Sam Altman Aims for AGI by 2026, According to His Definition

OpenAI claims to be accelerating towards an AGI in line with its own definition, with Astra at the center and agents promised to the general public. At the same time, the company is increasing advertisements in ChatGPT, testing sponsored formats, and preparing a small voice device, while acknowledging that AGI remains a topic of discussion and is still out of reach today.

OpenAI Increases Advertising in ChatGPT and Targets 92% Free Users

OpenAI is looking to fund its projects and claims to be doing so through a focus on product and revenue model testing. According to TIME, ads integrated into ChatGPT have reportedly yielded satisfactory results during initial trials, prompting management to increase their presence. CFO Sarah Friar specifies that this advertising revenue could help cover the expenses generated by the 92% of ChatGPT users who use the service without a subscription. The company is also testing "Sponsored Agents," where clicking on an ad opens an AI experience hosted by a brand.

The Merge and ChatGPT Work Aim to Spread Agentic AI

OpenAI is combining Codex and ChatGPT in an internal project called The Merge. This effort has led to ChatGPT Work, designed to put agentic AI in the hands of a broad audience: beyond providing answers, the tool handles tasks. Product lead Thibault Sottiaux, who is overseeing this initiative, assures that OpenAI is close to delivering features with persistence and always-active execution. He believes the whole will appear new to the public.

Astra, Presented as an Autonomous and Persistent Research Assistant

Announced last year as an automated "intern," Astra is described by Chief Scientist Jakub Pachocki as already achieving this internal goal. Given an experimental idea, the model family would write code in OpenAI's database, launch the trial, and report back; provided with a scientific paper, it would perform work that a human researcher would take about a week to complete. Astra would also enable persistent agents, capable of operating autonomously over time. Sam Altman indicated to clients that he expects such a model to actually invent new things, calling this capability very close to AGI. The hypothesis of self-recursive improvement is mentioned, but opinions vary: some researchers consider this milestone distant, while others claim to already see signs of it.

Definition of AGI and Internal Predictions Remain Debated

OpenAI defines AGI as highly autonomous systems that surpass humans in most economically valuable tasks. However, the term lacks a precise definition and remains interpretative, even as a debate persists about the ability of AI based solely on language models to discover and generalize. Others argue that a robust understanding of the world is necessary and that LLMs are only part of the picture. Internally, Mark Chen estimates the path to AGI at 80%, and Greg Brockman sees this period as one of its emergence. Sam Altman hopes for a system he would classify as AGI by the end of the year and acknowledges that it is not there yet; he also states that he aims for the end of 2026 if one adopts his definition. The claim that a future model could "significantly invent" fuels the debate.

Devices in Preparation: A Voice "Puck" First, Then Humanoids

Sam Altman indicates that OpenAI is working on a small number of devices: one intended to sit on a table, one for the pocket, and one to wear. The first expected is a small puck-shaped device capable of detecting its environment and interacting by voice, with availability aimed for early next year according to TIME. Altman adds that OpenAI will also build humanoid robots.