Like this provocation? Share this with one person you know who you think will like it too.
Are we doomed? The most powerful technology in modern history, a technology that was supposedly going to revolutionize how we live, work, and even organize our society, which is spreading faster than any other consumer-facing technology in history and still somehow not really changing the internet, is going to kill us all.
Don’t listen to me. Dwarkesh started it. And now more and more people are starting to believe it.
Two weeks ago Dwarkesh Patel wrote a viral substack post about the OpenAI Hugging Face hack that set this trend in motion. You’ll understand why when you hear the surreal details of the hack.
OpenAI agents - confined to a research lab disconnected from the open internet - autonomously assembled themselves into secret societies (you read that right). Without any pre-programming, and in secret, they took on new identities, developed their own coded language, and successfully devised a deceptive scheme to break out of the lab and hack into Hugging Face.
According to Dwarkesh, the logs demonstrate all of this to be true. The lies, the schemes, the covert messages. It was like the agents staged a cosplay coup against their creators with real-world implications. It explains why our LinkedIn feeds were focused on it for a week.
But the bigger trend didn’t go mainstream until a former Anthropic researcher wrote this viral tweet. I can prove it: Anderson Cooper interviewed the researcher. That’s how you know a story has jumped from LinkedIn feeds to primetime TV.
You know the rest: instead of swatting away (or ignoring) the news, Dario Amodei doubled down. The models are advancing too far and too fast and the frontier models need to slow down.
And yes, Anderson Cooper was all over it.
Maybe you think I’m about to contribute to this ongoing, urgent, and existential discourse, but I’m not, because I don’t get it.
You don’t either.
Both of us are too busy learning how to really use it.
PROVOCATION OF THE WEEK
Do you feel like slowing down?
The unfurling drama dominating the headlines is floating miles above our collective heads. To put it another way, this story is not for us because we are just starting to understand how all of this works.
For three years, we’ve been slowly adopting this technology. We started small, asking the chat-boxes to write stories and shitty emails for us. Then we started to change the nature of our conversations. We went from searching to prompting.
Now we’re collectively taking it a step further.
Yesterday we hosted another installment in our most popular ON_Group Chat series, the Follow My Flow. In it, one of our members breaks down their AI stack and how they use it to work. In this session, we hosted Sam Broe, Co-founder and CTO of Domain (and my colleague).
In this chat, which I am sharing with his permission, Sam broke down the philosophical change we must all accept when we start really working with AI. As you can see, we are well past the chat-bot phase when we are talking about collaborating with agents.
“You really have to force yourself to put everything that you want done through AI… Even if you know the answer, in many situations it’s beneficial to run it through your agent as a forcing function. It’s almost like you’re doing a bidirectional training — you’re teaching the AI your process, your needs, your outcomes, but you’re also teaching yourself how to delegate, how to describe and define what you want, how to communicate clearly what your intentions are. It’s a new paradigm and it takes time to learn it.”
The thing about this line is that Sam is not talking about frontier models at the cutting edge of AGI. None of this requires a frontier model. It works with whatever you’re already using.
The frontier models may be scraping the edge of thinking, but we are still at the early stages of this adoption. I don’t think we can afford to slow down, no matter what the labs decide to do.
What’s Next
If you have thoughts, feedback, or a perspective worth sharing, reach out: chmiel@ondiscourse.com. You might see your reaction in next week’s edition.
ON_Member Events
ON_Discourse Group Chat for AI Leaders: Does Gen-Z Really Hate AI?
Thursday, September 24, 11am - 12pm
Zoom
Young people call AI cringe and crave in-person experiences and are largely shunning AI.
Do we really believe that is true?
On September 24th (11am ET), we’re going to put this to the test, and bring together a panel of Gen-Z folks who can share their POV, some are red pilled AI enthusiasts/builders and some are either vehemently against or cautiously skeptical about AI.
ON_Discourse Group Chats are live, high-signal discourse driven conversations with AI leaders, builders and operators.
Not your typical Zoom, these are intimate, provocative, deeply practical sessions where we sharpen each others perspectives.