引导式测验 · 中等 等级
Magic AI 长上下文代码模型测验
探索 Magic 的 LTM 模型、长上下文检索、HashHop 评估、代码库推理以及其序列算法目标的成本问题。
问题 1 的 8
What software-development direction has Magic said it is pursuing?
预览每个问题8 问题
当您参加测验时，就会出现答案和解释。
- What software-development direction has Magic said it is pursuing?
- Developing AI models and agents for software-engineering tasks
- Replacing all software repositories with image datasets
- Building a spreadsheet-only OCR service
- Generating code without considering project context
- What maximum context capacity did Magic announce for LTM-2-mini in August 2024?
- Up to 100 thousand tokens
- 最多 1 亿个代币
- Up to 1 million tokens
- Up to 10 billion tokens
- What does Magic use HashHop to evaluate?
- How well a model predicts a familiar fact from pretraining
- Whether generated source code passes every production test
- Retrieval and chaining over random hash pairs placed in a supplied long context
- How quickly a model compiles a new programming language
- 为什么 Naive Transformer Attention 对于 1 亿个 token 的上下文来说是一个问题？
- 不花钱
- 它只能离线工作
- 它删除输入
- 它随序列长度呈二次方扩展，变得极其昂贵
- 2024 年 Magic 与哪家公司合作建造超级计算机？
- Google 云
- Apple
- Spotify
- Adobe
- What does a larger context window make possible for a coding task?
- 键盘打字速度更快
- Include more cross-file code and documentation in one context, without guaranteeing project-wide correctness
- 更好的屏幕分辨率
- 电费更便宜
- What comparison did Magic report for LTM-2-mini’s sequence-dimension algorithm?
- A 1,000-fold improvement in total cost for every deployed model
- A 1,000-fold increase in the accuracy of all code edits
- About 1,000 times lower per-token algorithmic cost than Llama 3.1 405B attention at 100M context in the reported configuration
- A fixed 1,000-token limit for every LTM model
- In a simple illustrative lookup with A → B, B → C, and C → D, what chain follows from A?
- A → C → B → D
- A → B → D → C
- A → D → C → B
- A → B → C → D