TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

3rd Place Solution for the Arc Prize 2024 Competition: Omni-Arc Approach

1 点作者 zurfer5 个月前

1 comment

zurfer5 个月前
&quot;Prompting is not enough, test-time fine-tuning is needed<p>Clearly this competition has shown that LLMs need test-time fine-tuning to do new tasks. Few-shot prompting is not enough for the model to learn novel tasks.&quot;<p>pretty interesting. However, only a pretty small model was used Qwen2.5-0.5B-Instruct. Bigger models were not available because of competition runtime constraints.