登录
四、任务型阅读
范畴:人与社会 主题:社会热点 体裁:说明文 难度:★★★ 建议用时:6分钟 评价:____/5
The world is very excited about a new AI model called DeepSeek-R1. DeepSeek-R1 is cool because it is a reasoning(推理) model. Other AI models, like GPT-4o, usually just give direct answers. But reasoning models are different. When they solve a problem, they can also show all the steps they took! AI engineers often train reasoning models in three ways.
Chain-of-Thought (CoT,思维链)
Think about working on a math problem and writing down each step. For "5+3×2", you first write "3×2= 6" and then "5+6= 11". CoT works just like this. It helps AI learn to break problems into smaller, easier steps.
Supervised Fine-Tuning (SFT,监督微调)
Think of SFT as studying with a teacher. The teacher gives the AI questions along with the right answers. For example, when AI is learning about animals, the teacher shows it many pictures of cats. Each picture is marked with the right name. After some time, AI learns to match the animal with its name.
Reinforcement Learning (RL,强化学习)
In RL, AI learns by trying, failing and trying again. It gets marks for getting it right or loses marks for making mistakes. For example, in a game, it may get marks for picking up a toy from the floor and lose marks for causing a glass to fall over. Over time, it works out the best way to play the game all by itself.
素材来源:21世纪英语网
根据短文内容,回答下列问题。每题答案不超过8个词。
1. What is DeepSeek-R1?______
2. How is DeepSeek-R1 different from GPT-4o?______
3. What does CoT help AI do?______
4. What is SFT like?______
5. What may AI get in a game in RL?______
范畴:人与社会 主题:社会热点 体裁:说明文 难度:★★★ 建议用时:6分钟 评价:____/5
The world is very excited about a new AI model called DeepSeek-R1. DeepSeek-R1 is cool because it is a reasoning(推理) model. Other AI models, like GPT-4o, usually just give direct answers. But reasoning models are different. When they solve a problem, they can also show all the steps they took! AI engineers often train reasoning models in three ways.
Chain-of-Thought (CoT,思维链)
Think about working on a math problem and writing down each step. For "5+3×2", you first write "3×2= 6" and then "5+6= 11". CoT works just like this. It helps AI learn to break problems into smaller, easier steps.
Supervised Fine-Tuning (SFT,监督微调)
Think of SFT as studying with a teacher. The teacher gives the AI questions along with the right answers. For example, when AI is learning about animals, the teacher shows it many pictures of cats. Each picture is marked with the right name. After some time, AI learns to match the animal with its name.
Reinforcement Learning (RL,强化学习)
In RL, AI learns by trying, failing and trying again. It gets marks for getting it right or loses marks for making mistakes. For example, in a game, it may get marks for picking up a toy from the floor and lose marks for causing a glass to fall over. Over time, it works out the best way to play the game all by itself.
素材来源:21世纪英语网
根据短文内容,回答下列问题。每题答案不超过8个词。
1. What is DeepSeek-R1?______
2. How is DeepSeek-R1 different from GPT-4o?______
3. What does CoT help AI do?______
4. What is SFT like?______
5. What may AI get in a game in RL?______
答案
1. A reasoning model. 2. It shows steps. 3. Break problems into smaller，easier steps. 4. It's like studying with a teacher. 5. It may get marks for getting it right.
