GPT-5 Pro for Developers Tease
Greg Brockman hinted at a "GPT-5 Pro for developers" offering. This suggests a potential new tier or specialized version of the GPT-5 model designed with features or optimizations tailored for developer use cases.
Chronological feed of everything captured from Greg Brockman.
Greg Brockman hinted at a "GPT-5 Pro for developers" offering. This suggests a potential new tier or specialized version of the GPT-5 model designed with features or optimizations tailored for developer use cases.
OpenAI is heavily investing in agentic AI for coding, emphasizing the co-evolution of model intelligence and user interfaces (harnesses). The company prioritizes both general AI capabilities and domain-specific optimizations for coding, exemplified by GPT-5 Codex. Future efforts focus on scalable oversight, AI-driven novel problem-solving, and increasing compute supply to support a future with widespread, continuously operating AI agents.
GPT-5 is OpenAI's first hybrid model, combining a fast non-reasoning model with a deep reasoning model via an internal router—an architectural acknowledgment that adaptive compute is more practically achieved through model orchestration than within a single architecture. The reasoning paradigm, rooted in reinforcement learning, emerged directly after GPT-4 as the identified mechanism to close the reliability gap: generating thousands of rollouts per task, learning from environment-grounded feedback, and compounding general problem-solving skills that transfer across domains (e.g., IMO proofs transferring to IOI code competition performance). Brockman frames compute as the primary bottleneck and fuel for intelligence, with inference costs dropping ~1,000x in 2.5 years since GPT-4 launch, and positions open-source model releases as a strategic move to anchor an American AI technology stack. The near-term frontier is seamless human-AI co-development—persistent agents that fluidly operate across local and remote environments, with layered security (instruction hierarchy, sandboxing) and codebases deliberately restructured around AI-optimized, self-contained modules.
Greg Brockman reveals that OpenAI did not set out to prove the scaling hypothesis — they observed it empirically during the Dota 2 project, where doubling compute consistently doubled performance with no plateau. The company was then built "backwards" from the typical startup playbook: chasing the technology without a defined problem, releasing a general API, and following emergent use cases. Brockman frames the current AI moment as sitting at "Level 3" of a 5-level AGI framework (chatbots → reasoners → agents → innovators → organizations), with energy infrastructure — not algorithms or data — now emerging as the primary bottleneck to further scaling.
AI has reached a utility threshold where advanced models are highly functional. This progress is primarily driven by precision execution on large-scale models, requiring significant compute and strong software engineering skills. Engineers, even without prior ML experience, are crucial for building, scaling, and managing these complex systems, and their contributions are as vital as those of researchers in advancing the field.
Greg Brockman, OpenAI co-founder and CTO, spent three years failing to transition into machine learning despite being embedded at one of the world's top AI labs — not due to lack of resources or intelligence, but due to psychological resistance to feeling like a beginner. His eventual breakthrough came from a deliberate three-month self-study sabbatical, a project-driven learning approach, and the willingness to make substantive modifications to existing ML codebases. His core thesis: the technical barrier to becoming an ML practitioner is lower than most experienced engineers assume, and the dominant blocker is identity-level discomfort with incompetence, not the material itself.
AGI development relies on a symbiotic relationship between generalizable algorithmic ideas and massive computational scaling. While scale often triggers qualitatively new emergent behaviors—such as long-term planning in OpenAI Five—the core goal is to establish positive initial conditions and technical safety mechanisms that align these scalable systems with human values.
OpenAI Five, an AI-powered agent, competed against human world champions in Dota 2, demonstrating advanced capabilities in deep reinforcement learning. This event highlighted the AI's ability to learn complex strategies through self-play, accumulating 45,000 years of game experience in 10 months. The underlying learning code is generalizable, evidenced by its prior application in robotic hand control, suggesting broad potential for future interactive AI systems. This marked a significant milestone for AI in esports, showcasing learned creativity and unexpected playstyles.
Deep learning's generality, competence, and scalability since 2012 enable sustained AI advances across domains like vision, speech, and robotics, making AGI—defined as systems outperforming humans at most economically valuable work—feasible on relevant timescales. AGI promises transformative applications, such as superhuman healthcare via scaled computerized doctors, but poses risks akin to unchecked companies, including misaligned goals or malicious subversion. OpenAI's new LP structure facilitates capital-raising to build safe AGI distributing benefits to all humanity, prioritizing preemptive safety measures.
OpenAI Five learns 5v5 Dota through self-play equivalent to 180 years of games daily, powered by five ant-brain-sized neural networks that enable teamwork, real-time strategy, and imperfect information handling without hand-coded rules. The system scaled from 1v1 pro victories to 5v5 feats using identical tech also applied to robotic dexterity. AI compute has doubled every 3.5 months since 2012, making such capabilities feasible now and commonplace soon.
doom-py offers Python wrappers around a modified bleeding-edge ViZDoom, enabling Python-based access to the Doom game AI research environment. It includes standard Python packaging files like setup.py and Makefile, with 13 commits on master but no releases. Ubuntu dependencies cover numpy, cmake, zlib, jpeg, boost, gcc, sdl2, wget, and unzip for building.
Greg Brockman recounts founding OpenAI by assembling a dream team around Ilya Sutskever's AI vision, initially handling all non-research tasks to maximize technical progress. He identifies engineering as the critical bottleneck in ML research, accelerating projects like Gym and Universe through intense, distraction-free coding sprints that unblock researchers. Organizational scaling now allows him to sustain a CTO role blending hands-on engineering with leadership, exemplified by hiring dedicated executors like Erika Reinhardt.
This script enables writing to CoreOS's read-only /usr by patching the rootfs superblock's ro_compat flag via dd to allow RW remount. It downloads and installs Google's safe_format_and_mount script to /usr/share/google/, required by Kubernetes for EBS volume handling, then restores RO state. Functions like disable_rw_mount and enable_rw_mount manipulate the filesystem at a precise byte offset (0x467) for idempotent operation.
Greg Brockman pivoted from high school chatbot experiments and college programming language research to scaling Stripe, before dedicating himself to safe human-level AI after recognizing deep learning's breakthrough potential. Key validations included impressive capabilities in image classification, speech recognition, and real-time translation via models that map diverse inputs to outputs using layered abstractions and backpropagation. A pivotal dinner with top talents like Ilya Sutskever, Elon Musk, and Paul Christiano crystallized the need for a nonprofit lab at AI's research frontier to ensure beneficial outcomes, leading Brockman to commit full-time.
Greg Brockman is joining the Stellar board. He believes digital currencies require protocols like Stellar that integrate with existing financial systems and prioritize user experience. Stellar, a non-profit, focuses on financial inclusion and has developed a provably-correct consensus algorithm, re-implemented its technology, and initiated a pilot program in South Africa.
Greg Brockman highlights the importance of collaborative environments for meaningful software development, referencing the Recurse Center as a prime example of a thoughtfully designed cultural dynamic. He notes that even experienced engineers seek such environments to explore new programming areas. Brockman
This content illustrates the process of creating a customer and subscribing them to a plan using the Stripe API. It highlights a common error when attempting to subscribe a customer without an attached payment method and demonstrates a successful subscription with a defined trial period. The output provides detailed JSON responses for both scenarios, offering insights into the API structure and response formats.
This document provides a technical overview of fundamental operations on the Stellar network, including account introspection, address creation, trustline establishment, credit issuance, and decentralized exchange functionalities. It details command-line interface (CLI) and Stellar helper functions for executing these operations, offering practical examples for developers. The content focuses on direct interaction with the Stellar network's core functionalities.
This content details a Keybase proof establishing the identity of GitHub user "gdb" on Keybase. The proof involves signing a JSON object containing Keybase, GitHub, and PGP key information with a specified PGP key, then publishing it as a GitHub Gist. This process validates the association between the GitHub account, the Keybase profile, and the PGP key.
Stellar allows users to issue custom tokens and facilitate their exchange via a decentralized exchange. This process involves creating an offer for the custom asset, which other users can then fulfill with counteroffers. The system emphasizes trust lines, where the value of an issued asset is dependent on the individual's trust in the issuer, enabling a flexible definition of currency across the network.
This GitHub Gist provides a minimal example of integrating Stripe for basic credit card payments using PHP and JavaScript. It demonstrates client-side tokenization with Stripe.js and a server-side PHP script for creating a charge. The code illustrates the essential steps for securely collecting payment information and processing a transaction through the Stripe API.