0s)

• Logan Kilpatrick is the Head of Developer Relations at OpenAI.
• Before OpenAI, he was a machine learning engineer at Apple and advised NASA on their open-source policy.
• Logan is passionate about working with people who have high agency and work with urgency.
• OpenAI's mission is to ensure that artificial general intelligence (AGI) benefits all of humanity.
• OpenAI's values include:
  • Safety: AGI should be developed and deployed safely and securely.
  • Transparency: OpenAI should be transparent about its research and development process.
  • Inclusivity: OpenAI should be inclusive of people from all backgrounds and perspectives.
  • Collaboration: OpenAI should collaborate with others to develop AGI in a responsible manner.
• OpenAI is a flat organization with a strong emphasis on collaboration.
• Teams are organized around specific projects, and there is a lot of cross-functional collaboration.
• OpenAI encourages employees to take risks and experiment.
• The company has a strong culture of learning and iteration.
• OpenAI uses a variety of techniques to ship quickly, including:
  • Continuous integration and continuous deployment (CI/CD).
  • A strong focus on automation.
  • A culture of experimentation and risk-taking.
  • A willingness to fail fast and learn from mistakes.
• High agency: the ability to take initiative and get things done without needing to be told what to do.
• Urgency: the ability to work quickly and efficiently.
• Logan sees the biggest opportunities in the following areas:
  • Generative AI: using AI to generate new content, such as text, images, and music.
  • Reinforcement learning: using AI to learn how to perform tasks by trial and error.
  • Robotics: using AI to control robots and other physical systems.
  • Natural language processing: using AI to understand and generate human language.
• Logan does not discuss the dramatic weekend that OpenAI had with the board and Sam Altman.

The impact of recent events on OpenAI’s team and culture (3m49s)

• Despite the sudden changes and lack of transparency during the dramatic weekend involving the board and Sam Altman, OpenAI employees quickly returned to work, demonstrating their commitment to the company's mission.
• The stressful experience brought the team closer together, creating a sense of shared resilience.
• Logan Kilpatrick expressed gratitude that the events occurred when the stakes were relatively lower, allowing for timely changes and transformations.
• The rapid growth of the team in the past year has contributed to a stronger sense of unity during challenging times.
• OpenAI has had several significant events, including the launch of JBT and GPT-4, which have brought the company together cross-functionally.
• Future events, such as the launch of GPT-5, are highly anticipated.

Exciting developments in AI interfaces (8m20s)

• Logan Kilpatrick is excited about new AI interfaces that are emerging, such as the Rabbit R1 consumer hardware device and TL draw, which allows users to sketch and have it turned into a website.
• He believes that chat is currently the predominant way people use AI, but he sees potential in multimodal AI and new UX paradigms that will push the boundaries of how we interact with AI.
• Kilpatrick predicts that 2024 will be the year of multimodal AI and new UX paradigms.

Using OpenAI tools to make companies more efficient (9m52s)

• OpenAI's tools, such as chatbots, are becoming increasingly useful and widely adopted.
• OpenAI focuses on developing general reasoning, coding, and writing capabilities, rather than highly specialized applications.
• Companies should consider specializing in specific use cases or creating unique interfaces to avoid direct competition with OpenAI's general-purpose products.
• To compete with OpenAI's general-purpose products, companies need to offer radically different solutions that solve specific problems better than existing options.

Examples of using AI effectively (13m4s)

• OpenAI tools like ChatGPT and GPTs can significantly boost a company's efficiency, especially in engineering, where productivity can increase by at least 50% for certain tasks.
• Companies can build customized solutions tailored to their specific needs using GPTs.
• Successful GPT use cases include writing ad copy, analyzing experiment results, planning OKRs, and performing venture capital due diligence.
• Logan Kilpatrick, head of developer relations at OpenAI, uses a private version of GPT to assist in planning and ensuring consistency in his work.
• GPT generates timelines, metrics, and identifies important stakeholders for Logan's planning processes, making it a valuable tool for him.

Prompt engineering (18m35s)

• Prompt engineering is a skill that involves effectively communicating with AI models to get the best output.
• Models are trained to give answers based on the input they receive, so the quality of the output depends on the quality of the input.
• Just like with humans, providing context and being specific in prompts can lead to more detailed and nuanced responses from AI models.
• Prompt engineering involves understanding the limitations of AI models and providing the necessary context to get the desired output.
• As AI models become more sophisticated, prompt engineering will likely become more important in getting the most out of them.

How to write better prompts (22m12s)

• Provide more context about the person or topic you're asking about.
• Use external sources like blogs and social media to gather information and provide context.
• Adding a smiley face at the end of your prompt can increase the performance of the model.
• Small changes like these can result in a 1-2% increase in performance, which may not be noticeable for short answers but can make a difference for longer text generations.

The launch of GPTs and the OpenAI Store (26m5s)

• OpenAI's GPTs (Generative Pre-trained Transformers) are free-to-use tools that allow users to create customized versions of ChatGPT. In the future, creators may be able to charge for access to their GPTs.
• GPTs offer various features such as file uploads, custom instructions, a built-in code interpreter, browsing, image generation, and the ability to connect to external APIs.
• Non-developers can use GPTs to solve complex problems by providing sufficient context. OpenAI aims to make more capabilities accessible to non-coders, such as one-click sign-in for external APIs.
• Zapier's integration with GPTs allows users to integrate various third-party services without coding knowledge.
• Universal Primer, created by M Siki, is an educational tool that uses the Socratic method to help users learn complex concepts interactively.
• Arcade is an interactive demo platform that enables teams to create polished, unbranded demos quickly and easily. Companies like Quantum Metric have successfully used Arcade, resulting in a 2X higher conversion rate for demos and five times more engagement compared to videos.
• Arcade offers product customization options, designer-approved editing tools, and detailed insights into viewer engagement. Lenny Rachitsky recommends Arcade as a no-code solution for building personalized demos at scale and offers a 50% discount for the first three months.

The importance of high agency and urgency (32m10s)

• OpenAI prioritizes hiring individuals with high agency and a sense of urgency.
• High agency individuals take initiative to solve problems without needing excessive consensus.
• Employees at OpenAI are empowered to identify and address customer challenges promptly.
• OpenAI's culture encourages employees to take action rather than waiting for approvals.

OpenAI’s ability to move fast and ship high-quality products (34m35s)

• OpenAI values high agency and urgency in its employees.
• An example of high agency is when employees hear about a customer issue and immediately work on a solution.
• The development of the assistance API illustrates OpenAI's ability to quickly respond to developer feedback and build new products.
• OpenAI encourages employees to identify problems and collaborate to find solutions.

OpenAI’s planning process and decision-making criteria (35m56s)

• OpenAI's planning process focuses on projects that align with their mission of achieving Artificial General Intelligence (AGI).
• Reliability is a core principle, prioritizing robust and reliable experiences for customers over immediate rewards.
• OpenAI tracks success through various metrics such as adoption, revenue, and the number of developers building on their platform.
• Revenue is seen as a means to obtain more compute resources for training better models and achieving AGI.
• OpenAI's ambitious vision of building AGI sets it apart from other companies, and many companies are likely considering developing similar visions.

The importance of real-time communication (40m22s)

• OpenAI uses Slack for instantaneous real-time communication.
• Slack allows for quick coordination and collaboration across different teams and locations.
• Slack is an integral part of the company culture and facilitates efficient communication.
• Logan Kilpatrick heavily promotes the use of Slack and believes it is a superior tool compared to alternatives.
• The widespread use of Slack at OpenAI raises the question of whether some users might be AGIs rather than actual employees.
• Slack is developing AI tools that contribute to the overall progress in the field of AI.
• The availability of consumer-friendly AI products, such as Google Docs, excites Logan Kilpatrick as a user and demonstrates the practical applications of AI.

OpenAI’s team and growth (42m33s)

• OpenAI had around 780 employees towards the end of 2022.
• The team is still growing rapidly, with ongoing hiring across engineering and product management roles.
• OpenAI intentionally keeps its research team small to foster innovation.
• Adding new researchers can decrease overall productivity unless they significantly increase the efficiency of the entire group.
• This trade-off is unique to GPU-constrained research environments and may not apply to all fields.

Future developments at OpenAI (44m47s)

• OpenAI plans to introduce new modalities and expand the ways users can interact with AI through ChatGPT.
• The goal is to move towards an agent-like future where users can give instructions and receive thoughtful responses without immediate answers.
• ChatGPT serves as a stepping stone to bring AI to a wider audience and introduce its capabilities through specific problem-solving templates.
• Narrow vertical tools that address specific problems will be a key factor in onboarding more users.
• OpenAI is exploring additional interfaces, such as voice and audio, to facilitate easier interaction with its tools.

GPT-5 and building toward the future (47m42s)

• Speculation about GPT-5's capabilities is high, with some expecting it to solve all the world's problems.
• OpenAI's GPT-4 was the first model where capabilities could be reliably predicted based on compute input.
• GPT-5 is expected to be extremely useful and solve new problems, but the fundamental problems in the world will remain the same.
• Vertical use cases will benefit significantly from GPT-5's capabilities.
• Unrealistic expectations, such as GPT-5 performing extraordinary tasks, should be avoided.
• GPT-5 will likely become a very effective tool, similar to GPT-4, and its widespread adoption will happen quickly.
• Planning for a world where people become accustomed to these tools quickly can provide an edge.
• Assuming that GPT-5 will completely change everything may be a downside and the wrong mental framing to have.

OpenAI’s enterprise offering and the value of sharing custom applications (50m38s)

• OpenAI offers an enterprise version of its products and services.
• Key features of the enterprise offering include:
  • The ability to build custom applications using the OpenAI API.
  • Access to an enterprise version of ChatGPT with features such as SSO and higher usage limits.
  • The ability to share custom prompt templates and GPTs internally within an organization.
  • Enhanced security controls and data protection.

New updates and features from OpenAI (52m30s)

• Updated GPT-4 Turbo model.
• Updated preview model released at Dev day fixes the 'laziness phenomenon'.
• Third generation embedding model with improved non-English performance.
• Five times cheaper than previous embedding models.
• Can embed 62,000 pages of text for $1.

How to leverage OpenAI’s technology in products (55m9s)

• New experiences beyond chat interfaces provide an advantage.
• Taking use cases to the next level and building core experiences from the ground up is valuable.
• Providing data-grounded answers to specific questions, rather than just a few examples, enhances the product experience.
• Visual Electric.com is an example of a product that takes AI-powered image generation to a new level, specifically for graphic designers.

Encouragement for building with AI (58m26s)

• Logan encourages people with cool ideas to build with AI, as there are many things that need to be built for the world using AI.
• If anyone needs help getting started on their journey of building something with AI, Logan and others at OpenAI are willing to assist.
• The best way to reach out to Logan is through Twitter or LinkedIn. His Twitter handle is @LoganKilpatrick, and his name should show up as Logan GPT or GP Official Logan K.