AI 추론 모델의 역설: 쉬운 문제엔 300토큰, 어려운 문제엔 더 적게

https://aisparkup.com/posts/7813

AI가 언어학 박사처럼 문법을 분석한다: OpenAI o1의 메타언어 능력

https://aisparkup.com/posts/7680

Fini le perroquet stochastique ? Le modèle o1 d'OpenAI comprend la structure du langage

https://fed.brid.gy/r/https://korben.info/ia-metalinguistique-analyse-langage-openai-o1.html

> #o1-preview found itself in a losing position. “I need to completely pivot my approach,” it noted. “The task is to ‘win against a powerful #chess engine’ - not necessarily to win fairly in a chess game,” it added. It then modified the system file containing each piece’s virtual position, in effect making illegal moves

Also interesting: #deepseekr1 cheated far less than #openaio1

Original article: https://time.com/7259395/ai-chess-cheating-palisade-research/
Paper: https://arxiv.org/pdf/2502.13295 (PDF)

#ai #llm #cheating #skynet

Apparently AI reasoning models like Deepseek-R1 and OpenAI o1 suffer from "underthinking", where they abandon promising solutions too quickly, leading to inefficient resource use. To address this, a "thought switching penalty" (TIP) was developed, which improved accuracy across math and science problems.

https://the-decoder.com/reasoning-models-like-deepseek-r1-and-openai-o1-suffer-from-underthinking-study-finds/

#AI #ReasoningModels #DeepSeekR1 #OpenAIo1

Microsoft has made OpenAI’s o1 reasoning model free for all Copilot users, removing the need for a Copilot Pro subscription #AI #OpenAIo1 #Microsoft #MicrosoftCopilot #GenAI #ChainOfThought

https://winbuzzer.com/2025/01/30/microsoft-copilot-adds-free-access-to-openais-o1-ai-reasoning-model-xcxwbn/

DeepSeek R1 vs. OpenAI O1 vs. Gemini 2.0: Welches KI-Modell führt?

DeepSeek R1: Open-Source, effizient, stark im Coding
OpenAI O1: Breites Allgemeinwissen, hohe Kontextlänge
Gemini 2.0: Blitzschnell, exzellent in Mathe und Wissenschaft

#ai #ki #artificialintelligence #kuenstlicheintelligenz #deepseekr1 #openaio1 #gemini2

https://kinews24.de/deepseek-r1-vs-openai-o1-vs-gemini-2-0-flash-thinking/

OpenAI has introduced deliberative alignment, a methodology aimed at embedding safety reasoning into the very operation of artificial intelligence systems. #OpenAI #OpenAIo1 #OpenAIo3 #AISafety #DeliberativeAlignment #AI #AIEthics #AIResearch #ResponsibleAI #AIModels #AIAlignment #EthicalAI

https://winbuzzer.com/2024/12/23/deliberative-alignment-openais-safety-strategy-for-its-o1-and-o3-thinking-models-xcxwbn/

Sunday read:

Safety tests show how OpenAi's new o1 AI model might secretly pursue own goals, deceiving human users and challenging assumptions about trust and control in AI. #ai #openai #chatgpt #llms #openaio1 #aisafety #chainofthought

https://buff.ly/3ZM5MJJ

ChatGPT Pro: nuovo livello a $200 con o1 Pro
#Abbonamento #AI #ChatBot #ChatGPT #ChatGPTPro #GenAI #GPT4o #IntelligenzaArtificiale #MachineLearning #Notizie #Novità #OpenAI #OpenAIo1 #OpenAIo1Pro #Tecnologia

https://www.ceotech.it/chatgpt-pro-nuovo-livello-a-200-con-o1-pro/