[sanxiyn]

It doesn't mean anything. The letter says:

> Therefore, we call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4.

StableLM is not an AI system more powerful than GPT-4, so the pause does not apply.

[samstave]

To whom, specificlally does the "pause apply"

Because, I can tell you that no state-agent is going to pause, but amplify.

Israel, Iran, China, Russia and any self-respecting NATO country is secretly pushing their AI as fast as fn possible.

You think the US is pausing anything with a 1 trillion dollar defense budget, especially when this AI has surfaced?

The NSA has been projecting these capabilities forever....

Look at the movie "enemy of the state" as a documentary on capabilities as early as 1998... now look at the fractal spiral that we are witness (and victim) of.

[goatlover]

As far as we know, OpenAI has the most advanced LLM in the world right now. Just because governments can fund all sorts of expensive military contracts or spying networks doesn't mean they have the cutting edge AI research. Lot of times it's companies that make the breakthroughs in tech.

[samstave]

Just so I understand how to come to that conclusion ; how did you come to that conclusion?

Do we have any transparent measure?

(My point is; do we think that what we can see now is the pinnacle of what is capable? or is this kindergarten to the PHDs that we cannot see in this field?

[moonchrome]

All these state actor stories imply that state actors are anywhere close to GPT4 levels. Do you have any evidence for that ?

[Zuiii]

My general test to judge whether a capable state actors has an ability is to answer this simple question: "Ignoring resource and morality costs, is it at all technically possible?"

If yes, then those actors almost certainly have this ability developed already and perhaps even deployed. If not, then maybe. This test has held up remarkably well in my experience.

And that's to say nothing about products that already exist: I would be extremely surprised if the US government and China didn't have a GPT4-level AI trained within one week of OpenAI's GPT4 announcement if not before.

[PoignardAzur]

> "Ignoring resource and morality costs, is it at all technically possible?"

If it were that simple, SpaceX wouldn't have revolutionized spaceflight.

Sometimes private actors have talents or organizational structure that gives them an edge in innovation that public actors can't keep up with for a while.

All competitors to OpenAI we've seen are struggling to reach GPT-3.5 level, let alone GPT-4 level, with years of catch-up time. It's not ridiculous to imagine that state actors are struggling as well.

[moonchrome]

You do realize that Google was the undisputed behemoth in AI research for a decade and now they are scrambling to catch up with OpenAI and are still not delivering.

You're saying that governments are both doing this secretly and more efficiently than Google and OpenAI ?

[samstave]

>>Do you have any evidence for tha

I do not, yet I am a SUPER SKEPTIC --> means I am a conspiracy weirdo that doesnt believe a gosh darn thing any government says, but I am also a technologist who is not ignorant to things which have been built in secrecy.

Thus ;; I summize that some crazy shit is going on with AI behind the scenes that we are not privy to -- and if one persons reality of "you cannot believe that they* are doing anything with AI that we dont know about"* ... to paraphrase a few "A nuke is literally about to fall on our heads"

--

We are moments away from realizing that it ALREADY happened....