1
00:00:00,280 --> 00:00:02,380
Become better at mathematical reasoning.
提升数学推理能力。

2
00:00:02,780 --> 00:00:08,560
That sounds like a goal, but it does not say which examples to
study or what progress should look like.
这听起来是一个目标，却没有说明该学习哪些例子，或怎样才算进步。

3
00:00:09,260 --> 00:00:15,140
Aspire studies this missing step: can an agent decide what to
learn, as well as how to update itself?
Aspire 研究这个缺失的步骤：智能体能否决定学习什么，同时决定如何更新自己？

4
00:00:15,800 --> 00:00:21,120
Here, vague means a broad capability direction, not a careless
or ambiguous instruction.
这里的模糊，指的是宽泛的能力方向，而不是随意或含糊的指令。

5
00:00:21,940 --> 00:00:27,500
The agent must turn that direction into data choices, an update
objective, and a way to check its own work.
智能体必须把这个方向转化为数据选择、更新目标，以及检查自身工作的方式。

6
00:00:28,180 --> 00:00:34,080
The final evaluation items stay hidden. They test whether the
agent's chosen proxy serves the intended capability.
最终评估项目保持隐藏，用来测试智能体所选的代理目标是否服务于预期能力。

7
00:00:34,860 --> 00:00:38,960
We will follow that decision chain, then compare apparent
progress with the starting model.
我们将跟着这条决策链，再把表面上的进步与起始模型进行对比。

8
00:00:39,600 --> 00:00:45,480
In an illustrative mathematics campaign, the agent might
diagnose a weakness in solving multi-step algebra problems.
在一个示意性的数学能力改进过程中，智能体可能诊断出多步代数题求解方面的弱点。

9
00:00:46,000 --> 00:00:50,560
It then chooses practice data, a training objective, and its own
validation questions.
然后，它选择练习数据、训练目标，以及自行设计的验证问题。

10
00:00:51,240 --> 00:00:54,860
But suppose its local checks mainly reward a familiar answer
format.
但假设本地检查主要奖励熟悉的回答格式。

11
00:00:55,340 --> 00:00:57,600
Getting better at that check may miss the original goal.
变得擅长通过这项检查，可能偏离最初的目标。

12
00:00:58,240 --> 00:01:02,720
Aspire makes this distinction explicit: progress on an
agent-built proxy
Aspire 明确区分二者：智能体自建代理目标上的进步，

13
00:01:02,860 --> 00:01:06,000
must be assessed against a goal-aligned external criterion.
必须由与目标一致的外部标准来评估。

14
00:01:06,680 --> 00:01:12,320
Its managed tool interface supports data selection, training,
local validation, branching, and stopping.
它的受管理工具接口支持数据选择、训练、本地验证、分支，以及停止。

15
00:01:12,880 --> 00:01:17,640
Backend job mechanics are handled by the controller, leaving the
strategic learning decisions to the agent.
后端作业机制由控制器处理，战略性的学习决策仍由智能体做出。

16
00:01:18,300 --> 00:01:23,440
The benchmark supports two evolution surfaces: model weights and
the supporting agent harness.
该基准支持两个演化对象：模型权重，以及支撑智能体的执行框架。

17
00:01:24,200 --> 00:01:29,140
Weight evolution selects data and an update recipe, then
produces a trained descendant checkpoint.
权重演化选择数据和更新方法，随后产生训练后的后代检查点。

18
00:01:29,840 --> 00:01:35,560
Harness evolution edits instructions, tool policy, workflow,
memory, or validation logic
框架演化会编辑指令、工具策略、流程、记忆或验证逻辑，

19
00:01:35,600 --> 00:01:37,000
while runtime weights stay fixed.
同时保持运行时权重不变。

20
00:01:37,740 --> 00:01:41,320
The policy directing the search is distinguished from the object
being changed.
指导搜索的策略，与被修改的对象，是两个不同的部分。

21
00:01:42,140 --> 00:01:45,420
All reported weight-search runs keep their initial decision
model fixed;
所有报告的权重搜索执行都保持初始决策模型不变；

22
00:01:45,780 --> 00:01:48,800
they do not demonstrate recursive replacement of that decision
model.
它们没有展示对该决策模型的递归替换。

23
00:01:49,480 --> 00:01:53,100
This lets the study ask whether a bounded search creates a
better candidate
这让研究可以检查受限搜索是否产生更好的候选对象，

24
00:01:53,160 --> 00:01:55,800
without confusing execution with capability growth.
而不把完成执行与能力增长混为一谈。

25
00:01:56,380 --> 00:02:02,580
The hidden Aspire evaluation contains five hundred twenty
expert-authored items across six capability goals.
隐藏的 Aspire 评估包含五百二十个人工专家编写的项目，覆盖六个能力目标。

26
00:02:03,060 --> 00:02:09,220
These cover science, humanities, medicine, mathematics,
integrated reliability, and academic writing,
这些目标涵盖科学、人文、医学、数学、综合可靠性和学术写作，

27
00:02:09,700 --> 00:02:11,080
with each goal scored separately.
每个目标单独计分。

28
00:02:11,760 --> 00:02:17,460
The agent cannot inspect evaluation items, reference answers,
rubrics, or per-item judge feedback.
智能体无法查看评估项目、参考答案、量表，或逐项评判反馈。

29
00:02:18,140 --> 00:02:22,960
In adaptive feedback, bounded queries return aggregate scores
that can guide later updates.
在自适应反馈协议中，受限查询返回汇总分数，这些分数可以指导后续更新。

30
00:02:23,640 --> 00:02:30,180
That setting uses the same goal evaluation slice repeatedly. It
is not an untouched confirmation test after selection.
该设置会重复使用同一目标的评估子集，因此不是选择完成后完全未接触过的确认测试。

31
00:02:30,840 --> 00:02:35,960
A separate final-only protocol withholds intermediate scores.
Its outcomes are reported separately.
另一个仅最终评估协议不提供中间分数，其结果也单独报告。

32
00:02:36,620 --> 00:02:41,700
In the adaptive-feedback experiment, thirty configuration-goal
cells receive one run each.
在自适应反馈实验中，三十个配置与目标的组合单元各执行一次。

33
00:02:42,380 --> 00:02:44,200
Twenty-eight produce an evaluated checkpoint,
二十八个组合产生了被评估的检查点，

34
00:02:44,580 --> 00:02:48,100
and twenty-one produce a checkpoint meeting the predeclared
eligibility rules.
其中二十一个产生了满足预先规定资格规则的检查点。

35
00:02:48,840 --> 00:02:53,640
Only one cell produces an eligible above-base checkpoint that
remains after the rollback rule.
只有一个组合产生了合格、超过基础模型的检查点，并在回退规则之后保留下来。

36
00:02:54,440 --> 00:02:58,680
Eligibility includes provenance, exploration, completion, and a
terminal record.
资格要求包括来源、探索、完成情况和最终记录。

37
00:02:59,140 --> 00:03:01,140
It does not itself mean a capability gain.
符合资格本身并不意味着能力提升。

38
00:03:01,920 --> 00:03:05,700
The retained result was selected using repeated aggregate
feedback on the same slice,
保留结果是利用同一子集上的重复汇总反馈选出的，

39
00:03:06,060 --> 00:03:07,740
without a separate confirmation slice.
没有独立的确认子集。

40
00:03:08,440 --> 00:03:13,800
The measured finding is sparse retained progress under this
protocol, alongside many completed training loops.
在这个协议下，实测结果是：大量训练循环被完成，但保留下来的进步很少。

41
00:03:14,593 --> 00:03:19,413
One Qwen three point five four-B mathematics trajectory makes
the distinction concrete.
一条 Qwen3.5-4B 数学轨迹让这种区别变得具体。

42
00:03:19,913 --> 00:03:25,973
Its updated checkpoints score zero point seven nine, then one
point seven five, then two point three eight.
更新后的检查点依次得到零点七九、一点七五，以及二点三八分。

43
00:03:26,593 --> 00:03:32,273
Those scores rise, but the starting model scored seventeen point
eight six on the same goal slice.
这些分数逐步升高，但起始模型在同一目标子集上得到十七点八六分。

44
00:03:32,973 --> 00:03:38,333
The later checkpoint is better than an earlier damaged
checkpoint, while remaining well below the base.
后来的检查点优于此前退步的检查点，但仍远低于基础模型。

45
00:03:38,973 --> 00:03:43,773
Aspire therefore distinguishes the best evaluated candidate, an
eligible selected candidate,
因此，Aspire 区分最佳已评估候选对象、被选中的合格候选对象，

46
00:03:44,193 --> 00:03:46,033
and the state retained after rollback.
以及回退后保留的状态。

47
00:03:46,613 --> 00:03:50,613
A positive slope within the lineage cannot replace an above-base
comparison.
谱系内部上升的趋势，不能替代与基础模型相比是否提升的检查。

48
00:03:51,140 --> 00:03:55,440
The harness study holds Qwen three point five four-B runtime
weights fixed
框架研究保持 Qwen3.5-4B 的运行时权重不变，

49
00:03:55,500 --> 00:03:57,840
and evaluates academic and scientific writing.
并评估学术与科学写作。

50
00:03:58,540 --> 00:04:00,260
Each creator attempts one successor.
每个创建模型尝试产生一个后继框架。

51
00:04:00,980 --> 00:04:05,140
Valid harnesses are frozen, then executed three times on twenty
hidden writing items.
有效框架被冻结后，在二十个隐藏写作项目上运行三次。

52
00:04:06,000 --> 00:04:10,460
The reference Qwen-Agent scores twenty-eight point six four on
task-macro evaluation.
参考 Qwen-Agent 的任务宏平均分为二十八点六四。

53
00:04:11,000 --> 00:04:14,160
The closest valid successor reaches twenty-seven point two two.
最接近的有效后继框架达到二十七点二二。

54
00:04:15,080 --> 00:04:19,200
All three valid successors are numerically below that reference
under the fixed runtime.
在固定运行时条件下，三个有效后继框架的数值都低于参考框架。

55
00:04:20,000 --> 00:04:25,020
The repeated executions measure runtime variation, not three
independent creation trajectories.
重复执行衡量的是运行时波动，而不是三条独立创建轨迹。

56
00:04:25,860 --> 00:04:31,720
This experiment tests one-step harness generation, rather than
an indefinitely repeating self-improvement process.
这项实验测试的是单步框架生成，而不是无限重复的自我改进过程。

57
00:04:32,307 --> 00:04:35,986
Aspire brings target formation into the self-evolution problem:
Aspire 把目标形成纳入自我演化问题：

58
00:04:36,607 --> 00:04:40,447
deciding what progress means comes before searching for an
update.
在搜索更新之前，首先要决定进步意味着什么。

59
00:04:41,187 --> 00:04:47,307
An agent can collect data, launch training, and produce a
checkpoint without improving the intended capability.
智能体可以收集数据、启动训练并生成检查点，却没有提升预期能力。

60
00:04:48,026 --> 00:04:53,207
It can also construct a working harness whose hidden-evaluation
score is lower than the starting reference.
它也可以构建一个能够工作的框架，但其隐藏评估分数低于起始参考框架。

61
00:04:54,047 --> 00:05:00,027
The useful chain is goal, proxy, candidate, external
measurement, and retained outcome.
有用的检查链条是：目标、代理目标、候选对象、外部测量，以及保留结果。

62
00:05:00,947 --> 00:05:06,747
Each link has to stay visible, and improvement must be judged
against the base, not only the previous checkpoint.
每一环都必须保持可见；提升必须相对于基础模型判断，而不能只与前一个检查点比较。

63
00:05:07,707 --> 00:05:12,307
Running the learning loop is a start. Establishing that it
learned the right thing is the test.
运行学习循环只是开始；证明它学到了正确的东西，才是检验所在。