Toàn bộ 14 bài trước của Tiếng Anh 12; bài chỉ tích hợp kiến thức đã mở. Không thêm điểm ngữ pháp mới, không hướng dẫn đưa dữ liệu bí mật vào công cụ và không coi AI là người chịu trách nhiệm pháp lý/đạo đức thay người dùng.
ENGLISH 6–12 • LEARN IT, USE IT, MAKE IT YOURSTHPT • ĐÍCH BẬC 3 • 105 TIẾT/NĂM
ƯỚC MƠ LỚNENGLISH EVERY DAY
Trang chủCHƯƠNG 5 • BÀI 15 • 85% MỚI + 15% NỀN CŨ
Show the help, keep the thinking
Công khai hỗ trợ, giữ lại tư duy
Chatbot xin lỗi vì bịa một nguồn, rồi nhanh nhẹn đưa thêm một nguồn bịa về ‘nghệ thuật xin lỗi’. Quy trình kiểm nguồn được mời vào phòng.
“The bot apologised for inventing a source so politely that it invented another source about apologies.”Bot xin lỗi vì bịa nguồn lịch sự đến mức nó bịa thêm một nguồn về lời xin lỗi.
Giọng lịch sự và câu trôi chảy không thay source trail; người nộp bài vẫn phải kiểm và chịu trách nhiệm.VOC • CHUNKS BEFORE SINGLE WORDS
Học sáu cụm để dùng ngay
Đọc nghĩa, nghe câu mẫu, che thẻ rồi gọi lại. Không chép một từ rời 10 lần.
publish an AI-use statement
/ˈpʌblɪʃ ən ˌeɪ ˈaɪ juːs ˈsteɪtmənt/công bố bản mô tả việc dùng AIPublish an AI-use statement with the final task.
separate assistance from authorship
/ˈsepəreɪt əˈsɪstəns frəm ˈɔːθəʃɪp/tách hỗ trợ khỏi tác quyềnThe policy separates assistance from authorship.
keep a source trail
/kiːp ə sɔːs treɪl/giữ dấu vết nguồnKeep a source trail for every retained claim.
protect confidential input
/prəˈtekt ˌkɒnfɪˈdenʃl ˈɪnpʊt/bảo vệ dữ liệu đầu vào mậtProtect confidential input by using a fictional file.
accept responsibility for the output
/əkˈsept rɪˌspɒnsəˈbɪləti fər ði ˈaʊtpʊt/nhận trách nhiệm về đầu raThe submitter accepts responsibility for the output.
request human review
/rɪˈkwest ˈhjuːmən rɪˈvjuː/yêu cầu con người xem xétRequest human review when the rule or risk is unclear.
GRAM • MEANING → FORM → USE
Final integration: condition, manner, comparison and result
Bài cuối không mở cấu trúc mới. Chọn linker theo quan hệ: condition đặt ranh giới, manner nói cách thực hiện, comparison đối chiếu cùng tiêu chí, result nêu hệ quả. Một câu dài không tốt hơn nếu quan hệ nghĩa sai hoặc trách nhiệm bị che.
Khung câu
- if/unless/as long as/provided that = condition
- as/as if/the way = manner or appearance
- as…as/than = comparison
- so/such…that = result
Câu dùng thật
If a source cannot be retrieved, do not retain the claim.
Document assistance as the policy requires.
A fluent answer is not as reliable as a verified answer is.
The rule was so vague that two teams recorded support differently.
Bẫy cần gỡ
Không xếp nhiều linker chỉ để câu khó. As if không xác nhận fact; unless không đi với phủ định kép; so that thường chỉ purpose; comparison phải cùng tiêu chí. Nếu câu che người chịu trách nhiệm, tách thành hai câu rõ.
SOUND • CLEAR ENGLISH
Final clarity: thought groups, contrastive stress, links and question tone
Ưu tiên người nghe hiểu rõ. Có hai mẫu Anh–Mỹ/Anh–Anh; giọng cụ thể phụ thuộc thiết bị, còn IPA và hướng dẫn khẩu hình là chuẩn đối chiếu.
Chia policy thành cụm nghĩa; nhấn action/limit/responsible role. Nối âm nhẹ nhưng không chạy nhanh. Dùng rising tone để kiểm yes/no, falling tone cho wh-question/kết luận và contrastive stress để sửa nhầm assistance–authorship.
- AI ASSISTED | I DECIDED
- Can the SOURCE be retrieved? ↗
- WHO accepts responsibility? ↘
- show the HELP | keep the THINKING
AI assisted, I decided; if the source cannot be retrieved, I will not retain the claim.1. Nghe không nói. 2. Nói đè theo cụm. 3. Tự ghi âm và so độ rõ.
LISTEN • 250 WORDS
The rule that failed its own test
Ghi RULE–TEST CASE–AMBIGUITY–RESULT–REVISION–OWNER–ESCALATION. Nghe contrastive stress để biết người nói sửa ‘tool decided’ thành ‘student decided’.
NGHE 3 LƯỢT
- Chỉ bắt ý chính.
- Nghe thông tin cần cho nhiệm vụ.
- Mở transcript để kiểm, rồi shadow một câu.
Chỉ mở transcript sau ít nhất hai lượt nghe
A class tests an AI-use rule before adding it to a learning handbook. The first version says, ‘AI may help with schoolwork if students use it responsibly.’ The sentence is so vague that three teams interpret it differently. Team One asks a tool for two headings, checks both and writes its own paragraph. Team Two asks for a full paragraph and changes three words. Team Three enters a role card containing a fictional phone number, believing that invented personal data is harmless. The review panel applies four questions. What assistance was permitted? Which input was prohibited? Who checked sources? Who accepts responsibility? Team One can answer all four. Team Two cannot separate assistance from authorship. Team Three used fictional data, but the panel notes that students should practise a safe habit: no personal-style identifiers are needed for the task at all. The panel revises the rule as the handbook requires. AI may suggest options provided that the learner keeps the prompt free of personal or confidential data, verifies retained claims, writes the final reasoning and publishes an AI-use statement. If a source cannot be retrieved, the claim must be removed or marked unverified. If policy or risk remains unclear, the learner must request human review. During the hearing, the chair stresses, ‘The tool ASSISTED; the student DECIDED and remains responsible.’ The revised rule is longer than the first, but each condition now has a test, owner and escalation route. The class schedules another trial instead of declaring the policy perfect.
- 1Why does the first rule fail the trial?
- 2What does the chair’s contrastive stress clarify?
TALK • ROLE + GOAL + REPAIR
Hội thoại có mục đích
Mục tiêu: Bảo vệ AI learning rule bằng condition–test–owner–escalation.
Vai: A: hội đồng phản biện • B: người viết chính sách
AWhat assistance does the rule permit?
BA tool may suggest options provided that the learner checks every retained claim.
AHow must the work be documented?
BThe learner must document it as the policy requires in an AI-use statement.
AWhat if the source cannot be found?
BRemove or label the claim, and request human review if the risk remains unclear.
READ • 297 WORDS
A usable AI rule must survive a test case
Lập bảng PERMITTED–PROHIBITED–VERIFY–DISCLOSE–RESPONSIBLE–ESCALATE. Với mỗi linker, hỏi quan hệ nghĩa có đúng không.
A useful AI policy is not a slogan about being responsible. It is a set of conditions that students and reviewers can apply to a real task. The policy should first define permitted assistance. Generating two possible headings is different from generating a final explanation, just as checking grammar is different from deciding which evidence supports a conclusion. The rule should then identify prohibited input. Personal, confidential or restricted material should not be entered merely because a tool can accept it. Verification comes next. If a generated claim is retained, the learner must trace it to an allowed, retrievable source or to the supplied data. A response that sounds as if it has researched a topic is not as reliable as a response whose evidence can be inspected. Documentation protects authorship and review. An AI-use statement should say what the tool did, what the learner retained or rejected, how sources were checked and which reasoning the learner wrote. The statement must not imply that disclosure transfers responsibility. The submitter remains accountable for the final output and for following the task rules. Test cases reveal ambiguity. If two careful readers apply the same policy differently, the wording may be so vague that it needs revision. Each condition should have an observable check and a responsible role. The rule also needs an escalation route. When a source, privacy boundary or permitted use remains unclear, the learner should pause and request human review rather than guess. No policy can predict every tool or task. A review date and correction route allow it to change transparently. The final goal is not to hide all assistance or to hand thinking to a machine. It is to show the help, preserve the source trail and keep human judgement visible from prompt to publication.
1What does a test case reveal?
2Why does the policy need an escalation route?
WRITE • 180–200 WORDS
Viết final AI learning policy 180–200 từ.
Viết cho người đọc thật: dựng ý, viết nháp, kiểm nghĩa rồi mới sửa hình thức.
Plan
- Permitted assistance
- Prohibited input
- Verification rule
- Disclosure and authorship
- Responsibility
- Escalation + review date
Model • 200 words
This policy permits AI to suggest questions, headings or alternative wording for a learning task. It does not permit a learner to submit generated reasoning as authorship. AI may assist provided that the prompt contains no personal, confidential or restricted material. The learner must verify every retained factual claim against a retrievable source or the supplied data. If a source cannot be found, the claim must be removed or clearly marked unverified. The final work must include an AI-use statement written as the class template requires: it names the tool role, important output retained or rejected, checks performed and reasoning written by the learner. A fluent output is not as reliable as verified evidence is. The rule was so vague that teams recorded assistance differently; this version gives each condition an observable test. The learner submitting the work accepts responsibility for accuracy, sources, data choices and compliance. When permitted use, privacy or source quality remains unclear, the learner must pause and request human review. The teacher owns the review route and will reconsider the policy after two fictional trials. This policy does not predict every tool. It keeps assistance visible, protects input and preserves human judgement from prompt to publication.
Self-edit
- 180–200 từ
- Tích hợp đúng 4 quan hệ mệnh đề
- Có permit–prohibit–verify–disclose
- Nêu responsibility–escalation–review
RETRIEVAL PRACTICE
Luyện tập 85%–15%
17 câu bài hiện tại + 3 câu nền đã học. Không có câu dùng kiến thức của bài sau.
Cụm “publish an AI-use statement” có nghĩa phù hợp nhất là gì?
Cụm “separate assistance from authorship” có nghĩa phù hợp nhất là gì?
Cụm “keep a source trail” có nghĩa phù hợp nhất là gì?
Cụm “protect confidential input” có nghĩa phù hợp nhất là gì?
Cụm “accept responsibility for the output” có nghĩa phù hợp nhất là gì?
Cụm “request human review” có nghĩa phù hợp nhất là gì?
___ a source cannot be retrieved, do not retain the claim as verified.
Document the assistance ___ the policy requires.
A fluent answer is not ___ a verified answer is.
The first rule was ___ vague that two teams recorded support differently.
Which words carry the contrast in ‘AI assisted; I decided’?
What tone commonly closes ‘Who accepts responsibility?’
How should a final policy be delivered?
Why does the first rule fail the trial?
What does the chair’s contrastive stress clarify?
What does a test case reveal?
Why does the policy need an escalation route?
It was ___ unclear rule that two teams produced different records.
Record the support ___ the handbook requires.
The trial can continue ___ the privacy boundary is respected.
Gợi mở, không nói hộ
Hỏi con bằng ba câu: ‘Công cụ giúp gì? Con tự kiểm gì? Ai chịu trách nhiệm?’. Gia đình không cần cấm hoặc thần thánh hoá AI; cần giữ dữ liệu riêng tư và đường hỏi người thật.
Dạy để giao tiếp có căn cứ
Đây là bài tích hợp cuối cấp, không mở grammar mới. Dùng fictional test cases; rubric phải chấm language relationship, verification, disclosure, authorship, accountability và escalation, không chấm theo thái độ ‘ủng hộ/chống AI’.
CHƯƠNG 5AI and lifelong learning