Creează un cont pentru a te abona la acest blog și a-i urmări articolele pe pagina Lectură.
-
A “slowdown” might actually improve OpenAI and Anthropic products
The frontier models have now reached a point where they can meaningfully contribute to scientific research.
-
A connection between capability-based security and lambda calculus
Motivation The security model commonly used in desktop operating systems nowadays is woefully inadequate: it was originally designed to protect user A from user B, but not user A from program A.
-
METR’s time-horizon of software engineering tasks does not mean what you think it means
Consider: METR: Measuring AI Ability to Complete Long Tasks METR’s “Measuring AI Ability to Complete Long Tasks” paper uses “a combination of RE-Bench, HCAST, and 66 novel shorter tasks” to measure “50%-task-completion time horizon”, which is defined as:…
-
The future of LLMs: cognitive core + cartridges?
Andrej Karpathy introduced a concept of a “cognitive core”: post #1: “a few billion param model that maximally sacrifices encyclopedic knowledge for capability” post #2: “the idea of stripping down LLMs, of making it harder for them to memorize, or active…
-
Experiments with automated ML experiments
How far is AI from being able to improve itself?
-
Can LLMs actually represent meaning? / Getting over the impenetrable wall
A discussion of AI and LLM often hits an impenetrable wall: “But it’s just a statistical model, and thus it cannot really …”.