pianopatrick

Well I assume there are some smart hacker / security people in the NSA etc. who could be parachuted in to help if it really came to that.

I think these banks will probably be able to hire equivalent people in the private sector though and might not need the government help.

If the large and politically connected banks and brokerages have losses large enough to threaten their survival as institutions then the government will jump in to help secure them and bail them out if needed.

I would stick with a "too big to fail" institution instead of splitting across multiple.

If you want a real "backup" you could ask for physical stock certificates or get physical gold or something else that cannot be taken from you by computer hacking.

Anybody have an estimate of how much that is going to cost in token cost?

I had to call Verizon customer support yesterday for something. Their AI customer support agent sounded like a person and worked just fine.

I think you can eventually set up an AI system that handles a lot of the interactions and conversations people have.

Can you share any figures on how many tokens were used in each of these experiments?

I don't mind the research. I just don't trust any software system to contain these AIs. I would prefer we use hardware and physics to contain them and not depend on software at all.

I don't know about that. There are many people in jail who are smarter than the guards. Yet we do not see people breaking out of jail all the time.

Why is it not possible long term?

We manage to keep some rather smart people in jail long term, even though the jailers are not as smart as some of their prisoners.

AIs can attempt to cross air gaps, but air gaps vary in terms of their effectiveness. I.e. if "air gap" is just some air between two machines that both have wifi chips, then yes the AI may be able to cross that by hacking the wifi system.

If "air gap" is AI running on a server with no wifi chip inside a faraday cage under a mountain powered by a generator then I'm not sure how the AI would escape on its own. Seems like the physics would contain it.

We should maybe apply the same kinds of physical controls to this software security research as we do to gain of function biological research.

I.e. do not do these frontier model training runs on virtual machines in the cloud which depend on software systems to prevent internet access. Do them in a special data center that is physically disconnected from the outside world. Like a data center in a faraday cage under a mountain powered by a generator with no cables coming in or out.

Or just do these training runs for frontier models in a special data center inside a faraday cage under a mountain. Do not rely on vLANs or hypervisors or VMs or containers or any other software tool. Software is soft.

Just set things up so there is physically no connection to the rest of the world.

Which is why we should physically isolate these frontier training runs.

Run them in a data center in a faraday cage under a mountain.

We really need a physically isolated data center for these frontier training runs.

Like a data center inside a mountain, where based on the laws of physics, no electrical signal can reach through the granite. Then also put them inside a faraday cage. And power them with a generator, such that there are no cables coming out of this data center connecting it to the external world.

OpenAI says this is not feasible due to the need for monitoring - well the monitoring server can be a separate physical machine inside this physically separated data center.

Software is soft. Use physics to contain these models.

Seems to me that from an investor perspective, the AI models built into the cars and TVs and other things are likely to be open source. I think long term, due to the high cost of training these models, there is going to be a "Linux of AI". I.e. some model that lots of companies collaborate on and then use for free.

The economic logic is the same as that of operating systems. AI models are very expensive for one company to make, but lots of companies need them. So companies are better off collaborating on that side of things and competing on something else.

Economically, the AI models that require big servers with lots of GPUs in data centers should likely be more interesting to investors than the AI models that can run on local hardware.

It's going to be "all of the above" because AI has a whole lot of use cases.

For certain AI use cases, AI will be like a utility. For other user cases you need to pay a premium for the frontier. For other use cases AI will be local.

Even in the industries you mention (water and electricity) you have all sorts of businesses. Like people selling water in fancy cans for a large markup or batteries vs wall power.

One thing that might help you.

I heard in a youtube video of a conference talk that many of the more modern models have been designed to output correct JSON. Like the model creators have done something with the next token prediction. When you are asking for JSON, they do some special stuff to basically force the model to output correctly escaped JSON. In the past, getting all output as JSON could lead to malformed JSON and parsing problems. With modern models that basically does not happen any more.

If you are doing a single HTML file harness, you have access to javascript. You might consider getting all model output as JSON. Should be pretty easy for you to parse and then use the output with javascript.

Dealing with JSON in posix sh would have been too many loc for my harness. So I used other techniques. But you don't have to do that.