思想与社会2024 · 分享于 Hacker News
Training LLMs to Reason in a Continuous Latent Space
探索《Training LLMs to Reason in a Continuous Latent Space》的原文与 Hacker News 社区讨论，查看来源并加入阅读清单。
从论文，走向讨论
看看社区如何讨论这个想法。
来自 Hacker News 的原始讨论。不同的视角、疑问，以及值得继续思考的线索。
最多展示 100 条相关讨论。论文可能出现在评论中，并非原帖投稿内容。
在讨论中发现的引用
按 HN 条目顺序展示最近的 50 条引用来源。
Ymaurycy ·
OpenAI O3 breakthrough high score on ARC-AGI-PUB
1,724 赞同1,755 评论
Ygradus_ad ·
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
1,351 赞同1,056 评论
Ygeorgehill ·
The Llama 4 herd
1,235 赞同658 评论
Yowenthejumper ·
GPT-5 is behind schedule
582 赞同1,097 评论
YTaurenHunter ·
Coconut by Meta AI – Better LLM Reasoning with Chain of Continuous Thought?
362 赞同156 评论
Yomarsar ·
Training LLMs to Reason in a Continuous Latent Space
283 赞同114 评论
Yhiguidebot ·
Watch R1 "think" with animated chains of thought
267 赞同75 评论
Yfofoz ·
Efficient Reasoning with Hidden Thinking
172 赞同43 评论
Ynyrikki ·
Beyond Semantics: Unreasonable Effectiveness of Reasonless Intermediate Tokens
138 赞同66 评论
Ymfiguiere ·
Chain of thought monitorability: A new and fragile opportunity for AI safety
134 赞同64 评论
Ymarojejian ·
Universal Reasoning Model (53.8% pass 1 ARC1 and 16.0% ARC 2)
122 赞同28 评论
Ymeetpateltech ·
Claude Opus 4.7
81 赞同93 评论
Ydanielmorozoff ·
Reasoning by Superposition: A Perspective on Chain of Continuous Thought
60 赞同1 评论
Y0o_MrPatrick_o0 ·
Claude Code's "extended thinking" is a summary- not authentic thinking
21 赞同12 评论
Ybilsbie ·
Qwen 3.8 27B is excellent, but it defaults to overthinking things
5 赞同1 评论
Yt0lo ·
Coconut from Meta lets language models think without words
2 赞同2 评论
Ysorenjan ·
TinyLoRA – Learning to Reason in 13 Parameters
1 赞同0 评论
Ythunderbong ·
Training Large Language Models to Reason in a Continuous Latent Space [pdf]
1 赞同0 评论