Chronological feed of everything captured from Greg Brockman.
You can now just build amazing voice agents, with the GPT-Realtime-2 reasoning model in our API:
Systems and Algorithms for Convolutional Multi-Hybrid Language Models at Scale — We introduce convolutional multi-hybrid architectures, with a design grounded on two simple observations. First, operators in hybrid models can be tailored to token manipulation tasks such as in-context recall, multi-token recall, and compression, with input-dependent convolutions and attention offering complementary performance. Second, co-designing convolution operators and hardware-aware algorithms enables efficiency gains in regimes where previous alternative architectures struggle to surpas — Citations: 15.
Genome modeling and design across all domains of life with Evo 2 — All of life encodes information with DNA. While tools for sequencing, synthesis, and editing of genomic code have transformed biological research, intelligently composing new biological systems would also require a deep understanding of the immense complexity encoded by genomes. We introduce Evo 2, a biological foundation model trained on 9.3 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life. We train Evo 2 with 7B and 40B parameters to have an unprecedente — Citations: 224.
OpenAI introduces gpt-oss-120b and gpt-oss-20b, open-weight language models designed for advanced reasoning with an efficient mixture-of-expert transformer architecture. These models, trained with distillation and reinforcement learning, demonstrate strong agentic capabilities, offering a balance of accuracy and inference cost. Their release under an Apache 2.0 license promotes broader research and application, particularly in areas requiring deep research browsing, Python tool use, and flexible instruction following.
Most LLMs universally excel at generating code for high-resource programming languages (HRPLs) like Python , a capability that has become standard due to the abundance of training data. However, they struggle significantly with low-resource programming languages (LRPLs) such as D , exacerbating the digital divide. This gap limits developers using LRPLs from equally benefiting and hinders innovation within underrepresented programming communities. To ma — Citations: 3.
A modern, simple and very fast Mysql library for Ruby - binding to libmysql. Stars: 2279
A Really Ruby Mail Library. Stars: 3661
Active Merchant is a simple payment abstraction library extracted from Shopify. The aim of the project is to feel natural to Ruby users and to abstract as many parts as possible away from the user to offer a consistent interface across all supported gateways.. Stars: 4596
Active Utils extracts commonly used modules and classes used by Active Merchant, Active Shipping, and Active Fulfillment. Stars: 95
ruby bindings to libgit2. Stars: 2288
A cross-platform, linkable library implementation of Git that you can use in your application.. Stars: 10401
Ruby MIME type registry library. Stars: 337
The Official MongoDB Ruby Driver. Stars: 1431
The Ruby Programming Language. Stars: 23504
Open-source headless eCommerce platform with REST API, TypeScript SDK, and Next.js storefront for cross-border, B2B or marketplace eCommerce.. Stars: 15332
Greg Brockman, a co-founder of Stripe, announced his departure from the company after helping to scale it from a small startup to a global enterprise. He cited a desire to create something new and the feeling of having reached the limit of his learning about company building at Stripe as key motivations. Brockman emphasized that his decision was not due to any issues with Stripe, which he states is performing exceptionally well, but rather a personal drive to explore new entrepreneurial endeavors.
OpenAI is prioritizing the development of a unified "Super App" that integrates chat, browsing, and coding functionalities, moving away from specialized models like Sora for video generation (except in robotics research). This strategic shift is driven by the belief in imminent AGI, rapid technological maturation, and the need to focus resources on applications with the most significant real-world impact, particularly in knowledge work and personal assistance. The company faces the challenge of scaling compute while navigating public skepticism and competitive pressures.
The Codex app server provides a unified platform for developing and deploying agentic applications, enabling seamless integration between different devices and environments. It allows users to leverage their existing ChatGPT accounts and offers direct app building capabilities on its infrastructure. This design streamlines the development experience by providing automatic access to user-specific data such as skills, agents, sessions, and prompts across various devices.
Large Language Models are proving to be invaluable in real-world scenarios for managing critical healthcare information across multiple family members. They enable families to consolidate, comprehend, and act upon vast amounts of fragmented medical data, significantly empowering them within disjointed healthcare systems. This application highlights their immediate, practical utility beyond current limitations in areas like design or API stability, demonstrating their potential to improve patient advocacy and decision-making.
A field experiment with 515 startups demonstrated that exposure to AI case studies significantly increases AI adoption. This increased adoption correlates with nearly double the revenue and substantially reduced capital requirements, indicating that AI use is a critical factor in accelerating business growth and fostering entrepreneurship. The primary barrier to leveraging these benefits appears to be a lack of understanding regarding practical AI applications.
OpenAI's Codex platform now features a direct integration with Vercel, enabling developers to deploy applications with enhanced efficiency. This partnership streamlines the process from project setup to deployment, leveraging Codex's capabilities to facilitate rapid application delivery through the Vercel plugin. The integration is designed to simplify developer workflows and accelerate the deployment pipeline.
OpenAI has hired a developer specializing in Codex, their AI model for code generation. This recruitment suggests a continued and potentially accelerated investment in advancing AI-driven code development capabilities. The new hire is actively soliciting feedback from the community, indicating a user-centric approach to future development.
The Codex App has emerged as the primary interaction surface for users, overtaking traditional developer tools like CLIs and IDE extensions. This shift indicates a preference for standalone application interfaces over integrated plugin environments for AI-assisted coding. OpenAI is currently accelerating enterprise adoption through credit-based incentives.
Codex has updated its pricing model to eliminate upfront commitments and reduce annual seat costs. This strategy aims to lower the barrier to entry for teams and individuals, encouraging wider adoption and usage of the AI coding assistant.
AI presents significant opportunities for entrepreneurs, particularly in enabling rapid development and market penetration, as exemplified by "vibecoded" companies. However, this acceleration also introduces substantial risks, including a propensity for ethical shortcuts, regulatory non-compliance, and deceptive practices. The case of Medvi highlights both the potential for high revenue with minimal staffing and the pitfalls of neglecting regulatory integrity and marketing ethics.
OpenAI's internal models have successfully generated short and elegant proofs for three previously unresolved mathematical problems posed by Erdős, signaling a potential paradigm shift in scientific discovery. This advancement highlights the growing capability of AI in complex problem-solving within mathematics, and suggests a future where AI significantly accelerates the pace of scientific breakthroughs.
OpenAI has launched a gallery of Codex use cases, demonstrating practical applications for both coding and non-coding tasks. These use cases are designed to function like "skills" for human users, providing starter prompts that can be directly opened within the Codex application. This initiative aims to broaden the accessibility and utility of Codex beyond traditional coding environments.
OpenAI has rolled out plugin functionality for Codex, allowing it to seamlessly integrate with popular developer tools. This integration aims to enhance the utility of Codex by enabling direct interaction with platforms such as Slack, Figma, Notion, and Gmail, streamlining workflows for builders.
The content suggests the potential application of GPT-5.4 in frontend development. This indicates an expanding role for large language models in specialized technical domains beyond general natural language processing, potentially automating or assisting in code generation and UI/UX design. Further information is needed to determine the specific capabilities and implications.
Greg Brockman notes a substantial improvement in Codex's capabilities, specifically in its ability to identify bugs and errors in task planning. This development suggests advancements in AI-driven code analysis and automated planning, potentially enhancing software development workflows and autonomous systems. The observation is a direct affirmation of progress by a key figure in AI development.
Codex now supports subagents, enabling developers to organize complex tasks, parallelize work, and maintain a cleaner primary context window. This feature allows for more efficient large-scale project execution by delegating specialized functions to individual agents and providing real-time steering.
OpenAI has launched GPT-5.4 Mini, a new model designed for advanced coding, computer interaction, multimodal understanding, and subagent functionality. This iteration is also noted for its performance, boasting a two-fold speed increase compared to its predecessor, GPT-5 Mini. The model is immediately available across OpenAI platforms including ChatGPT, Codex, and via its API.
GPT-5.4 has demonstrated unprecedented rapid adoption within its first week on the API, processing 5 trillion tokens daily. This volume surpasses the entire API usage from a year prior, indicating substantial scalability and demand. The model has also generated an annualized net-new revenue run rate of $1 billion, highlighting its immediate economic impact.
AI models are demonstrating the capability to generate functional game environments within Excel, complete with integrated AI opponents. This indicates a potential for leveraging AI in complex spreadsheet-based applications beyond traditional data processing, extending to interactive and simulated environments. The varying success across models highlights differences in their ability to translate high-level requests into executable, formula-driven logic.
AI tools, such as ChatGPT, are capable of handling comprehensive real estate sales processes, from pricing and marketing to drafting contracts. This demonstrates AI's potential to significantly expedite property transactions. Increased exposure to and utilization of these AI tools correlates with greater user confidence in their capabilities.
Once considered an "impossible" internal goal, AI models can now generate coherent 1000-line programs. This signifies substantial progress in AI's code generation capabilities, far exceeding previous expectations. The rapid advancement underscores a significant shift in what is considered achievable in AI-assisted software development.
OpenAI CTO Greg Brockman discusses the company's internal culture, including the lessons learned from past conflicts and their current "agent-first" approach to software development. He highlights the rapid advancements in AI, emphasizing the exponential pace of breakthroughs and the increasing integration of AI into various aspects of work and daily life. Brockman also shares insights into OpenAI's origin, its relationship with Microsoft, and the philosophical underpinnings of developing AGI for the benefit of humanity, emphasizing the importance of resilience and diverse applications.
The increasing consensus among AI developers and researchers points to a significant and ongoing need for more powerful computational resources, specifically GPUs. This demand is driven by the escalating complexity of AI models and the computational intensity of training and inference processes. The industry anticipates continued growth in this area, underscoring GPUs as a critical bottleneck.
The concept of a "token budget" for AI models, while currently a niche consideration, is predicted to become a significant factor in future employment negotiations. This shift reflects the increasing integration of AI into workflows and the potential for access to computational resources to become a competitive advantage or constraint for employees.
This discussion with OpenAI President Greg Brockman highlights the transformative potential of AI beyond simple task automation. It emphasizes the concept of AI as a "goal amplifier," where intelligent agents proactively assist users in achieving objectives. The conversation also delves into the evolving nature of engineering and business in an AI-first world, stressing the importance of rapid iteration, domain expertise, and adapting workflows to leverage AI's capabilities.
OpenAI views Artificial General Intelligence (AGI) as a continuous process, not a fixed destination, driven by scaling compute and algorithmic advancements. The company prioritizes compute as the fundamental driver of innovation, actively managing its scarcity and advocating for increased supply. This strategy underpins their approach to model development, the evolution of user interfaces, and the broader societal impact of AI.