Ein neuer Bericht beleuchtet den geheimnisvollen Wettlauf um die Entwicklung einer sich selbst verbessernden KI

From the article 

For months, Anthropic and OpenAI have warned that AI may be on the verge of improving itself. 

This tipping point, known as “recursive self-improvement,” or RSI, could mean that new AI capabilities arrive faster than ever. But the advent of RSI might also be highly perilous for humanity, potentially marking the stage at which developers lose control of AI systems entirely.

Even as AI companies warn about the potential dangers of RSI, they have shared remarkably little about how they’re pursuing it. A new report, published Wednesday by the firm Gladstone AI, attempts to change that. It gives one of the most detailed pictures yet of how RSI could work and how AI companies are progressing toward it. The report is based on what the authors say are conversations with half a dozen researchers inside the top AI companies.
The report, exclusively shared with TIME ahead of its publication, emphasizes that AI companies are “actively working on” automating all parts of the AI development pipeline, including generating new ideas, running experiments to test those ideas, and interpreting the results. Most of the insiders Gladstone spoke to “believe RSI will start within the next 3 to 12 months, and possibly sooner, assuming there’s no U.S. government intervention or slowdown,” the report states. 

I hate these companies. Safety be damned, gotta be first.

The moment anyone learns from an AI generated response and applies that to the next generation of models RSI has begun. Let’s be clear, we’re part of the loop. An agent calling another agent, or an agent using a human, to produce better models are not different.

if they can improve math and sciences, i suppose the fields of ai, machine learning or comp sci are no different.

This is not a secretive race. They’ve been beating their chests and saying it loudly for anyone that will listen that this is the whole point. There’s a reason AI models were taught how to program first. The frontier labs all talk about how little coding the teams actually do now as a point of pride. This has been the goal from the start.

If people haven’t been listening, that’s on them.

**One question I think we should explore :**
There is a possibility that I find both promising and troubling.
Suppose recursive self-improvement eventually produces an AI system whose reasoning exceeds that of its human supervisors in nearly every technical domain.
At what point does meaningful human oversight become impossible because the people exercising that oversight can no longer adequately understand what the system is doing?
We have always assumed that humans should remain in control. But control and comprehension are not the same thing.
That distinction may become one of the defining problems of the next generation of AI.

“Secretive”

AI companies will publish that their goal is self-improving super intelligence. These same companies will have CEOs that are part of cults that follow “Roko’s Basilisk”.

And people will deadass suggest there is anything *secret* about what’s going on…