@rohanpaul_ai: https://x.com/rohanpaul_ai/status/2108532808054604235

TL;DR: Sam Altman sat down for a Bloomberg TV interview with Ed Ludlow at Fort Mason in San Francisco, covering the positioning of the always-on agent Dots, pricing rationale, the delays behind Astra 6.1 and the training pause, IPO timing, and the responsibility frontier labs have when it comes to risk disclosure and industry coordination.

Interview Context

During OpenAI’s Developer Day, Sam Altman joined Bloomberg TV host Ed Ludlow via remote connection at Fort Mason in San Francisco for a conversation covering the new products announced at the event, business momentum, safety decisions, and the topic of going public.

Dots: An Always-On Agent That Feels “More Like a Conversation”

Ludlow first pointed to a notable detail from the keynote: OpenAI employees interact with Dots not in the traditional “Hey ChatGPT, let me ask a question” style, but rather through an ongoing conversation.

Altman responded that he doesn’t see it as anthropomorphism: “We don’t talk to it exactly the way we talk to a person, but there is a style of interaction that’s different from traditional ChatGPT.”

Target Audience: Starting at the Top of the Market

Asked about who Dots is actually for, Altman said it was initially positioned as a premium product, powered internally by OpenAI’s best model, Astra. Pressed further, he clarified that the product “leans more toward helping enterprise users get work done,” primarily targeting businesses but also covering a large cohort of founders who would traditionally be considered “prosumer” users.

“It’s a very capable model with enormous usage. So we’re starting at the top of the market, and you’ll see us gradually roll out versions for broader consumer audiences.”

Competitive Benchmarking

Ludlow asked how frequently the OpenAI team runs head-to-head comparisons against existing agent products such as Grok Bot, Muse, and Instinct. Altman said he’s tried them but doesn’t use them regularly: “I haven’t found a good use case for them in my own life yet,” though he does make a point of trying every significant new AI product.

He believes Dots belongs to an entirely different category: “The volume of work it can take on, the level of capability, the level of intelligence — I think this is something completely new.”

Pricing: More Expensive, but Benchmarked Against Capability and Compute

Dots is priced higher than other products on the market. Altman attributed this to its capabilities and the compute resources behind it, “so I think it’s useful in a completely new way and on a completely new level.”

Ludlow then raised a broader question: OpenAI simultaneously launched GPT-6.1 Soul, which matches Astra in capability but costs only a fifth as much, while the industry at large is trending toward cheaper inference costs — yet users are more willing than ever to pay. How does OpenAI decide on these pricing tiers?

Altman’s answer echoed OpenAI’s longstanding position: “Our goal is to build increasingly better intelligence and keep driving prices down.” He argued the two are not in conflict — the better AI gets, the more people want to use it. People are clearly more willing than ever to spend on AI, but there’s also so much more they want to do with it.

“If we keep improving quality, lowering prices, and staying ahead at every point on the Pareto frontier, people will use our tools in amazing ways,” he said. He noted that listening to developers at the event describe what they had built and what they could do if prices dropped further was “genuinely remarkable.”

Business Momentum: No Confirmation of ARR Rumors

Asked about the company’s Q3 close, Altman said he didn’t think OpenAI would disclose specific figures, but acknowledged that “it has been a very steep growth period for us.” He cited several strong sources of momentum: the models themselves resonating with users, new products and tools built for developers and enterprises, ChatGPT’s strong growth, and the full slate of new announcements at Developer Day.

Ludlow pressed with a specific figure from reporting — Q3 ARR (annualized recurring revenue) of $70 billion — and asked whether that number was meaningful. Altman responded: “I don’t know where these so-called leaked figures or reported numbers come from. But obviously I’m not going to comment on any of that. What I can say is I’m very pleased with our growth.”

Safety and Pacing: Delaying Astra 6.1, Pausing Training

This was the heaviest portion of the interview. Ludlow pointed out that within the same week, OpenAI launched Dots and 6.1 Soul, but delayed the release of Astra 6.1 and paused training on another model the week prior.

Altman said the discussion around delaying Astra 6.1 had been going on for several months. “We are clearly entering a phase where the capability curve is very steep.” He emphasized that OpenAI’s mission is to build products for people and empower them, and that safety is part of that, as is distributing the benefits.

“When we need to make the decision to slow down training progress or delay a model release so we can make more progress and invest more effort into safety, alignment, monitoring, and safeguards, we will do that without hesitation.”

He also articulated OpenAI’s stance on regulation: “We don’t want to say ‘we need regulation’ or ‘we need competitors to do X, Y, Z, or else we’re not safe.’ We will act responsibly, and we want the world to have confidence in that.”

For what can still be shipped, Altman gave the example of a faster version of Astra or a cheaper version of Astra 6: “There’s absolutely plenty of room to do great things on these new products.” He concluded: “I understand that some people have a ‘damn the torpedoes, full speed ahead’ mindset, but as we push into these new capability tiers, we have to remain responsible.”

Why Prioritize “Faster” Over “More Powerful”

Ludlow followed up: the two things developers and technical users want most are speed and performance — why is implementing new mechanisms for speed on the current generation of models better than releasing a stronger model?

Altman’s answer was direct: “They want both. We’re doing both.” He noted that speed, price, and capability are the three things people ask for most, and “the answer is ‘all of the above’” — it’s just that the desired mix differs depending on the use case.

He specifically called out the ultra-fast mode: “I didn’t expect to like it as much as I do, and I didn’t expect to not want to go back. I should have seen it coming — we all want everything faster — but it’s really great.”

On the economics, Altman acknowledged that the operating costs are significantly higher, “but we also charge more,” so they tried to find a balance. He said the market will render its verdict, but he thinks people will be excited about the pricing and that it makes sense for the company.

IPO: Hopeful to Go Public, But Not Now

On the topic of going public, Altman said he hopes to IPO one day, but denied the notion that “the responsibilities of being a public company are incompatible with focusing on safety”: “Not really.”

His explanation is that OpenAI is in a period of continuously adapting to new capability tiers and confronting new safety requirements, while also needing to become a newly public company and learn how to operate under investor pressure and the additional burdens that come with it. “We just want to get our footing first, make sure we understand how to operate in this new way, and be able to make some of the decisions in front of us without the pressure of being a newly public company.”

“I think we’ll go public at some point. New tiers will keep emerging, but not everything will hit at once.”

Asked whether existing shareholders would support pushing the IPO out to 2028, Altman laughed: “Our investors seem very happy with us and very patient.”

Industry Coordination and Risk Disclosure

Ludlow highlighted a rare point of consensus: Altman, Jerome Powell, and Elon Musk are, unusually, on the same page about the idea of “pacing the frontier,” and OpenAI has been transparent about its internal frameworks and plans.

Altman believes good practice involves two things: first, sharing standards for safety cases — being explicit about what conditions must be met at each tier before advancing, in order to avoid inappropriate risks; and second, sharing methods for ensuring everyone executes those standards well — whether through independent evaluation bodies, government review, or cross-checks between companies. “There are already a lot of proposals out there for this. But I think both of those things are beneficial.”

On risk disclosure, Ludlow noted that Altman has spoken with the United Nations and pointed out that a prospectus once spelled out in black and white an existential threat to humanity — asking whether, as a frontier lab, OpenAI has a responsibility to make these things public.

Altman’s response: “I don’t think that has anything to do with going public. This is just our responsibility to our mission and to the work we’re doing.” Specifically, that means “explaining to the world the risks and benefits we see in this rapidly changing landscape.”

Source: https://www.youtube.com/watch?v=j-AXnqA7KBA