Sam Altman (@sama) profile picture

The mission of OpenAI is to ensure that AGI benefits all of humanity

SF
Joined July 2006
Sam Altman retweeted
A lot has happened in the past couple months, to say the least. Today we are sharing a recap of what it takes to move a frontier model from training to internal use to public release. As model capabilities rise, so do the safety and alignment standards required to move forward to the next stage. To meet these standards, we rely on four layers of protection: 1. Model alignment: train models to reliably act as intended and respect human oversight. 2. Security and sandboxing: restrict what models can access and affect. 3. Blocking safeguards: stop harmful or unauthorized actions before they happen. 4. Monitoring and response: detect concerning behavior and intervene. The post brings together efforts from across the company, with links to recent research, technical work, and incident reports we’ve published. This is the baseline operational discipline we owe people as we continue training and deploying models - necessary work, but not a solution to alignment in the limit. There’s a lot more to do, and will continue keeping you informed as the thinking and work evolves. 🏗️ openai.com/index/managing-fr…
32
44
5
388
124,981
Sam Altman retweeted
Proud of what we built here! Codex can use the tools you already work with, and you decide what it can access. Here’s how we built the sandbox that makes that possible: openai.com/index/inside-the-…
6
3
39
6,493
So fun to do DealBook with my friend and partner of 11 years, @gdb. Thrilled the world got to see the brilliant, fiercely determined leader I have the privilege of working with every day. Greg’s role at OpenAI has evolved from writing code 100 hours a week in the early days to now leading some of the hardest and most consequential parts of our business. One thing that has not changed is his extraordinary commitment to our mission. I can say with confidence that OpenAI would not exist without Greg, and I believe the world would be on a much worse path toward the AI future without him.
349
113
31
3,304
348,934
6.2 Sol and Astra are out. These models can do amazing things and are very token-efficient; we hope people will get great value from them. In particular, 6.2 Sol ultrafast is something I think people will really enjoy.
GPT-6.2 Sol & Astra are *finally* here. We've also expanded ultrafast to Pro 200 plans for GPT-6.2 Sol, it's my go-to model for most day to day things.
944
451
138
9,863
804,584
Although the new models are very smart, we have more work ahead of us on making the models nice to work with, and will prioritize that for upcoming releases. But even if it takes a little extra work, the intelligence and capability are extraordinary.
335
105
89
3,357
393,169
Sam Altman retweeted
Today’s GPT-6 Cyber release is a meaningful step for defenders: stronger vulnerability research and better alignment. We’ve already been using GPT-6 Cyber alongside GPT-6 Astra with Daybreak to harden our own systems, finding weaknesses, validating their impact, and developing fixes. Even after years @OpenAI, model capabilities continue to surprise me. One small example came from testing a prerelease version of GPT-6 Astra earlier this summer. We were examining tools used across our research infrastructure and one of our research infra leads, Michael Dalton, was looking into some low risk areas like a math task grader that used SymPy, a mathematics library, to check answers. In an offline test, the submitted answer used characters that looked like ordinary letters to get past the grader’s checks. It caused the grader to run a shell command that printed “4242”, while still returning the expected result, 1/2. The grader marked the answer correct. A valid math answer had become a way to run code inside the grader - diagram below. This is the kind of vulnerability that can be very hard for a human to find, even with a fuzzer. It takes connecting how input validation and evaluation interact, constructing an input that passes the checks but triggers unintended behavior, and recognizing the failure even when the final answer is correct. Models can reason across those pieces and help uncover classes of vulnerabilities that are difficult to spot when reviewing components in isolation. That’s why it’s important for more defenders to use models with Daybreak to find and fix vulnerabilities in their own systems. Whether it’s a potentially internet-melting bug like PHP2Shell or a subtle but important issue like this grader bug, finding it before an attacker does gives defenders time to act. Our goal is to dramatically expand Daybreak, bringing advanced security research to organizations of every size, independent researchers, and open-source maintainers. We’re also expanding Daybreak Blue alpha access for GPT‑6 Astra to more organizations and individual security researchers - let us know if you’re interested: docs.google.com/forms/d/e/1F…
Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment. The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work. GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time. openai.com/index/introducing…
49
58
8
508
120,069
Sam Altman retweeted
Day 25/ We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today. We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage. And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
7,368
2,701
3,648
39,132
7,579,147
People should always be in control of the future. OpenAI is against the notion that it would be acceptable for people to be relegated to a role of doing things that “AI systems think makes sense to reward in humans”. People should not cede control of the future to superintelligence, and that requirement should ultimately constrain AI progress.
6,177
1,044
2,765
20,972
10,890,172
6.1 Terra is here! It is good at a lot of things, but we are especially excited for this new era of computer use.
GPT-6.1 Terra: delivering fast and capable computer use at a lower price. GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it.
2,800
499
191
10,195
1,776,880
I am very grateful Aleksander accepted our request to come take this on; the safety and security challenges in front of us as we confront superintelligence are immense, and he has an excellent set of skills to work on this.
Personal update: Turns out that what I thought was leaving OpenAI became more of a sabbatical. When I left a few months ago, I felt that the most important thing to focus on was making AI more useful to people. But we’ve since entered a different regime of AI progress. One in which getting the safety and security challenges of increasingly capable AI right has to become a central priority. So I’ll be returning to OpenAI, joining @merettm to help chart a path through these challenges. There’s much work to do, so let’s get going.
949
254
71
7,337
1,814,987
The back of the Navier-Stokes milestone shirt is the last line of the chain-of-thought from the solution:
919
497
173
11,238
1,303,896
Wow the new roadster looks quite amazing, good job @elonmusk
2,689
1,874
1,043
65,933
7,067,726
Very happy to see this from Nick!
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/
282
144
24
3,825
924,254
ChatGPT can now generate a custom UI for you!
GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone. Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.
1,102
559
136
12,276
2,191,047
Have been waiting for this one for a long time, and I would hate to have to go back to the old version of Chat!
365
18
9
1,109
432,984
looking up at the stars with extra awe tonight. "thy sea is so great and my boat is so small."
1,220
671
310
12,700
2,007,199
thank you to the machines, and the structure of reality, for letting us understand a little more.
260
105
65
3,705
474,880
and thank you to the untold number of people who put in the technical work, brick by brick over the generations, to get us to the point where such a wonder is possible.
223
127
34
3,963
373,488