Create accountLogin

Recent searches

No recent searches

Search options

Only available when logged in.
mathstodon.xyz is one of the many independent Mastodon servers you can use to participate in the fediverse.
A Mastodon instance for maths people. We have LaTeX rendering in the web interface!

Administered by:

Server stats:

2.8K
active users

mathstodon.xyz: AboutStatusProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.7

#reasoningmodels

2 posts2 participants0 posts today
Public
Global Threads

🤖 AI
🔴 OpenAI Unveils o3 & o4-mini Reasoning Models

🔸 o3 outperforms all models in math, coding & visual tasks; o4-mini balances price & power.
🔸 First OpenAI models to "think with images" — can analyze blurry PDFs or sketches.
🔸 Both run Python, browse the web, and will be accessible via APIs & ChatGPT.

#OpenAI#AI#o3
Public
Alex Jimenez

Anthropic’s Evaluation of Chain-of-Thought Faithfulness: Investigating Hidden Reasoning, Reward Hacks, and the Limitations of Verbal #AI Transparency in #ReasoningModels

marktechpost.com/2025/04/05/an

MarkTechPost · Anthropic’s Evaluation of Chain-of-Thought Faithfulness: Investigating Hidden Reasoning, Reward Hacks, and the Limitations of Verbal AI Transparency in Reasoning ModelsA key advancement in AI capabilities is the development and use of chain-of-thought (CoT) reasoning, where models explain their steps before reaching an answer. This structured intermediate reasoning is not just a performance tool; it’s also expected to enhance interpretability. If models explain their reasoning in natural language, developers can trace the logic and detect faulty assumptions or unintended behaviors. While the transparency potential of CoT reasoning has been well-recognized, the actual faithfulness of these explanations to the model’s internal logic remains underexplored. As reasoning models become more influential in decision-making processes, it becomes critical to ensure the coherence between
#CoTs#LLMs
Public
WetHat💦

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.

the-decoder.com/reasoning-mode

THE DECODER · Reasoning models like Deepseek-R1 and OpenAI o1 suffer from 'underthinking', study findsChinese researchers have discovered why AI models often struggle with complex reasoning tasks: They tend to drop promising solutions too quickly, leading to wasted computing power and lower accuracy.
#AI#ReasoningModels#DeepSeekR1
SearchLive feeds
About