AI that can improve itself: A deep dive into self-improving AI and the Darwin-Gödel Machine. https://
richardcsuwandi.github.io/blog/2025/dgm/ Excellent blog post by @richardcsuwandi reviewing the Darwin Gödel Machine (DGM) and future implications.
SAFETY
-
Self-Improving AI and the Darwin-Gödel Machine Deep Dive
By
–
-
Test Examples Extractability and Privacy in Saturated AI Models
By
–
– hrunning these experiments in a clean setting with perfectly deduplicated texts tells us a lot about privacy: – once capacity is sufficiently saturated, the **test examples** are slightly more extractable than the training examples — maybe extraction is a bit of a myth?
– the -
Negligence in AI Due Diligence Requirements
By
–
If that's the case it's proof of negligence. They should have done their due diligence and lack of such information is also problematic.
-
Dwarkesh podcast guests debate divergent AI singularity and job market scenarios
By
–
I love how Dwarkesh‘s guests alternatingly construct detailed scenarios of futures where singularities manifest in 2 or 30 years, followed by a world in which openAI does not interfere with the job market of tax attorneys unless Anthropic releases irresponsible improvements
-
Dario Amodei Warns of Imminent Job Bloodbath and Universal Basic Income
By
–
Pour faire suite à ce tweet où je parle du revenu universel de base comme d’un besoin vital pour éviter l’effondrement de notre modèle économique : J’en discute en détail ici, dans ma dernière vidéo : Dario Amodei parle d’un "bain de sang" imminent… et d’un chômage qui
-
Sam Altman Warns World Must Prepare for Massive AI Impact
By
–
Sam Altman déclare que le monde doit se préparer collectivement à l’impact massif de l’IA. Il dis: « Il y aura des périodes effrayantes à venir » « OpenAI publie volontairement des modèles imparfaits dès maintenant pour que le monde puisse les voir, s’adapter, et contribuer à
-
LLM Security Awareness Gap: Red Team Patching Challenge
By
–
"As soon as the companies realise this, red team it and patch the LLMs it should stop being a problem. But it's clear that they're not aware of the issue enough right now.” How do people still believe this?
-
DAIR Receives Emails About AI-Related Delusions
By
–
We also get many emails from people suffering from these delusions at DAIR.
-
Future AI predictions: only one perspective can be right
By
–
And on that day in the future, zero, one, or two of us will *be unambiguously right.* Of those scenarios, "two" seems like the most unlikely number.
-
AGI and Populism: Future Political Risks
By
–
Oh if you think that populism is bad now, just wait to see what happens once the fully blown AGI is here.