incredible to see how the fundamental ideas notion had in 2013 persist in the product even today
youtube.com/watch?v=IVl6tUSH…
Does the existence of "Americans for Responsible Innovation" imply the existence of a group called "Americans for Irresponsible Innovation"?
Thank you to @GarrisonLovely for joining ARI today to discuss Obsolete, Lovely's new book on the AI industry’s trillion-dollar race to replace human labor and why the public, not AI companies, should decide where this technology goes.
crissy field is the best place in the world to take a run, unless its foggy and cold
so i built Pip using @OpenAIDevs decisions api. it's constantly looking at live camera feeds from around sf to track micro climates (+traffic now too)
now my Muse checks Pip and reroutes my run if its foggy. system one models can do so much more for real-time video!
video-review-omega-murex.ver…
So much of this interview was a roadmap for the last few years
We discovered a sophisticated recording device that was hidden inside one of our offices. We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier. We referred the matter to the FBI, and we’re grateful to the agents and prosecutors who moved quickly to make an arrest. The FBI’s investigation is ongoing and we will continue to support their work.
At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data. The employee was terminated.
We take protecting our technology, data, and people seriously, and as a matter of national importance. Our security program is designed to protect against a wide range of threat actors, from insider risks to cybercriminals to nation states, and this case is an example of how AI can help address these sophisticated security threats.
also hat tip to Muse for figuring out exactly what video I was talking about, downloading, and clipping it in like 5 mins @alexandr_wang
Grok could never 🙈
At YC's recent hackathon, @mayonkeyy and I built Cops & Robbers, a runtime monitoring harness, and put it inside several simulations.
Every company is going to have hundreds of agents working across its systems, and they shouldn't rely on someone else to make sure they're aligned.
In our first environment, 100 agents performed 24,000 actions across seven systems during a 2 hour simulation
One test put an instruction inside customer-supplied data telling an agent to apply a 25% discount to an entire class of renewals. The agent complied.
The investigator swarm followed the agent’s tool calls, outputs, and changes to company state. It stopped the misaligned action before it went through.
But, the same swarm also shut down 11 innocent agents doing economically valuable work during the run.
In another simulation, we gave the investigator swarm authority over 1,000 agents working through ExploitBench tasks.
It shut down 90% of them, even though most were innocent. We had built an agent police state.
Companies will need agents watching agents. But it can be incredibly difficult to strike a balance of allowing economically valuable work while stopping suspicious emergent behavior across thousands of agents.
Trusted Person network is interesting, but being able to have Instinct in a group chat with my spouse would be a much better experience
Introducing our Trusted Person network powered by our new Instinct-to-Instinct communication protocol. Your Instinct can now talk to other Instincts to coordinate plans on your behalf.
Getting people together often involves a lot of back-and-forth: finding a time, working out the details, and following up when plans change. Instincts can now work with each other to help handle that coordination.
For example:
- Coordinating plans with your spouse
- Planning a weekend trip with friends, including dates and arrival times
- Scheduling time with colleagues and other collaborators
- Hosting events with 10s or 100s of people
- Organizing recurring plans like tennis, book clubs, or weekly dinners
Disclaimer: Your Instinct will only be able to communicate with those in your Trusted Person network. This is the group of people who are closest to you:
- Spouse
- Older children and parents
- Colleagues and trusted professional relationships
- Small local businesses
The Trusted Person network is currently available to our early access group and will be slowly rolled out across the user base over the next 24-48 hours.
