#arcprize — Public Fediverse posts on home.social

diva.exchange @[email protected] · 2025-05-23 · 20:08 UTC

There are two realities in AI today: one sees AGI just around the corner, the other urges caution. The #ARCPrize offers a clear test — can an AI truly understand and generalize, or is it just memorizing patterns?
#divaexchange
Learn more: https://www.diva.exchange/en/privacy/will-there-be-a-real-artificial-intelligence-revolution/

#arcprize #divaexchange

diva.exchange @[email protected] · 2025-05-19 · 15:50 UTC

#OpenAI’s O3 model reportedly scores 76–88% on ARC benchmarks. François Chollet calls it a breakthrough - but warns: this still isn’t AGI. The #ARCPrize helps us tell real intelligence from clever shortcuts. #divaexchange

More: https://www.diva.exchange/en/privacy/will-there-be-a-real-artificial-intelligence-revolution/

#openai #arcprize #divaexchange

diva.exchange @[email protected] · 2025-05-12 · 17:40 UTC

Chollet says today’s AI isn’t brilliant - it’s just “big databases.” These systems retrieve information well but fail at novel problems. The #ARCPrize sets real challenges that demand understanding and generalization, not just memorization.
#divaexchange

More info: https://www.diva.exchange/en/privacy/will-there-be-a-real-artificial-intelligence-revolution/

#arcprize #divaexchange

GOMOOT :mastodon: @[email protected] · 2025-03-25 · 17:07 UTC

💡 ARC-AGI-2 mette in crisi i modelli IA più avanzati

https://gomoot.com/arc-agi-2-mette-in-crisi-i-modelli-ia-piu-avanzati/

#agi #arcagi2 #arcprize #benchmark #blog #chatgpt #claude #deepseekr1 #geminiflash #news #openai #picks #sonnet #tech #tecnologia

#agi #arcagi2 #arcprize #benchmark #blog #chatgpt

Miguel Afonso Caetano @[email protected] · 2024-12-24 · 11:25 UTC

"Yesterday OpenAI announced some very impressive results from their not-yet-released o3 model. According to the announcement, o3 has made enormous progress over its predecessors on several “reasoning” benchmarks, in particular, two quite difficult ones: Frontier Math, a benchmark containing hundreds of unpublished math problems that are known to be hard even for human math whizzes, and the Abstraction and Reasoning Corpus (ARC), a collection of concept-induction tasks which I’ve written about here, here, and here.

In this post I’ll discuss the o3 results on ARC. If you’re interested in AI and active on social media, you’ve likely already heard about these results, but I’ll try to add more context and my own thoughts here."

https://aiguide.substack.com/p/did-openai-just-solve-abstract-reasoning

#AI #GenerativeAI #OpenAI #o3 #ArcPrize #AbstractReasoning #LLMs