unSafe.sh - 不安全
我的收藏
今日热榜
公众号文章
导航
Github CVE
Github Tools
编码/解码
文件传输
Twitter Bot
Telegram Bot
Search
Rss
黑夜模式
Ask your LLM for receipts: What I learned teaching Claude C++ crash triage
好的,我需要帮用户总结一篇文章,控制在100字以内。首先,我要仔细阅读文章内容,理解主要观点。 这篇文章讲的是作者用Claude Code分析和分类C++代码中的崩溃问题。一开始效果不好,Claude生成了很多不切实际的原因。经过几天的迭代,最终效果不错。作者贡献了一些.md文件到GitHub项目raptor,特别是crash-analyzer-agent的部分。 文章还提到LLM在处理无法立即验证的任务时表现不佳,比如崩溃分析需要构建叙述。作者通过让LLM提供详细的数据支持( receipts)来解决这个问题,并由另一个LLM验证这些步骤,从而提高准确性。 总结时要抓住关键点:使用Claude Code进行崩溃分析,通过迭代改进效果,提供详细数据支持并由另一个模型验证。控制在100字以内,不需要开头用“文章总结”之类的词。 最后检查一下是否符合要求,确保内容准确且简洁。 </think> 作者通过训练Claude Code分析C++代码中的崩溃问题,在ffmpeg bug tracker上测试后发现初期结果不理想。通过迭代改进,最终开发出一种基于文本数据的崩溃分析方法:利用子代理生成假设并提供详细数据支持(如指针分配、修改记录等),再由另一代理验证这些假设的正确性。该方法通过分步验证提高结果可信度,并已开源至GitHub项目raptor中。...
2025-12-12 11:17:0 | 阅读: 7 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
llm
receipts
claude
llms
rr
Understand Neural Nets better, post 5 of N -- Code Assistant shootout
文章描述了一个使用全连接Leaky ReLU网络进行图像训练的项目,并通过可视化边界研究网络生成的多面体结构。随着网络规模扩大,计算激活模式哈希的过程变得缓慢。为优化性能,作者尝试使用AI工具 Gemini 和 Claude 来改进代码。Claude 的解决方案更高效且保持输出一致,而 Gemini 则因随机数问题导致训练不稳定。最终优化使代码运行更快,支持更大规模的网络训练和可视化研究。...
2025-7-11 13:28:0 | 阅读: 10 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
nn
polytope
epoch
2000000
A non-anthropomorphized view of LLMs
文章探讨了大语言模型(LLMs)的本质及其在安全与对齐问题上的挑战。作者认为LLMs是生成单词序列的函数,而非具备意识或伦理的实体,并批评将AI拟人化的倾向。文章指出,尽管LLMs在NLP等领域表现出色,但将其视为具备人类特质的对象会混淆讨论,并强调应以数学模型和概率分布的角度审视其功能与风险。...
2025-7-6 19:58:0 | 阅读: 16 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
sequences
undesirable
mathbb
llm
llms
Some experiments to help me understand Neural Nets better, post 4 of N
作者计划阅读四篇论文,主题涉及深度神经网络(DNNs)作为样条、ReLU网络的激活模式及其结构特性。...
2025-5-22 11:58:0 | 阅读: 16 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
relu
arxiv
1274831
fulli
Some experiments to help me understand Neural Nets better, post 3 of N
作者进行了神经网络实验,试图诱导过拟合现象。尽管使用了高参数网络(27,000参数),但未出现预期的过拟合迹象。相反,网络倾向于生成几何形状(如环形)。即使在稀疏训练数据(如312个点)下,网络仍表现出对几何形状的偏好而非记忆个体点。...
2025-4-10 07:36:0 | 阅读: 3 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
network
overfitting
experiments
312
shape
Some experiments to help me understand Neural Nets better, post 2 of N
文章探讨了神经网络的数学结构及其训练过程。通过将偏置向量融入权重矩阵,并使用ReLU或泄漏ReLU作为激活函数,作者展示了神经网络如何将输入空间划分为线性区域(多面体)。文章还通过可视化不同网络结构(如深层窄瓶颈或过参数化网络)的训练动态,揭示了神经网络如何逐步调整这些区域以逼近目标函数。...
2025-4-5 14:40:0 | 阅读: 14 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
network
relu
dots
nn
overline
The German debt brake is stupid!
这篇文章批评了德国的债务刹车机制,并认为支持该机制的人在经济领域是愚蠢的。作者列举了十个理由:该机制不合理、干预市场、导致2015年移民危机处理不当、基于难以估算的结构性赤字、缺乏民主解决方案、限制企业融资、支持“饿死政府”理论、忽视基础设施需求、限制国防开支等。作者认为该机制不仅无益于德国和世界,还可能削弱民主决策能力。...
2025-3-2 14:35:0 | 阅读: 6 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
debt
brake
german
germany
politicians
What I want for Christmas for the EU startup ecosystem
Hey all,I have written about the various drags on the European tech industry in the past, and recen...
2024-12-5 18:7:0 | 阅读: 8 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
instruments
employ
grants
payroll
Someone is wrong on the internet (AGI Doom edition)
The last few years have seen a wave of hysteria about LLMs becoming conscious and then suddenly att...
2024-7-11 01:36:0 | 阅读: 19 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
humanity
agi
theory
coin
Some experiments to help me understand Neural Nets better, post 1 of N
While I have been a sceptic of using ML and AI in adversarial (security) scenarios forever, I also...
2024-7-4 15:33:0 | 阅读: 16 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
network
creases
plane
intuition
neuron
The end of my Elastic/optimyze journey ...
Hey all,== tl;dr ==Today is my last day at Elastic. I'll take an extended break and focus on rest,...
2024-1-31 19:43:0 | 阅读: 7 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
optimyze
father
mother
intense
consulting
A list of factors that act(ed) as drag on the European Tech/Startup scene
This post is an adaption of a Twitter thread where I listed the various factors that in my experien...
2023-12-11 21:54:0 | 阅读: 5 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
capital
european
fragmented
cultural
tax
Book Review: "This Is How They Tell Me the World Ends"
tag:blogger.com,1999:blog-14114712.comments2020-12-10T10:32:08.468-08:00ADD / XOR / ROLhalvar.flakeh...
2021-02-24 16:13:00 | 阅读: 82 |
收藏
|
addxorrol.blogspot.com
14114712
468
flakehttp
Book Review: "This Is How They Tell Me the World Ends"
This blog post is a review of the book "This Is How They Tell Me the World Ends" by Nicole Perlroth...
2021-2-24 15:13:0 | 阅读: 3 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
security
falsehoods
chapters
epilogue
software
Book Review: "This Is How They Tell Me the World Ends"
This blog post is a review of the book "This Is How They Tell Me the World Ends" by Nicole Perlroth...
2021-2-24 07:13:0 | 阅读: 12 |
收藏
|
addxorrol.blogspot.com
security
falsehoods
software
epilogue
chapters
The missing OS
tag:blogger.com,1999:blog-14114712.comments2020-09-19T00:51:49.311-07:00ADD / XOR / ROLhalvar.flakeh...
2020-09-17 02:55:00 | 阅读: 68 |
收藏
|
addxorrol.blogspot.com
14114712
19t00
00add
flakehttp
The missing OS
Preface:When I joined Google in 2011, I quoted a quip of a friend of mine:"There are roughly one an...
2020-9-17 01:55:0 | 阅读: 3 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
datacenter
security
machine
cobbled
The missing OS
Preface:When I joined Google in 2011, I quoted a quip of a friend of mine:"There are roughly one an...
2020-9-16 17:55:0 | 阅读: 4 |
收藏
|
addxorrol.blogspot.com
datacenter
security
machine
decades
My Twitter-Discussion-Deescalation Policy
tag:blogger.com,1999:blog-14114712.comments2020-09-19T00:51:49.311-07:00ADD / XOR / ROLhalvar.flakeh...
2020-08-14 18:20:00 | 阅读: 59 |
收藏
|
addxorrol.blogspot.com
rolhalvar
14114712
19t00
311
My Twitter-Discussion-Deescalation Policy
Twitter is great, and Twitter is terrible. While it enables getting in contact and starting loose di...
2020-8-14 17:20:0 | 阅读: 4 |
收藏
|
ADD / XOR / ROL - addxorrol.blogspot.com
discussions
security
convey
importantly
conveying
Previous
1
2
3
4
5
6
7
8
Next