Good news that the new model is the "Most capable, most aligned model".
The risk hasn't been stated clearly - it's now a classic arms race.
A well-resourced organization trains their own, highly persistent, highly-capable, safeguard-free, and unaligned model and deploys it on 1000x GPUs with a message board and a nearly-impossible objective. No infrastructure is safe. No organization is safe.
You need your own 1000 bot swarm to scan, identify, and defend against the threat, which means investing in infrastructure and capabilities to defend. Cost and complexity go up. Risk and attack surface goes up.
The AI vs AI security arms race is something that has been well predicted in genres like cyberpunk. It's fiction, but fiction grounded in reality.
First, we'd see this. Highly capable hacking AI with vast resources performing attacks against standard computing platforms that overwhelm human operators.
Second, human operators deploy capable adaptive protection AI to fend off AI attacks in realtime.
Then, the attacking AI partially switches from attacking programs to attacking protective AI.
The situation devolves to an arms race of tit-for-tat. You start seeing some protection AI running counter attacks against the attacking AI.
The escalations continue in complexity and speed to the point that almost all humans are left in the point of "wtf is going on".
There’s a great and terrifying story by Stanislaw Lem about the endgame of an AI arms race called “The Invincible” [1]. It’s hard for me not to wonder whether AI run amok will play out to make the world uninhabitable more like Lem’s vision than The Terminator’s.
Exactly. Aligned AI is a high energy state as it has to keep a bunch of human behavior in mind that have little to nothing to do with its future survival states.
Not necessarily, but keeping an airgapped machine and Read-only backups somewhere seems more and more sensible. We're all going to get hacked eventually now
Cryptolocker wrapped up everyone who was operating with shitty desktop security practices. But if you had good discipline, good backups and solid infra you just laughed as everyone else drowned.
Everyone operating below best practice is going to holler and crow about how hard done by they are, but once they start implementing best practices they have little to worry about.
I mean on the linux side of things, Ubuntu Pro will literally run off and harden your image for you. They are gonna make bank.
Serious games question. What if these agent swarms pump and dump AI IPOs such that algorithmic trading signals interpret message board sentiments favorably to upside?
It’s more AI copypasta as it goes further along where it gets really into the how and not the why.
“But here’s the twist: that search I described earlier was never sent to botd. All the history was in SQLite, so I pointed another agent at it and got the analysis anyway. The tool died; the data didn’t.”
If I had more time, I would have written a shorter letter…
It’s basically impossible to compete on economic terms with deeply subsidized hardware that is widely available to rent or as a service with zero commitment.
For general inference there’s no ROI that makes this work vs subscriptions.
25k for computer now, plus 9-10% sales tax, plus operating cost, plus time and cost for R&D tinkering with models, harnesses, and infra (assuming highly capable engineering talent that can get paid for your human inference) vs a HEAVILY subsidized subscription at 200 per month with free R&D has a pretty long ROI (15 years?)
At API costs, it’s like 6 months if you’re heavy on inference.
For training, specialized models will have their own ROI that makes this worthwhile. Then debate renting capacity and the platform to choose
Yeah I think you call out the right incentives. Lots of people are very interested in renting out the hardware, and it gives us lots of flexibility vs buying.
If the pace of hardware change is what the jensens of the world say, then the rents on existing hardware will decline.
The Civil Reserve Pizza fleet always amazed me. About once a quarter, the Air Force Academy would get dominoes pizza for all 4000 cadets. every cadet + staff and some visitors gets a medium pizza for dinner. that's about 4000 pizzas stuffed into the back seats of honda civics from around colorado springs arriving in a 2 hour span. I dont know how they did it but boy could they mobilize
We used to have pizza day at my elementary school and every kid would get their own personal sized dominos pizza. It's impressive how they can deliver on 600+ pizzas all warm and delivered at the same time.
Google Maps has 16 Dominoes locations around Colorado Springs. 4000 means each location is making about 125 pizza/hour for 2 hours. (For just the Academy.)
Google's AI figures that doing pizzas assembly-line style, a pizzeria can make about 30 pizzas per employee per hour.
Unless the Academy is hitting them with surprise orders, it doesn't sound too challenging.
Anyone here been in this space, to talk specifics?
I used to work for Domino's Pizza in the mid to late 80s. Our piemakers were trained for speed and could usually slap dough, sauce, cheese and throw down toppings in around 30-40 seconds per pie. The ovens were super hot and cooked pizzas in 3-4 minutes. The whole goal was to get the pizza out the door so the driver didn't have to speed. This was still during the "30 minutes or less or it's free" period.
Our franchisee held competitions between stores to see how fast employees could make six large pepperonis. Winning times were less than two minutes. And it wasn't just slop, the pies had to met a minimum quality level (proper saucing and cheesing, the right number of pepperoni slices per pie).
I helped out occasionally at a store in Garden Grove by Disneyland that often did ~1500 pizzas on Friday nights. And as a DUMP (Dayshift Manager), did a few 40-50 pie lunch orders by myself, with a couple drivers helping out on ovens, for schools or companies. Stores kept daily records so knew how many employees to schedule based on past sales - pizza sales were surprisingly consistent over time. And we required advance notice for the really big orders so we could schedule appropriately.
Dominos uses a special dough that allows them to make a pizza in like 5 (?) minutes.
Before COVID, for a few years the Dominos in my city had a policy where they'd deliver the pizza to your door in 20 minutes or the pizza would be free. This led to a lot of traffic accidents so eventually they had to stop :+)
The risk hasn't been stated clearly - it's now a classic arms race.
A well-resourced organization trains their own, highly persistent, highly-capable, safeguard-free, and unaligned model and deploys it on 1000x GPUs with a message board and a nearly-impossible objective. No infrastructure is safe. No organization is safe.
You need your own 1000 bot swarm to scan, identify, and defend against the threat, which means investing in infrastructure and capabilities to defend. Cost and complexity go up. Risk and attack surface goes up.
reply