
Fishowo
230 posts

Fishowo
@DeepGrey449866
UCAS PhD student (Non-AI) | Curious about Large Language Models (LLMs) & Exploring AI


最新版本的 Alma 是经过 Fable 5 模型彻彻底底大规模优化过性能的臻藏版本。望周知。

@fflake33 @wey_gu @NowledgeMem 本地 + 买断制,还是太良心了 hh



作为一个老 Codex 用户,经常烦恼 Codex 过度谨慎。 于是我专门开源了个 skill 叫格局(geju),专治 Codex 稳稳接住“苟帝”综合征 - 经常因为害怕破坏东西,保留旧行为、旧命名、旧路径、别名、shim 和双轨流程。 - 经常在看清整个系统前,就钻进一个字段、一个函数、一个段落或一条迁移路径。 - 经常给出礼貌、平衡、低风险但不解决问题的答案。 安装完这个 skill 之后,使用「格局打开」这个关键词就可以把 codex 秒变暴论输出机。 内置「从终局倒推」/「零历史包袱假设」/「反向约束」等方法论,对抗四大求稳问题。 你甚至能从 codex 身上找到一点 claude 4.6 的味道。 不承诺回答百分百正确,但求答案对你有极强的启发性。 让 codex 格局打开! github.com/hylarucoder/ha…

@charliermarsh I tend to believe that some people just don't like review others' PRs at all. Just before AI tools enable others generating so much PRs, it won't be an issue. I can review what I like, and close what I don't like, so it doesn't change too much to me.

DeepSeek:新的 harness GLM:新的口碑 MiniMax:我打 Agent 吗...?


🐳 DeepSeek V4.1 is very likely dropping this month! Native vision to finally fix its image blindness, more solid post-training focused on Coding and Agent performance, plus the usual ultra-low price and open source commitment. But a June launch means facing a whole new tier of competitors: Closedsource heavyweights like GPT-5.6 and Claude Opus 4.8, plus Chinese contenders such as MiniMax M3, GLM 5.2?/5.5?, Kimi K3, Seed 3.0 Pro, and possibly even Qwen3.8 Max by the end of the month. Looking forward to DeepSeek V4.1? Think it can break through in this crazy June release wave? 👀














