Our car window was smashed and my backpack stolen. Men rolled up in a car wearing ski masks and jumped out. Us and our four kids a few feet away. Laptop, phone, etc all gone. The city is suffering. This is not normal and we should demand better.

After leading product at OpenAI for two and a half years I’ve made the decision to move on. I’ll be telling the story of modern AI and investing in OpenAI alumni and other remarkable founders. More in the thread below and at: moreentropy.com/about

text-davinci-003 ChatGPT Whisper v2 Busy week.

In March 2021 I contemplated leaving @OpenAI to start a company around the ideas that open source and open research were going to win. I decided it was too early but obviously that's no longer the case. Here was my thinking at the time. Who is building this? I Want to help them.

On the one hand, I feel for @ylecun and Meta/Google researchers. Not seeing your research released in usable products due to big tech bureaucratic bs would be hard to stomach. Yet what they have released has been so fumbled it suggests a profound ignorance about shipping product

This wasn't meant to be controversial. I'm saying the same thing as LeCun: "It's nothing revolutionary, although that's the way it's perceived in the public," the computer scientist said. "It's just that, you know, it's well put together, it's nicely done."

Where do AI optimists hang out?

If you're an eng or researcher who wants to work on the forefront of AI, I want to help. I know everyone at @openai, Spark has invested in @AnthropicAI, @AdeptAILabs and others. I am an angel in @trychroma, @perplexity_ai, @dust4ai, @contextualai, etc. DM me

Empire of AI by @_karenhao is by far the most accurate telling of the era when I was at OpenAI, which was an important few years – from the first commercial step to shortly after the launch of ChatGPT. There is one important piece that is incorrect: the portrayal of @sama He’s presented as some machiavellian and reckless leader and the facts don’t support that. I joined OpenAI when we were about 100 people and purely a research lab. As head of product, I helped transition OpenAI from a research org to one deploying our research as products. During this time a number of large and complex decisions were worked through. There were no easy and obvious solutions to any of these and many of these decisions were seemingly at odds with past decisions. Complex situations often look very different to people and there were dynamics at OpenAI during this time that made everything more challenging – from the org’s structure to philosophical belief structures and much in between. The weirdness of OpenAI at this time appealed to me – the unusual structure felt like it created space for something different and the differing beliefs (while exhausting at times) felt necessary for navigating genuinely novel territory. But that same weirdness created real tensions as we worked through three major challenges. First, the Microsoft partnership: how do we take billions from a tech giant without compromising independence and our mission? Second, productization: how do we go from a research lab to shipping products without abandoning our original purpose? Third, deployment: how do we deploy AI research fast enough to matter while being careful enough to be responsible? In the moment, none of these had obvious answers. The right path forward was uncertain, and reasonable people disagreed – often strongly – about what we should do. Led by Sam, we worked through each of these tensions carefully and deliberately. With the fullness of time and the ability to see how things actually played out, I believe the evidence shows we reached the right decisions on all three. When negotiating the early Microsoft deal the entire term sheet was shared with everyone at the org. We’d add questions and comments and then Sam would host an endless meeting where we’d talk through the questions, discuss the spirit of what we cared about, gather feedback on what missed the mark, etc. Each iteration of the term sheet, month after month, progressed like this. Some opposed the partnership, but their voices were always heard and attempts to address their concerns were made. In hindsight, a deal of this sort was required – there was no other viable path – but Sam ensured that our independence and our mission were preserved while spending time working through everyone’s concerns. The first product roadmap spent considerable time articulating why shipping product supported our mission and how we could do so safely. I spent significant time working through my colleagues’ concerns about productization because getting buy-in across the org on the why was essential to doing it right. With Sam’s full support, we consistently slowed down our product work and made decisions that hurt our business and metrics. We refused to allow entire use-cases we felt we couldn’t handle responsibly. We learned what was required – technically and operationally – to comfortably support select use-cases and prioritized that work. We fired some of our biggest customers because we were concerned about misuse. We didn’t get everything right during this era, but we did an excellent job identifying, sizing, and mitigating risk while building one of the most widely-used products in history. This wasn’t luck, it was the result of the deliberate, sometimes frustrating culture Sam insisted we work through. On deployment, many of us believed that deployment was essential to the safety strategy (not separate and something to fear). Learning to deploy the research responsibly would require practice, and the time to practice was when the stakes were lowest. And so we embraced an iterative deployment strategy. While other labs struggled with misuse and PR crises, we consistently deployed without major incidents and we learned and improved with each model release. We all understood that being able to shape the norms and standards of AI was critical to our mission. Sam argued that writing policy memos could only go so far and we’d be in a much stronger position to define norms aligned with our values if we were consistently the first to deploy responsibly. His argument proved more correct than many of us realized at the time. One question I’ve reflected on a lot is why brilliant, well-intentioned people have such different views of this era and Sam’s leadership. I have respect for many who have framed Sam’s leadership negatively, and count many of them as friends, and so it’s somewhat uncomfortable to share my conclusion. Over the years, when I’ve listened to people share examples of what they saw as problematic behavior, I’ve noticed that it often traces back to one of these dynamics: someone who lost an internal debate and attributed it to bad faith rather than legitimate disagreement; someone who struggled to accept that complex situations made previous plans untenable; someone unfamiliar with how large organizations with multiple stakeholders actually function; or someone who pursued power and lost. I don’t say this to dismiss the substance of these perspectives – the concerns about Microsoft, productization, and deployment were real. But I think these underlying dynamics shaped how people interpreted complex, ambiguous situations. When I joined I was told we’d only ever be 200 people. For reasons I understood, we had to abandon this idea. I didn’t feel lied to or misled. I understood we were navigating novel territory where plans had to evolve. Not everyone experienced it that way, and I understand why. But those different experiences don’t mean Sam was acting in bad faith. With several years of distance, I believe the major decisions from that era have held up remarkably well. That doesn't mean we got everything right or that the concerns weren't legitimate – but it does suggest Sam was navigating these tensions with more wisdom than many give him credit for.

While building ChatGPT at @OpenAI I was repeatedly told by our internal teams that our corporate docs were the "Crown Jewels" of the business. I get it. Nobody trusts AI. That's why Spark is excited to support credal.ai in building the security layer for AI

The Financial Times wrote that "OpenAI ranks as one of the most unusual organizations on the planet". I’ve been here just over a year and I must agree! Here’s a thread.

I believe strongly that: 1) The best products that will emerge from this moment are “full stack”, with teams training their own models, and the models & UI informing one another. 2) This requires researchers who care deeply about what’s best for the product, including data

I’m thrilled to share that Spark Capital is co-leading a $22m investment in @elicitorg Today, Elicit’s AI Research Assistant automates the ability to understand what is known. We are excited for a world where Elicit automates the scientific method More in the thread below

I joined Spark two weeks ago and wired my first investment today. I love entrepreneurship and technology. I love that a small team can ship and change the world. As a former founder, I’m excited to support founders building remarkable products

Working closely with @gdb was a joy. Talented, kind people want to work with him not only because he's one of the most talented technologists alive but because he helps great people deliver what they're capable of -- the opposite of creating a hostile work environment.

In the summer I heard from about ~500 people because of the note below and helped many of them. In the next few weeks, I'd like to connect with more engineers and researchers looking to work on the frontier of AI. If you are interested in @OpenAI @AnthropicAI @AdeptAILabs @perplexity_ai @credal_ai @ContextualAI @dust4ai @GlaiveAI or startups pursuing the AI scaling hypothesis in biology.... DM me!

Profluent has successfully edited human DNA with an AI designed gene editor. Remarkable for many reasons, particularly the future it points to: one where AI designs specifically what’s needed to cure diseases. Coverage from @nytimes below in thread.

nothing but love for @sama and @gdb and all that they’ve built. they’ve changed the trajectory of the world for the better

I just used ChatGPT to interpret ultrasound results for my son (all is well) and it felt utterly remarkable. It is hard to even comprehend how much better our lives are going to be because of AI.

Here are some of my thoughts on the future of AI, in response to the leaked google memo: The case for open ai tl;dr: there will be an oligopoly and a vibrant open ecosystem. moreentropy.com/p/the-case-f…

Appreciate the kind words. We are all OK. Kids range from confused to shaken. I’m troubled that I have become so desensitized to the ills of the city that my immediate response was a resigned sigh at the nuisance this was going to create for us.

But we’re at the earliest stages of a technology wave that will make the world dramatically better, in countless ways

I did not foresee that the champions of this movement would be Facebook. Rather wild! But we can thank @GuillaumeLample, @ylecun, and others

Proud to share that Spark Capital is leading a $35m round in Profluent. We’ve seen what happens when the scaling hypothesis is applied to language and are excited for a future where AI makes biology programmable.

Congratulations to all of my friends at OpenAI! Many chapters will be written in the years ahead about the ingenuity that went into building GPT-4. They won't say it, but I can: this is a truly remarkable accomplishment by a team unlike any other in the world.

We’re now entering the era of AI-first products and the world is changing in beautiful ways. I’ll be exploring this change with something new, telling the story of modern ai, and investing in OpenAI alumni and other remarkable founders. Follow along moreentropy.com/about

This is such a stunningly good post: maithraraghu.com/blog/2023/d… AI systems focused on specific high-value use-cases are already emerging

Yesterday was my last day at @airbnb, which marks the end of the six year Koko story, from startup through sale. On Monday I start at @openai and am excited beyond belief about this next adventure.

“It’s just… well put together,” is doing a lot of work in that sentence 😉

And I’d like to be closer to people who share this view and who are building to this

"A badass sheep wearing a lab coat in a science lab, 1980s Miami vibe, digital art"

Awe inspiring magic.

New funds, same love for founders and those who build.

2022 was especially sweet for me. From leading the DALLE product release to building and leading the team that launched ChatGPT — all while relocating a family of 6 to the Bay Area — it was a fun year (albeit busy).

Maybe it’s because I came into AI accidentally, using it to help scale a product that supported people at risk of self harm and suicide, and later applied it to help Airbnb provide safer experiences for their hosts and guests

What a fabulous year. Thank you to everyone at @sparkcapital and the many friends at @OpenAI who helped make this possible. And thank you to the many founders I’ve gotten to know and get to work with.

The challenge in the proposal is defining a compute threshold. Every time one has been set in recent years based on “danger” we’ve quickly learned the risks were ~zero. GPT-2 was deemed too dangerous to release. Evidence to date suggests a lack of credibility in forecasting risk

Stakeholders gotta stake hold. And that’s the real super alignment effort of our time

In summer 2020 many couldn’t appreciate why I’d join a research lab. After selling our AI startup to Airbnb in 2018 I realized that the AI world had changed because of the transformer and that soon AI would change the world.

Congrats to all my friends at @openai - what a day! Clearly so much grinding to make it happen

We're introducing DALL-E 2 today, AI that generates images from text. It's delightful, fun, and feels completely magical. labs.openai.com/s/q3Gq4APxkp…

In January, we introduced DALL-E, AI that generates images from text, and one of the most delightful pieces of technology I’ve ever used. venturebeat.com/2021/01/05/o…

It’s been enjoyable to help the world’s most ambitious research lab embrace product development. When I wrote the first product roadmap for OpenAI a lot of discussion focused on answering why product was important. The culture has come a long way!

Just over a year ago, we released GPT3. The MIT Technology Review named it one of the 10 Breakthrough Technologies of 2021 (alongside mRNA vaccine, no less) technologyreview.com/2021/02…

Set up a Lindy that gets cc'd, looks at my schedule, and replies with times to meet. Once confirmed, the Lindy delegates creating the calendar event to another Lindy, which informs the original Lindy once done so it can reply to the thread. Magic from @Altimor and team.

Congratulations for all that you helped create, the products, yes, but especially the team you built and led. What a run and excited for all that you'll do next.

If you love building products with exceptionally talented and compassionate teammates, come shape the future of AI. We're hiring on the product team for curious, determined, and kind individuals: boards.greenhouse.io/openai/…

I believed that with their talent and ambition OpenAI was the group most likely to shepherd in this change. It’s been an honour supporting the researchers and engineers who have done exactly that. They are the best at what they do.

Earlier this month @github released Copilot, the first product powered by OpenAI Codex, which is coming to our API later this summer. Instagram’s cofounder calls Copilot “the single most mind-blowing application of ML I’ve ever seen.” copilot.github.com

He is, of course, out of his mind

We also introduced CLIP, a state-of-the-art image classifying AI. An “AI-generated art scene has exploded as hackers have been modifying an OpenAI model to make astonishing image-generation tools.” vice.com/en/article/n7bqj7/a…

Perplexity is amazing with R1.

Today, @elicitorg is launching Elicit Reports, the best way to understand the state of science. AI's ability to automate work is already profound -- PhDs estimate that Elicit saves them 16 hours of research per report.

This is why maintaining and prioritizing against a detailed backlog is a mistake as it creates a false sense of progress and distracts from larger efforts to find pmf

you are a hero of mine. thank you for everything so far and for all that will come next.

This was a remarkable effort from a remarkable team. @joannejang led this from the product side, one of her first projects since joining a few months ago. If you are determined, curious, and kind -- and want to ship product this ambitious -- join us. fraser@openai.com

For Canadians in California the July 1st - July 4th combo is always sweet, especially so when Canada Day falls on a Saturday!

I agree! Would value your thoughts on this: moreentropy.com/p/novelty-vs…

Testing by✨vibes✨ and controlling the product experience moreentropy.com/p/testing-by…

"Directly Competing With Open Source Is a Losing Proposition" purported leaked memo from Google semianalysis.com/p/google-we…

Their focus on the researcher as the user will preclude them from delivering on this, though they're obviously building part of the ecosystem that will enable it. I think the closest today is @bfirsh / @replicatehq and @NaveenGRao / @MosaicML

One year ago today the technique to have GPT follow instructions was released, which went on to enable ChatGPT, and which will enable many startups to create differentiated product experiences moreentropy.com/p/startups-a…

"No gpu before pmf" - @spolu

So much magic has radiated out of the research org, a constant stream of moments that made the world gasp. One. After. Another. Many talented researchers behind them but one constant behind the researchers. What a run, the rest is well earned.

No, white Lexus

Success, as the saying goes, has 1,000 fathers. Undoubtedly ChatGPT was the result of many brilliant people and a lot of good luck. But you were the one who breathed life into it. Congrats on all that you have done and all that you will do during this historical moment.

@ravin_tham and jack are the ideal founders to solve this problem. They worked together at Palantir for many years, deploying AI into enterprises where data security was critical. The future of work tools may be smart but they also have to be safe. Protect the jewels!

The same trajectory that took us from GPT-2 to ChatGPT is playing out in robotics and biology: Era 1: it's a toy, easily dismissible Era 2: starts to become useful, not valuable Era 3: useful, starts to become valuable Era 4: valuable In a decade marginal cost of labor for many tasks will be zero, AI-designed molecules will cure many diseases We just need the data

Started from the bottom

:heart-eyes:

This research by @IreneSolaiman and @cbd was so surprising that in response I wrote a memo arguing that OpenAI should become the best in the world at human data annotation. Their work was foundational for a lot of what's to come with OSS openai.com/research/improvin…

Congrats to you all! Wishing you, @barret_zoph, @Luke_Metz, @lilianweng, @johnschulman2, and everyone else a lot of success

We’ve shipped a tonne of ambitious product beyond these major releases - a fine-tuning product, specialized endpoints, the Instruct Series of models, and much, much more. And we’ve done this all with a small, collegial team.

“Everything around you that you call life was made up by people that were no smarter than you and you can change it, you can influence it, you can build your own things that other people can use. Once you learn that, you'll never be the same again. " - Steve Jobs

Something else mentioned by @amasad, which I didn’t discuss: @Replit maintains a high shipping velocity by _shunning_ process. The common response to a broken process is to add more process, and people to manage the process. But the optimal thing is usually to remove process.

"We weren't at neurips last week" Product motivated researchers are an under appreciated reason for why what's happening is happening within AI

It was a lot of fun to share stories about the months leading up to the chatgpt release with @nabeel hallwaychat.transistor.fm/ep…

Too dangerous to release the weights, even in the horcrux-like fashion you describe here

An organization is in trouble the moment it prioritizes doing things right ahead of doing the right things

Bill Gates on the age of AI: "The development of AI is as fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone. It will change the way people work, learn, travel, get health care, and communicate with each other."

"In a moment when all these large language models are converging to become roughly functionally equivalent, no one is going to win the game on technology alone. AI is moving from a science problem to a product... problem" platformer.news/p/how-google…

Too much talent across the world, too much entrepreneurial energy emanating from researchers, too much wonder to explore... for a future like this not to take shape moreentropy.com/p/the-case-f…

Novelty vs utility, science vs belief, and where we are with AI-first products moreentropy.com/p/novelty-vs…

Yep. And I would like to help! I do believe that the diff between "building this" and "building a lot of this" is itself a meaningful company, alongside yours.

it's disingenuous and so intellectually dishonest that it's grotesque

It’s been amazing to see all that you’ve done. The developer community, and all of OpenAI, is fortunate to have you in the role!

More should appreciate Noah's prescience.

"no moat"

务必要疯狂地怀抱雄心，且还要疯狂地真诚.

Developers are among the most creative artists of our time. I’m excited to see what they build with our API.

Kids range from confused to shaken. Sorry that you also had to experience this. Appreciate the kind words

Welcome @thisismadani and the talented team at Profluent! medium.com/p/773117378ffa

We are excited to be co-leading this round with @footworkvc with participation from earlier investors like @fiftyyears. We should all celebrate the application of AI to advance reasoning and accelerate progress.

We are moving from having to cherry pick what worked to cherry picking what doesn’t

Legend. What a fantastic run. Excited for all that you’ll do next