WEBVTT

1
00:00:00.420 --> 00:00:04.560
An agent can play a game many times without learning a strategy
that helps on the next board.
智能体可以玩很多次游戏，却未必学到一种能帮助它处理下一块棋盘的策略。

2
00:00:05.320 --> 00:00:09.200
S-three-Gym tests whether interaction experience can become
better future behavior,
S3Gym 测试交互经验能否转化为更好的未来行为，

3
00:00:09.640 --> 00:00:11.600
rather than only measuring a fixed policy.
而不只衡量固定策略。

4
00:00:12.120 --> 00:00:17.320
It separates three capabilities: self-testing gathers
experience, self-judging evaluates it,
它区分三种能力：自我测试收集经验，自我评判评价经验，

5
00:00:17.680 --> 00:00:19.760
and self-improvement changes later decisions.
自我改进改变后续决策。

6
00:00:20.500 --> 00:00:25.880
The benchmark instantiates this process in seven text-based
games with executable environment verifiers.
该基准在七款基于文本的游戏中实现这个过程，并提供可执行的环境验证器。

7
00:00:26.480 --> 00:00:29.800
A model's own score is kept separate from the environment's
verified outcome.
模型给出的分数，与环境验证的结果保持分离。

8
00:00:30.520 --> 00:00:36.260
We will use an original Minesweeper illustration to follow the
gap between a useful observation and a transferable rule.
我们将使用一个原创扫雷示例，展示有用观察与可迁移规则之间的差别。

9
00:00:36.913 --> 00:00:42.933
Suppose a revealed Minesweeper tile reports one adjacent mine,
and that mine is already known and marked.
假设一个已揭开的扫雷格子提示相邻位置有一颗雷，而那颗雷已经确定并标记。

10
00:00:43.573 --> 00:00:48.453
The remaining unknown neighbors can be uncovered safely because
the local constraint has already been satisfied.
由于局部约束已满足，其余相邻的未知格子可以安全揭开。

11
00:00:49.233 --> 00:00:52.573
A reusable lesson is to count known mines against the clue.
可复用的经验是对照提示计算已知地雷。

12
00:00:53.313 --> 00:00:55.993
Remembering only the clicked coordinate loses that reasoning.
仅记住点击坐标，会丢失这个推理过程。

13
00:00:56.753 --> 00:01:02.713
On a new board, the same coordinate may require a different
action. A rule must be applied to the current state.
在新棋盘上，同一坐标可能需要不同操作，因此规则必须结合当前状态来应用。

14
00:01:03.453 --> 00:01:07.793
This is an illustrative constraint example, not a recorded
benchmark run or a reported score.
这是一个示意性的约束示例，不是基准中记录的执行，也没有报告实测分数。

15
00:01:08.673 --> 00:01:13.713
S-three-Gym checks whether agents make that transformation from
past experience to decisions under new conditions.
S3Gym 检查智能体能否完成这种转化：把过去经验变成新条件下的决策。

16
00:01:14.353 --> 00:01:18.193
During exploration, agents interact with more permissive game
configurations
探索阶段，智能体与条件较宽松的游戏配置交互，

17
00:01:18.253 --> 00:01:20.833
and collect successful and unsuccessful trajectories.
收集成功和失败的轨迹。

18
00:01:21.413 --> 00:01:27.113
For Minesweeper, exploration gives two extra lives. Under
evaluation, the first mine hit ends the episode.
以扫雷为例，探索阶段提供两条额外生命，而评估时，第一次踩雷就会结束该局。

19
00:01:27.773 --> 00:01:32.553
Other games change board sizes, failure handling, or opponent
difficulty between the two phases.
其他游戏会在两个阶段之间改变棋盘大小、失败处理方式，或对手难度。

20
00:01:33.473 --> 00:01:39.673
Exploration and evaluation use disjoint random seeds, and
evaluation trajectories do not enter memory or training data.
探索和评估使用不重叠的随机种子，评估轨迹也不会进入记忆或训练数据。

21
00:01:40.533 --> 00:01:44.953
In the main context experiment, thirty exploration episodes are
checked every three episodes,
主要上下文实验中，三十局探索每隔三局接受一次检查，

22
00:01:45.253 --> 00:01:47.733
using three strict evaluation episodes per checkpoint.
每个检查点使用三局严格评估。

23
00:01:48.473 --> 00:01:53.993
The measured question is transfer to stricter conditions, rather
than reproducing a lucky exploration outcome.
测量的问题是经验能否迁移到更严格的条件，而不是复现一次幸运的探索结果。

24
00:01:54.453 --> 00:02:00.813
History in-context learning places previous observations,
actions, and self-scores directly into later context.
历史上下文学习，把先前的观察、动作和自评分数直接放进后续上下文。

25
00:02:01.453 --> 00:02:06.333
It preserves detailed evidence, but that evidence consumes
context and must still be used correctly.
它保留详细证据，但这些证据占用上下文，而且仍需要被正确使用。

26
00:02:07.293 --> 00:02:12.673
Summary Memory compresses trajectories into rules, mistakes, and
future directions that guide the next episode.
摘要记忆把轨迹压缩为规则、错误和后续方向，用来指导下一局。

27
00:02:13.513 --> 00:02:17.353
Parameter Training converts selected or corrected experience
into supervised updates,
参数训练把选定或修正后的经验转化为监督更新，

28
00:02:17.753 --> 00:02:19.333
so the change resides in model weights.
使变化体现在模型权重中。

29
00:02:20.293 --> 00:02:24.873
These are different information bottlenecks: retain the detail,
compress it, or internalize it.
这些方式形成了不同的信息瓶颈：保留细节、压缩细节，或把经验内化。

30
00:02:25.913 --> 00:02:30.413
S-three-Gym compares their outcomes without assuming that more
interaction automatically creates improvement.
S3Gym 比较它们的结果，而不假设更多交互会自动带来改进。

31
00:02:31.213 --> 00:02:36.093
At each step, the model proposes an action and estimates its
immediate reward under the game rules.
每一步中，模型提出一个动作，并按照游戏规则估计该动作的即时奖励。

32
00:02:36.793 --> 00:02:38.953
That estimate is not merely a confidence label.
这个估计不只是置信度标签。

33
00:02:39.393 --> 00:02:43.893
It influences how the experience is selected, summarized, or
turned into training data.
它影响经验如何被选取、摘要，或转成训练数据。

34
00:02:44.753 --> 00:02:50.933
The environment returns visible feedback, but verifier-computed
rewards remain on the benchmark side during exploration.
环境返回可见反馈，但在探索阶段，验证器计算的奖励仍保留在基准侧。

35
00:02:51.933 --> 00:02:55.353
The paper compares self-judgments with those rewards across
ninety-eight runs
论文在九十八次执行中对照自我评判与这些奖励，

36
00:02:55.713 --> 00:02:58.033
and over one hundred sixteen thousand transitions.
共涉及超过十一万六千次状态转移。

37
00:02:58.973 --> 00:03:03.653
Agreement about whether an outcome is positive can look high
when many transitions have zero reward;
当很多状态转移奖励为零时，对结果是否为正的一致率可能显得很高；

38
00:03:04.333 --> 00:03:06.213
reward magnitude requires another check.
奖励大小还需要另外检查。

39
00:03:07.173 --> 00:03:12.313
Accurate local scoring alone does not reliably predict improved
performance at the next evaluation checkpoint.
准确的局部评分本身，并不能可靠预测下一个评估检查点的表现会改善。

40
00:03:12.953 --> 00:03:16.173
The summary comparison uses the same model and game in each pair
摘要比较在每组中使用相同模型和游戏，

41
00:03:16.233 --> 00:03:19.213
and evaluates sustained improvement above the starting baseline.
并评估相对于起始基线的持续改进。

42
00:03:19.733 --> 00:03:25.873
On Tetris, summaries win five of seven paired comparisons; raw
history wins one, with one tie.
在俄罗斯方块中，摘要在七组配对比较中赢了五组；原始历史赢了一组，还有一组持平。

43
00:03:26.573 --> 00:03:30.273
On Minesweeper, raw history wins four pairs and summaries three.
在扫雷中，原始历史赢了四组，摘要赢了三组。

44
00:03:31.113 --> 00:03:37.433
The paper associates helpful summaries with compact reusable
abstractions, such as preserving a stable Tetris surface.
论文认为，有效摘要通常包含精简且可复用的抽象，例如保持俄罗斯方块堆叠表面的稳定。

45
00:03:38.173 --> 00:03:44.033
State-sensitive games instead need local board configurations,
timing, and hazards that compression may discard.
对状态敏感的游戏，则需要局部棋盘配置、时机和危险位置；压缩可能丢弃这些信息。

46
00:03:44.833 --> 00:03:48.153
A good memory therefore preserves the information needed to
execute the rule,
好的记忆需要保留执行规则所需的信息，

47
00:03:48.613 --> 00:03:50.693
not only a sensible-sounding recommendation.
而不仅是一条听起来合理的建议。

48
00:03:51.213 --> 00:03:56.893
The parameter-training study evaluates Qwen-three eight-B at
twenty checkpoints, including the original model.
参数训练研究在二十个检查点上评估 Qwen3-8B，其中包括原始模型。

49
00:03:57.473 --> 00:04:04.033
Strict evaluation uses no History in-context learning or Summary
Memory, so this pathway tests the parameter update.
严格评估不使用历史上下文学习或摘要记忆，因此这一路径测试的是参数更新。

50
00:04:04.593 --> 00:04:07.953
On Trust Evolution, the score rises from zero to a peak of
thirty
信任演化游戏中的分数从零升至最高三十，

51
00:04:08.293 --> 00:04:11.733
and exceeds the initial baseline at eighteen of nineteen updated
checkpoints.
并在十九个更新检查点中的十八个超过初始基线。

52
00:04:12.393 --> 00:04:17.473
On Plants-versus-Zombies, the initial score is twenty-three, but
every updated checkpoint scores six.
在植物大战僵尸中，初始分数为二十三，但每个更新后的检查点都得到六分。

53
00:04:18.113 --> 00:04:22.213
The gains and regressions belong to this model and training
setting; game scores use different scales.
这些提升和退步对应这一模型和训练配置；不同游戏的分数采用不同尺度。

54
00:04:23.052 --> 00:04:26.373
Persistent parameter changes can therefore internalize useful
experience
持久的参数变化可能内化有用经验，

55
00:04:26.753 --> 00:04:29.393
while also overwriting behavior that previously worked.
同时也可能覆盖先前有效的行为。

56
00:04:30.067 --> 00:04:35.167
S-three-Gym turns experience-driven improvement into a process
whose stages can be inspected separately.
S3Gym 把经验驱动的改进转化为一个可以分别检查各阶段的过程。

57
00:04:35.867 --> 00:04:39.707
Testing should gather informative evidence. Judging should
identify its value.
测试应收集有信息量的证据。评判应识别其价值。

58
00:04:40.207 --> 00:04:42.687
Reuse should produce actions that work in a new state.
复用应产生适合新状态的动作。

59
00:04:43.587 --> 00:04:48.847
The Minesweeper illustration shows why a constraint rule and a
past coordinate carry different information.
扫雷示例说明了为什么约束规则与过去的坐标包含不同信息。

60
00:04:49.687 --> 00:04:55.567
The measured results show task-dependent memory benefits and
both positive and negative effects of parameter training.
实测结果显示，记忆的收益依赖于任务，参数训练既可能带来正面作用，也可能带来负面作用。

61
00:04:56.407 --> 00:04:59.727
A convincing improvement loop needs transfer measured by the
environment,
有说服力的改进循环，需要由环境测量迁移效果，

62
00:04:59.807 --> 00:05:02.286
alongside careful judgment and experience selection.
同时仔细评判和选择经验。

63
00:05:02.967 --> 00:05:08.267
The important question is what changes in the next decision
because the agent learned something it can actually use.
关键问题是：智能体学到了真正能用的经验之后，下一次决策发生了什么变化？