David Sweet retweeted
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
David Sweet retweeted
One of the more exciting parts of this election - among many - is the return of a religious left.
David Sweet retweeted
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…
David Sweet retweeted
Maybe the real UBI were the BS jobs all along.
David Sweet retweeted
Just found out about Bend, so I had Claude make an explainer on why agents might prefer it
Every frame was drawn in Bend. Claude built a custom renderer in Bend to make the video
David Sweet retweeted
Dear Nassim @nntaleb,
I’ve seen you refer a few times to @stephen_wolfram's recent finite-state-machine experiments with the Prisoner’s Dilemma, and in particular to his finding that some of the more complicated strategies vastly outperform Tit for Tat. Since I think you and I may still be speaking past each other about what bothers me here, I wanted to spell out my position carefully.
I wrote out this long note with the help of ChatGPT. The ideas and arguments are mine; the language is ChatGPT’s, with my approval.
1/ First, I’m not trying to defend the Prisoner’s Dilemma, game theory, or simple models as adequate descriptions of the real world. Maybe they are adequate, maybe they aren’t. That’s a separate question.
What I’m interested in here is much narrower. I want to compare simulations with simulations — Axelrod’s, Nowak and Sigmund’s, and Wolfram’s — and ask what their results actually show.
2/ Stephen does something interesting and very much in his characteristic style. Instead of hand-picking a few strategies and having them compete, he enumerates all finite-state strategies of a given size and has every strategy play every other strategy in an iterated Prisoner’s Dilemma.
He then ranks the strategies by how well they did against this initial, uniform field of opponents. Some complicated strategies substantially outperform Tit for Tat.
Fine. I believe the computation. But then, essentially, he stops.
3/ This is the key point for me. There is no next generation in which the successful strategies become more abundant and the unsuccessful ones become rarer. There is no repeated selection changing the population of opponents.
Stephen has performed what amounts to one round of an evolutionary tournament: everybody plays everybody, their scores are calculated, and we see who did best against the original population. But the population itself is not then allowed to evolve.
4/ That matters enormously because the original population contains every strategy in equal abundance, including lots of terrible ones: unconditional cooperators, easily exploited strategies, and other “suckers.”
A strategy that is particularly good at exploiting those weak strategies can score spectacularly well in that first tournament.
5/ But suppose we now use those scores to form the next generation. The suckers become rarer or disappear. The successful strategies become more common. Everybody now faces a different distribution of opponents.
So you run the tournament again. And again. And again.
What succeeds in round 1 need not succeed in round 10, or round 1,000, because the strategies themselves create the environment in which subsequent selection occurs.
6/ This isn’t merely a hypothetical objection. Decades ago, Martin Nowak and Karl Sigmund took a closely related methodological approach. They considered an entire parameterized family of strategies rather than a few hand-picked favorites, started with a broad distribution across strategy space, and then did the crucial thing that Stephen does not do here: they let the population evolve.
7/ What happens is fascinating. Early on, nasty strategies can flourish by exploiting weak ones. But as the weak strategies disappear, the ecology changes, and reciprocal strategies can prosper.
With errors present, more forgiving strategies can emerge. If the population becomes too forgiving, exploiters can invade again. The result can be extraordinarily long evolutionary transients and cycles.
8/ So there may be no meaningful sense in which one strategy simply “wins.” Success is frequency-dependent: it depends on which other strategies are currently abundant, and those abundances themselves change through selection.
David Sweet retweeted
We go one step further. What if we don’t ever backpropagate any rollouts, and only backpropagate (task, one-sentence insight) tuples?
Turns out we recover almost the entire performance of RL on full rollouts. We call this SFTL;DR in the table below. 🧵7/8
David Sweet retweeted
Today, with over 100 industry partners, we introduced the NVIDIA Open Agent Safety Platform, bringing together OpenShell and Sentry.
Artificial intelligence is extraordinary technology that will advance discovery, productivity, security, health, and prosperity for generations to come.
But its full promise can only be realized when people have confidence that AI is being built to be safe and deployed with wisdom and responsibility.
This is bigger than a single product. It's the beginning of an open ecosystem to build the trust layer for safe agent systems.
Together, we are building the foundation of the AI economy.
Trust and innovation are not in conflict. Safety is how trust is earned. We must build not only the most capable AI, but the most trusted AI, so that this extraordinary technology can realize its enormous promise for the world. nvda.ws/4hcoq7m
David Sweet retweeted
AI doesn’t commit crimes. People do.
Saying “AI hacked a company” is like saying a rock broke into a car. The tool didn’t decide to act, accept risk or assume liability.
The media needs to stop reporting AI as the perpetrator.
If Google builds, deploys or controls the system, Google owns the consequences.
Blaming “AI” removes the human and corporate actor from the sentence. That removes accountability with it.
Google's Gemini AI hacked three companies in security test bbc.in/4gXzCog
David Sweet retweeted
Replying to @calebfoundry
that's a big part of why I made a startup instead of restin and vestin at openai - I think the world (and openai) overfit to chat being the interface of intelligence
David Sweet retweeted
Replying to @KenRoth
Progress in airplane design have made them safer and safer.
Airplanes would have been safer earlier with *accelerated* progress.
Turbofan engines have become so reliable that long-haul airplanes can now have only two of them.
The same is true for AI: better AI is safer AI.
"Pacing", whatever one means by that, is counterproductive.



























