Connect with us

Noticias

Meet the Power Players at OpenAI

Published

on

  • OpenAI has been elevating research and technical talent to leadership roles after recent departures.
  • The company has also brought on some new faces to fill the vacancies in its executive suite.
  • Here are some of the key people to watch going forward.

Last year, OpenAI found itself navigating a storm of departures. Recently, the company has been busy elevating its research and technical talent to leadership positions while strategically bringing in new hires to patch up the holes in its executive suite.

This shuffle in leadership couldn’t come at a more critical time, as the company faces intensified competition from heavyweights like Microsoft, Google, Anthropic, and Elon Musk’s xAI. Staying ahead means securing top-flight talent is essential. After all, “OpenAI is nothing without its people,” or so employees declared on social media after the failed Sam Altman ouster.

Meanwhile, the company is juggling a cascade of legal challenges, from copyright lawsuits to antitrust scrutiny, all while navigating the shifting sands of regulatory guidance under President Donald Trump. On top of that, OpenAI is trying to restructure as a for-profit business, raise tens of billions of dollars, and build new computer data centers in the US to develop its tech.

It’s a high-wire act that hinges on the expertise and execution of its new and newly promoted leaders. Below are some of the key power players who are helping to shape OpenAI’s future.


Leadership


Nextdoor CEO Sarah Friar sits on stage in front of a blue background smiling.

Sarah Friar.

Photo By Stephen McCarthy/Sportsfile via Getty Images



Sarah Friar, chief financial officer

Friar joined last year as the company’s first financial chief and a seasoned addition to the new guard. Formerly Square’s CFO, Friar knows how to turn a founder’s vision into a story that investors want to be a part of. She took two companies public: Square and Nextdoor, the hyperlocal social network she led through explosive growth during pandemic lockdowns.

At OpenAI, Friar leads a finance team responsible for securing the funds required to build better models and the data centers to power them. In her first few months on the job, she helped the company get $6.5 billion in one of the biggest private pools of capital in startup history.

She inherited a business with a colossal consumer-facing business and high-profile partnerships with Microsoft and Apple. At the same time, OpenAI is burning through billions of dollars as it seeks to outpace increasingly stiff competition from Google, Meta, and others. Friar is expected to bring much-needed financial acumen to OpenAI as the company moves to turn its research into mass-market products and a profitable business.

Jason Kwon, chief strategy officer

In his role as chief strategy officer, Kwon helps set the agenda for a slew of non-research initiatives, including the company’s increasingly active outreach to policymakers and the various legal challenges swirling around it. His background as the company’s former general counsel gives him a strong foundation in navigating complex legal and regulatory landscapes.

Kwon works closely with Anna Makanju, the VP of global impact, and Chris Lehane, the VP of global affairs, as they seek to build and strengthen OpenAI’s relationships in the public sector.

Kwon was previously general counsel at the famed startup accelerator Y Combinator and assistant general counsel at Khosla Ventures, an early investor in OpenAI.

Che Chang, general counsel

Being at the forefront of artificial intelligence development puts OpenAI in a position to navigate and shape a largely uncharted legal territory. In his role as general counsel, Chang leads a team of attorneys who address the legal challenges associated with the creation and deployment of large language models. The company faces dozens of lawsuits concerning the datasets used to train its models and other privacy complaints, as well as multiple government investigations.

OpenAI’s top lawyer joined the company after serving as senior corporate counsel at Amazon, where he advised executives on developing and selling machine learning products and established Amazon’s positions on artificial intelligence policy and legislation. In 2021, Chang took over for his former boss, Jason Kwon, who has since become chief strategy officer.

Kevin Weil, chief product officer


Kevin Weil talking and making gestures with his hands while speaking at Web Summit.

Kevin Weil.

Photo by Horacio Villalobos/Corbis via Getty Images



If Sam Altman is OpenAI’s starry-eyed visionary, Weil is its executor. He leads a product team that turns blue-sky research into products and services the company can sell.

Weil joined last year as a steady-handed product guru known for playing key roles at large social networks. He was a longtime Twitter insider who created products that made the social media company money during a revolving door of chief executives. At Instagram, he helped kneecap Snapchat’s growth with competitive product releases such as Stories and live video.

Weil is expected to bring much-needed systems thinking to OpenAI as the company moves to turn its research into polished products for both consumer and enterprise use cases.

Nick Turley, ChatGPT’s head of product

In the three years since ChatGPT burst onto the scene, it has reached hundreds of millions of active users and generated billions in revenue for its maker. Turley, a product savant who leads the teams driving the chatbot’s development, is behind much of ChatGPT’s success.

Turley joined in 2022 after his tenure at Instacart, where he guided a team of product managers through the pandemic-driven surge in demand for grocery delivery services.

OpenAI’s chatbot czar is likely to play a crucial role as the company expands into the enterprise market and adds more powerful, compute-intensive features to its famed chatbot.

Srinivas Narayanan, vice president of engineering

Narayanan was a longtime Facebook insider who worked on important product releases such as Facebook Photos and tools to help developers build for its virtual reality headset, Oculus. Now, he leads the OpenAI teams responsible for building new products and scaling its systems. This includes ChatGPT, which is used by over 400 million people weekly; the developer platform, which has doubled usage over the past six months; and the infrastructure needed to support both.

Research

Jakub Pachocki, chief scientist

Ilya Sutkever’s departure as chief scientist last year prompted questions about the company’s ability to stay on top of the artificial intelligence arms race. That has thrust Pachocki into the spotlight. He took on the mantle of chief scientist after seven years as an OpenAI researcher.

Pachocki had already been working closely with Sutskever on some of OpenAI’s most ambitious projects, including an advanced reasoning model now known as o1. In a post announcing his promotion, Sam Altman called Pachocki “easily one of the greatest minds of our generation.”

Mark Chen, senior vice president of research

A flurry of executive departures also cast Chen into the highest levels of leadership. He was promoted last September following the exit of Bob McGrew, the company’s chief research officer. In a post announcing the change, Altman called out Chen’s “deep technical expertise” and commended the longtime employee as having developed as a manager in recent years.

Chen’s path to OpenAI is a bit atypical compared to some of his colleagues. After studying computer science and mathematics at MIT, he began his career as a quantitative trader on Wall Street before joining OpenAI in 2018. Chen previously led the company’s frontier research.

He has been integral to OpenAI’s efforts to expand into multimodal models, heading up the team that developed DALL-E and the team that incorporated visual perception into GPT-4. Chen was also an important liaison between employees and management during Sam 0Altman’s short-lived ouster, further cementing his importance within the company.

Liam Fedus, vice president of research, post-training

Fedus helps the company get new products out the door. He leads a post-training team responsible for taking the company’s state-of-the-art models and improving their performance and efficiency before it releases them to the masses. Fedus was the third person to lead the team in a six-month period following the departures of Barret Zoph and Bob McGrew last year.

Fedus was also one of seven OpenAI researchers who developed a group of advanced reasoning models known as Strawberry. These models, which can think through problems and complete tasks they haven’t encountered before, represented a significant leap at launch.

Josh Tobin, member of technical staff

Tobin, an early research scientist at OpenAI, left to found Gantry, a company that assists teams in determining when and how to retrain their artificial intelligence systems. He returned to OpenAI last September and now leads a team of researchers focused on developing agentic products. Its flashy new agent, Deep Research, creates in-depth reports on nearly any topic.

Tobin brings invaluable experience in building agents as the company aims to scale them across a wide range of use cases. In a February interview with Sequoia, Tobin explained that when the company takes a reasoning model, gives it access to the same tools humans use to do their jobs, and optimizes for the kinds of outcomes it wants the agent to be able to do, “there’s really nothing stopping that recipe from scaling to more and more complex tasks.”

Legal

Andrea Appella, associate general counsel for Europe, Middle East, Asia

Appella joined last year, bolstering the company’s legal firepower as it navigated a thicket of open investigations into data privacy concerns, including from watchdogs in Italy and Poland. Appella is a leading expert on competition and regulatory law, having previously served as head of global competition at Netflix and deputy general counsel at 21st Century Fox.

Regulatory scrutiny could still prove to be an existential threat to OpenAI as policymakers worldwide put guardrails on the nascent artificial intelligence industry. Nowhere have lawmakers been more aggressive than in Europe, which makes Appella’s role as the company’s top legal representative in Europe one of the more crucial positions in determining the company’s future.

Haidee Schwartz, associate general counsel for competition

OpenAI has spent the last year beefing up its legal team as it faces multiple antitrust probes. Schwartz, who joined in 2023, knows more about antitrust enforcement than almost anyone in Silicon Valley, having seen both sides of the issue during her storied legal career.

Between 2017 and 2019, she served as the acting deputy director of the Bureau of Competition at the Federal Trade Commission, one of the agencies currently investigating Microsoft’s agreements with OpenAI. Schwartz also advised clients on merger review and antitrust enforcement as a partner at law firm Akin Gump. She’ll likely play an important role in helping OpenAI navigate the shifting antitrust landscape in President Donald Trump’s second term.

Heather Whitney, copyright counsel

Whitney serves as lead data counsel at OpenAI, placing her at the forefront of various legal battles with publishers that have emerged in recent years. She joined the company last January, shortly after The New York Times filed a copyright lawsuit against OpenAI and its corporate backer, Microsoft. OpenAI motioned to dismiss the high-profile case last month.

Whitney’s handling of these legal cases, which raise new questions about intellectual property in relation to machine learning, will be crucial in deciding OpenAI’s future.

Previously, Whitney worked at the law firm Morrison Foerster, where she specialized in novel copyright issues related to artificial intelligence and was a member of the firm’s AI Steering Committee. Prior to her official hiring, she had already been collaborating with OpenAI as part of Morrison Foerster, which is among several law firms offering external counsel to the company.

Policy

Chan Park, head of US and Canada policy and partnerships

Before OpenAI had a stable of federal lobbyists, it had Park. In 2023, the company registered the former Microsoft lobbyist as its first in-house lobbyist, marking a strategic move to engage more actively with lawmakers wrestling with artificial intelligence regulation.

Since then, OpenAI has beefed up its lobbying efforts as it seeks to build relationships in government and influence the development of artificial intelligence policy. It’s enlisted white-shoe law firms and at least one former US senator to plead OpenAI’s case in Washington. The company also spent $1.76 million on government lobbying in 2024, a sevenfold increase from the year before, according to a recent disclosure reviewed by the MIT Technology Review.

Park has been helping to guide those efforts from within OpenAI as the company continues to sharpen its message around responsible development of artificial intelligence.

Anna Makanju, vice president of global impact

Referred to as OpenAI’s de facto foreign minister, Makanju is the mastermind behind Sam Altman’s global charm offensive. On multiple trips, he met with world leaders, including the Indian prime minister and South Korean president, to discuss the future of artificial intelligence.

The tour was part of a broader effort to make Altman the friendly face of a nascent industry and ensure that OpenAI will have a seat at the table when designing artificial intelligence regulations and policies. Makanju, a veteran of Starlink and Facebook who also served as a special policy advisor to former President Joe Biden, has been integral in that effort.

In addition to helping Altman introduce himself on the world stage, she has played an important role in expanding OpenAI’s commercial partnerships in the public sector.

Chris Lehane, vice president of global affairs


FILE PHOTO: Airbnb head of global policy and public affairs Chris Lehane speaks to Reuters in Los Angeles, California, U.S. November 17, 2016.  REUTERS/Phil McCarten

Chris Lehane.

Thomson Reuters



Lehane joined OpenAI last year to help the company liaise with policymakers and navigate an uncharted political landscape around artificial intelligence. The veteran political operative and “spin master” played a similar role at Airbnb, where he served as head of global policy and public affairs from 2015 to 2022 and helped it address growing opposition from local authorities.

He previously served in the Clinton White House, where Newsweek referred to him as a “master of disaster” for his handling of the scandals and political crises that plagued the administration.

Lehane is poised to play a crucial role in ensuring that the United States stays at the forefront of the global race in artificial intelligence. When President Trump introduced Stargate, a joint venture between OpenAI, Oracle, and SoftBank aimed at building large domestic data centers, Lehane was on the scene. From Washington, he traveled to Texas to meet with local officials, engaging in discussions about how the state could meet the rapidly growing demand for energy.

Lane Dilg, head of infrastructure policy and partnerships

In her newly appointed role, Dilg works to grease the wheels for the construction of giant data centers needed to build artificial intelligence. She took on the position in January after two years as head of strategic initiatives for global affairs, working with government agencies, private industry, and nonprofit organizations to ensure that artificial intelligence benefits all of humanity.

In hiring Dilg, OpenAI gained an inside player in the public sector. Dilg is a former senior advisor to the undersecretary of infrastructure at the US Department of Energy and was interim city manager for Santa Monica, California, managing the city through the COVID-19 pandemic.

Dilg will undoubtedly play an important role in expanding and nurturing OpenAI’s relationships in Washington as it seeks to secure President Trump’s support for building its own data centers.

Have a tip? Contact this reporter via email at mrussell@businessinsider.com or Signal at meliarussell.01. Use a personal email address and a nonwork device; here’s our guide to sharing information securely.

Darius Rafieyan contributed to an earlier version of this story.

Continue Reading

Noticias

Operai deja a Microsoft Copilot en el polvo con memes de Gibli

Published

on

He estado a bordo del tren de IA generativo desde los primeros días de Openai, antes del lanzamiento de Chatgpt y, finalmente, Microsoft invirtiendo miles de millones de dólares en OpenAi.

En el papel, Microsoft se supone que se encuentra entre las principales compañías tecnológicas en el espacio de IA, debido a su inversión multimillonaria y los estrechos lazos con OpenAI como su mayor inversor y proveedor exclusivo de la nube (bueno, al menos hasta que SoftBank bombardeó el mejor “Bromance” tecnológico en la historia con su ambiente proyecto de $ 500 mil millones).

Continue Reading

Noticias

Llega la búsqueda profunda abierta para desafiar la perplejidad y la búsqueda de chatgpt

Published

on

Únase a nuestros boletines diarios y semanales para obtener las últimas actualizaciones y contenido exclusivo sobre la cobertura de IA líder de la industria. Obtenga más información


Los investigadores de la Fundación Sentient han publicado Open Deep Search (ODS), un marco de código abierto que puede igualar la calidad de las soluciones de búsqueda de IA patentadas, como la perplejidad y la búsqueda de CHATGPT. ODS equipa modelos de idiomas grandes (LLM) con agentes de razonamiento avanzado que pueden usar la búsqueda web y otras herramientas para responder preguntas.

Para las empresas que buscan herramientas de búsqueda de IA personalizables, ODS ofrece una alternativa convincente y de alto rendimiento a las soluciones comerciales cerradas.

El paisaje de búsqueda de IA

Las herramientas de búsqueda de IA modernas como Perplexity y ChatGPT Search pueden proporcionar respuestas actualizadas combinando las capacidades de conocimiento y razonamiento de LLMS con la búsqueda web. Sin embargo, estas soluciones son típicamente patentadas y de código cerrado, lo que dificulta la personalización de ellas y adoptarlas para aplicaciones especiales.

“La mayoría de la innovación en la búsqueda de IA ha sucedido a puerta cerrada. Los esfuerzos de código abierto se han retrasado históricamente en usabilidad y rendimiento”, dijo a VentureBeat de Himanshu Tyagi, cofundador de Sentient. “ODS tiene como objetivo cerrar esa brecha, mostrando que los sistemas abiertos pueden competir, e incluso superar, las contrapartes cerradas sobre la calidad, la velocidad y la flexibilidad”.

Arquitectura de búsqueda profunda (ODS) abierta

Open Deep Search (ODS) está diseñado como un sistema plug-and-play que puede integrarse con modelos de código abierto como Deepseek-R1 y modelos cerrados como GPT-4O y Claude.

ODS comprende dos componentes centrales, ambos aprovechando la base elegida LLM:

Herramienta de búsqueda abierta: Este componente toma una consulta y recupera información de la web que se puede dar al LLM como contexto. La herramienta de búsqueda Open realiza algunas acciones clave para mejorar los resultados de búsqueda y asegurarse de que proporcione un contexto relevante para el modelo. Primero, reformula la consulta original de diferentes maneras para ampliar la cobertura de búsqueda y capturar diversas perspectivas. Luego, la herramienta obtiene resultados de un motor de búsqueda, extrae el contexto de los resultados superiores (fragmentos y páginas vinculadas), y aplica técnicas de fragmentación y reanimación para filtrar el contenido más relevante. También tiene un manejo costumbre para fuentes específicas como Wikipedia, ARXIV y PubMed, y puede solicitarse a priorizar fuentes confiables al encontrar información contradictoria.

Agente de razonamiento abierto: Este agente recibe la consulta del usuario y utiliza la base LLM y varias herramientas (incluida la herramienta de búsqueda abierta) para formular una respuesta final. Sentient proporciona dos arquitecturas de agentes distintos dentro de ODS:

ODS-V1: Esta versión emplea un marco de agente React combinado con el razonamiento de la cadena de pensamiento (COT). Los agentes reaccionados intercalan pasos de razonamiento (“pensamientos”) con acciones (como usar la herramienta de búsqueda) y las observaciones (los resultados de las herramientas). ODS-V1 usa reaccionar iterativamente para llegar a una respuesta. Si el agente React lucha (según lo determinado por un modelo de juez separado), es predeterminado a una autoconsistencia de COT, que muestra varias respuestas de cuna del modelo y usa la respuesta que aparece con más frecuencia.

ODS-V2: Esta versión aprovecha la cadena de código (COC) y un agente CodeAct, implementado utilizando la biblioteca de Sumolagents Face. COC utiliza la capacidad de LLM para generar y ejecutar fragmentos de código para resolver problemas, mientras que CodeAct usa la generación de código para las acciones de planificación. ODS-V2 puede orquestar múltiples herramientas y agentes, lo que le permite abordar tareas más complejas que pueden requerir una planificación sofisticada y iteraciones de búsqueda potencialmente múltiples.

Agente de razonamiento abierto ODS
Crédito de arquitectura ODS: ARXIV

“Si bien herramientas como ChatGPT o Grok ofrecen ‘investigación profunda’ a través de agentes de conversación, ODS opera en una capa diferente, más similar a la infraestructura detrás de la perplejidad de IA, que proporciona la arquitectura subyacente que impulsa la recuperación inteligente, no solo los resúmenes”, dijo Tyagi.

Rendimiento y resultados prácticos

Sentient evaluó ODS emparejándolo con el modelo de código abierto Deepseek-R1 y probándolo contra competidores populares de código cerrado como Perplexity AI y la vista previa de búsqueda GPT-4O de OpenAI, así como LLMS independientes como GPT-4O y LLAMA-3.1-70B. Usaron los marcos y los puntos de referencia de SimpleQA Pregunta-Respuesta, adaptándolos para evaluar la precisión de los sistemas de IA habilitados para la búsqueda.

Los resultados demuestran la competitividad de ODS. Tanto ODS-V1 como ODS-V2, cuando se combinan con Deepseek-R1, superaron a los productos insignia de Perplexity. En particular, ODS-V2 combinado con Deepseek-R1 superó la vista previa de búsqueda GPT-4O en el complejo punto de referencia de marcos y casi lo coincidió en SimpleQA.

Una observación interesante fue la eficiencia del marco. Los agentes de razonamiento en ambas versiones de ODS aprendieron a usar la herramienta de búsqueda juiciosamente, a menudo decidieron si era necesaria una búsqueda adicional en función de la calidad de los resultados iniciales. Por ejemplo, ODS-V2 utilizó menos búsquedas web en las tareas SimpleQA más simples en comparación con las consultas más complejas y múltiples en marcos, optimizando el consumo de recursos.

Implicaciones para la empresa

Para las empresas que buscan potentes capacidades de razonamiento de IA basadas en información en tiempo real, ODS presenta una solución prometedora que ofrece una alternativa transparente, personalizable y de alto rendimiento a los sistemas de búsqueda de IA patentados. La capacidad de enchufar LLM y herramientas de código abierto preferidos brinda a las organizaciones un mayor control sobre su pila de IA y evita el bloqueo del proveedor.

“ODS fue construido con modularidad en mente”, dijo Tyagi. “Selecciona qué herramientas usar dinámicamente, en función de las descripciones proporcionadas en la solicitud. Esto significa que puede interactuar con herramientas desconocidas con fluidez, siempre y cuando estén bien descritadas, sin requerir exposición previa”.

Sin embargo, reconoció que el rendimiento de ODS puede degradarse cuando el conjunto de herramientas se hincha, “un diseño tan cuidadoso importa”.

Sensient ha lanzado el código para ODS en GitHub.

“Inicialmente, la fuerza de la perplejidad y el chatgpt era su tecnología avanzada, pero con ODS, hemos nivelado este campo de juego tecnológico”, dijo Tyagi. “Ahora nuestro objetivo es superar sus capacidades a través de nuestra estrategia de ‘Entradas abiertas y salidas abiertas’, lo que permite a los usuarios integrar sin problemas a los agentes personalizados en un chat sensible”.

Continue Reading

Trending