拟题：
基于混合现实的智能体交互

具体功能：
通过hololens2获取场景数据（识别小车/其他物体的位置，网格信息）。
通过hololens2的立体空间交互能力，指挥小车进行移动，执行任务（譬如抓取柱体并移动到相应位置）。

• 关于移动，可以是在眼镜中通过手部动作画一条路径
• 关于执行任务，可以是眼镜识别出物体的位置后规划小车移动和动作序列(A*or强化学习)->相关信息发给小车->小车收到后执行
  通过hololens2的虚拟物体放置功能，譬如可以放置小车的车库，柱体移动到的目标位置等
  可以通过hololens2显示小车将要移动的意图

The example site: https://chen-yulin.github.io/Besiege-Modern-Docs/

Initialize the local project

node -v >= 18.0

1

npm init docusaurus@latest Besiege-Modern-Mod-docs classic

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

/developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Array/sort#browser_compatibility
✔ Which language do you want to use? › JavaScript
[INFO] Creating new Docusaurus project...
[INFO] Installing dependencies with npm...
...
[SUCCESS] Created Besiege-Modern-Mod-docs.
[INFO] Inside that directory, you can run several commands:

  `npm start`
    Starts the development server.

  `npm run build`
    Bundles your website into static files for production.

  `npm run serve`
    Serves the built website locally.

  `npm run deploy`
    Publishes the website to GitHub pages.

We recommend that you begin by typing:

  `cd Besiege-Modern-Mod-docs`
  `npm start`

Happy building awesome websites!

Configuration

Add a source code button

src/pages/index.js,in function HomepageHeader(), add a <div>.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

function HomepageHeader() {
  const {siteConfig} = useDocusaurusContext();
  return (
    <header className={clsx('hero hero--primary', styles.heroBanner)}>
      <div className="container">
        <Heading as="h1" className="hero__title">
          {siteConfig.title}
        </Heading>
        <p className="hero__subtitle">{siteConfig.tagline}</p>
        <div className={styles.buttons}>
          <Link
            className="button button--secondary button--lg"
            to="/docs/intro">
            Getting Started ⏱️
          </Link>
        </div>
        // +++++++++[[
        <div className={styles.buttons}>
          <Link
            className="button button--secondary button--lg"
            to="https://github.com/Chen-Yulin/Besiege-Modern-Mod">
            Source code
          </Link>
        </div>
        // ]]+++++++++
      </div>
    </header>
  );
}

Two button too close to each other, change the css style for button.
In src/pages/index.module.css

1
2
3
4
5
6

.buttons {
  display: flex;
  align-items: center;
  justify-content: center;
  margin-top: 1rem; // +++
}

Deployment

Add the remote repository on github.

1

git remote add origin git@github.com:Chen-Yulin/Besiege-Modern-Docs.git

Change the deployment configurations

1
2
3
4
5
6
7
8
9
10
11
12

// Set the production url of your site here
url: 'https://chen-yulin.github.io',
// Set the /<baseUrl>/ pathname under which your site is served
// For GitHub pages deployment, it is often '/<projectName>/'
baseUrl: '/Besiege-Modern-Docs/',

// GitHub pages deployment config.
// If you aren't using GitHub pages, you don't need these.
organizationName: 'Chen-Yulin', // Usually your GitHub org/user name.
projectName: 'Besiege-Modern-Docs', // Usually your repo name.
deploymentBranch: 'deploy',
trailingSlash: false,

Change the page setting in repository settings, Branch set to deploy branch.

Deploy command: yarn deploy

1
2
3
4
5

[INFO] `git commit -m "Deploy website - based on aa492c3f0934f0177ca946345c2d32940c1900c3"` code: 0
To github.com:Chen-Yulin/Besiege-Modern-Docs.git
 * [new branch]      deploy -> deploy
[INFO] `git push --force origin deploy` code: 0
Website is live at "https://Chen-Yulin.github.io/Besiege-Modern-Docs/".

Purpose:

We use OpenROAD to generate timing analysis reports for each design. However, manually modifying parameters, running commands, and intercepting important information will be too arduous work. It is mainly due to the following points.

• First of all, the report generated by OpenROAD for each design is too long, and the report file may be as large as hundreds of megabytes. It is almost impossible to open and find tns, wns data using a text editor.
• Secondly, the locations of the configuration files running OpenROAD are very scattered. Moreover, the configuration file directory for timing modification is stored in the tcl script. Manual search and modification are too anti-human and prone to misoperation.
• Furthermore, the amount of data we need is very large. For each design, we hope to test its performance under 7 different clock cycles and 5 different core area sizes. In other words, for each design, it is necessary to manually modify parameters, run commands, and intercept important information 35 times, which is unbearable, not to mention that we have 7 designs.
  Due to these drawbacks, I decided to use shell script to automate the process of OpenROAD testing. I have the following requirements for the script.
• The script can modify the core_area, die_area parameter in the tcl script given the path to the tcl script and the desired scale of area.
• The script can modify the create_clock‘s -period parameter in the sdc configuration file referred by the tcl script, given the path to the tcl script and the desired clock period.
• The script can set all the parameters and run OpenROAD analysis one by one, given the choice set of clock period and area scale.
• After the OpenROAD generated the report file, the script can read the report and extract the information of tns & wns for each stop point and export these data to a csv table.

Main Test Framework

The entry script is named as runit.gxy, which can receive the name of design as an argument.
For each design, the corresponding tcl script is fixed, so I decided to map the name of design to the tcl script.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35

design="$1"
if [ "${design}" = "APU" ]; then
  file_path_orig="../dataset/run/greedy/APU/10n.tcl"
  file_path="../dataset/run/greedy/APU/10n_bak.tcl"
elif [ "${design}" = "ppu" ]; then
  file_path_orig="../dataset/run/greedy/ppu/5n.tcl"
  file_path="../dataset/run/greedy/ppu/5n_bak.tcl"
elif [ "${design}" = "xtea" ]; then
  file_path_orig="../dataset/run/greedy/xtea/5n.tcl"
  file_path="../dataset/run/greedy/xtea/5n_bak.tcl"
elif [ "${design}" = "yhuff" ]; then
  file_path_orig="../dataset/run/greedy/yh/5n.tcl"
  file_path="../dataset/run/greedy/yh/5n_bak.tcl"
elif [ "${design}" = "dynamic_node" ]; then
  file_path_orig="../dataset/run/greedy/dy/5n2.tcl"
  file_path="../dataset/run/greedy/dy/5n2_bak.tcl"
elif [ "${design}" = "ibex" ]; then
  file_path_orig="../dataset/run/ibex_tcl/C0A0.tcl"
  file_path="../dataset/run/ibex_tcl/C0A0_bak.tcl"
elif [ "${design}" = "salsa20" ]; then
  file_path_orig="../dataset/run/greedy/sal/5n.tcl"
  file_path="../dataset/run/greedy/sal/5n_bak.tcl"
else
  echo "no such design"
  exit 1
fi

if [ ! -f "$file_path" ]; then
    echo "File $file_path not found."
    if [ ! -f "$file_path_orig" ]; then
      echo "Origin file ${file_path_orig} not found."
      exit 1
    fi
    cp $file_path_orig $file_path
fi

The reason I use _bak post-fix is that, I want to reserve the original core_area and die_area configuration in the origin tcl file, since the original configuration is needed in the following steps.

Then I use nested loop to change the configuration one by one. The choice set can be modified in the condition statement of each for loop.

1
2
3
4
5
6
7
8
9
10
11

for period in 2 4 6 8 10 12 14
do
  # set clk period
  for scale in 0.2 0.4 0.6 0.8 1
  do
    # set area scale
    #run openroad
    timeout 5 ../../OpenROAD-flow-scripts/tools/OpenROAD/build/src/openroad $file_path > $output_path
    # collect data and export to csv
  done
done

The set clk period, set area scale, export to csv function will be covered in the following steps.

Set the Period of Clock

I create a new shell script called set_clk.sh to manage the automation of changing clock period.
It takes two arguments, the path to the tcl script (string) and the desired clock period (integer).
I first preprocess the period parameter, namely, divide it by 10.

1
2
3
4
5

time="$2"
period=$(echo "scale=1; $time / 10.0" | bc)
if [[ $period =~ ^\. ]]; then
	period="0$period"
fi

The clock configuration can be modified in the sdc file, while the path to the sdc file can be extracted from the tcl file, with the format of set sdc_file "PATH_TO_SDC"
I decided to use grep to locate the line of set sdc_file and extract the path to sdc file using regex substitution.

1
2

tcl_path="$1"
sdc_path=$(grep 'set sdc_file' "$tcl_path" | sed 's/.*"\(.*\)".*/\1/')

In the sdc file, the configuration has a format of create_clock ... -period 0.x.
Still, use regex substitution to replace the original period with the input period and everything is done.

1

sed -i "s/^\(create_clock.*-period \)[0-9]*\.*[0-9]\(.*\)$/\1$period\2/" "$sdc_path"

Set the core_area and die_area

The area of the design is expressed in the form of {x1 y1 x2 y2}, where x1 y1 stands for the coordinate of bottom-left corner and x2 y2 stands for the coordinate of upper-right corner.
Since the area of different design varies a lot, I need to change the area proportionately. As a result I need to reserve the original area stored in tcl script as mentioned before.

First, I read the origianl core_area and die_area from the original script.

1
2
3
4
5
6
7
8
9

file_path="$1"
orig_path="${file_path//_bak}" # remove the _bak postfix_

core_area=$(grep 'set core_area' "$orig_path" | sed -n 's/.*{\([^}]*\)}.*/\1/p')
die_area=$(grep 'set die_area' "$orig_path" | sed -n 's/.*{\([^}]*\)}.*/\1/p')

IFS=' ' read -r core1 core2 core3 core4 <<< "$core_area"

IFS=' ' read -r die1 die2 die3 die4 <<< "$die_area"

Then I calculate the desired core/die area size based on the scale and original area.

1
2
3
4
5
6
7
8
9
10
11
12
13

core_x_size=$(echo "$core3" - "$core1" | bc)
core_y_size=$(echo "$core4" - "$core2" | bc)

core1=$(echo "$core1 + $core_x_size * (1 - $2) / 2" | bc)
core3=$(echo "$core3 - $core_x_size * (1 - $2) / 2" | bc)

core2=$(echo "$core2 + $core_y_size * (1 - $2) / 2" | bc)
core4=$(echo "$core4 - $core_y_size * (1 - $2) / 2" | bc)

die1=$(echo "$core1" - "10" | bc)
die2=$(echo "$core2" - "10" | bc)
die3=$(echo "$core3" + "10" | bc)
die4=$(echo "$core4" + "10" | bc)

Finally, use regex substitution to modify the area configuration in _bak.tcl script.

1
2

sed -i "s/die_area {[^}]*}/die_area {$die1 $die2 $die3 $die4}/g" "$file_path"
sed -i "s/core_area {[^}]*}/core_area {$core1 $core2 $core3 $core4}/g" "$file_path"

Collect data and export to .csv

The key information I need from the report generated by OpenROAD is tns & wns.
By observation, the block containing the data has the following pattern.

1
2
3
4
5

***
result of: stoppoint*
tns: [number1]
wns: [number2]
***

Therefore, I use the grep and awk command to get the last word in each data block. In this case, it’s stoppoint*, [number1], [number2].

1
2

input_file="$1"
tnswns=$(grep -A 2 "result of" "$input_file" | awk '/result of/{p=3} p&&p--' | awk '{print $NF}')

Each report contains totally 6 valid stop points, so I use a loop to get the desired data.

1
2
3
4
5
6
7
8
9

csv_file="${design}.csv"
for nr in 2 5 8 11 14 17
do
	stoppoint=$((stoppoint+1))
	tns=$(echo "$tnswns" | awk -v nr="$nr" 'NR==nr')
	nr_wns=$(echo "$nr + 1" | bc)
	wns=$(echo "$tnswns" | awk -v nr="$nr_wns" 'NR==nr')
	# export to csv table
done

After getting the desired tns and tws, I want to export these data to a csv table.
I decided to let the table has the following columns, period, scale, stop point number, tns and wns, one such table for each design.
This function can be achieved with the following script

1
2
3
4

design="$2"
period="$3"
scale="$4"
echo "${period},${scale},stoppoint${stoppoint},${tns},${wns}" >> $csv_file

Replace the comment in the loop and everything is done.

Result

I assemble runit_gxy.sh in a single script named runit_cyl.sh to run the analysis of all designs at once.

1
2
3

./runit_gxy.sh APU
./runit_gxy.sh ppu
...

Then redirect the output to runit_cyl.log.

Here is the content of APU.csv and runit_cyl.log

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45

Period,Scale,Stoppoint,tns,wns
2,1,stoppoint1,-17.2309532166,-0.4609909952
2,1,stoppoint2,-13.7819681168,-0.4073848128
2,1,stoppoint3,-13.7819681168,-0.4073848128
2,1,stoppoint4,-14.1436796188,-0.3814869523
2,1,stoppoint5,-14.1436796188,-0.3814869523
2,1,stoppoint6,-14.1436796188,-0.3814869523
2,0.95,stoppoint1,-17.2309532166,-0.4609909952
2,0.95,stoppoint2,,
2,0.95,stoppoint3,,
2,0.95,stoppoint4,,
2,0.95,stoppoint5,,
2,0.95,stoppoint6,,
2,0.9,stoppoint1,-17.2309532166,-0.4609909952
2,0.9,stoppoint2,,
2,0.9,stoppoint3,,
2,0.9,stoppoint4,,
2,0.9,stoppoint5,,
2,0.9,stoppoint6,,
4,1,stoppoint1,-8.3930120468,-0.2609909773
4,1,stoppoint2,-4.7969579697,-0.2073848099
4,1,stoppoint3,-4.7969579697,-0.2073848099
4,1,stoppoint4,-4.5129084587,-0.1783644110
4,1,stoppoint5,-4.5129084587,-0.1783644110
4,1,stoppoint6,-4.5129084587,-0.1783644110
4,0.95,stoppoint1,-8.3930120468,-0.2609909773
4,0.95,stoppoint2,,
4,0.95,stoppoint3,,
4,0.95,stoppoint4,,
4,0.95,stoppoint5,,
4,0.95,stoppoint6,,
4,0.9,stoppoint1,-8.3930120468,-0.2609909773
4,0.9,stoppoint2,,
4,0.9,stoppoint3,,
4,0.9,stoppoint4,,
4,0.9,stoppoint5,,
4,0.9,stoppoint6,,
6,1,stoppoint1,-1.8987674713,-0.0609909929
6,1,stoppoint2,-0.0572821237,-0.0073847598
6,1,stoppoint3,-0.0572821237,-0.0073847598
6,1,stoppoint4,0.0000000000,0.0000000000
6,1,stoppoint5,0.0000000000,0.0000000000
6,1,stoppoint6,0.0000000000,0.0000000000
...

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

use design: APU
use tcl file path ../dataset/run/greedy/APU/10n_bak.tcl
get sdc path: ../dataset/run/greedy/gcd/1/gcd_nangate45.sdc
Change period to 0.2
create_clock [get_ports clk] -name core_clock -period 0.2
core area difference 60.18 59.8
set die_area {.07 1.2 80.25 81}
set core_area {10.07 11.2 70.25 71}
collecting data
core area difference 60.18 59.8
set die_area {1.07 2.2 79.25 80}
set core_area {11.07 12.2 69.25 70}
collecting data
core area difference 60.18 59.8
set die_area {3.07 3.2 77.25 79}
set core_area {13.07 13.2 67.25 69}
collecting data
get sdc path: ../dataset/run/greedy/gcd/1/gcd_nangate45.sdc
Change period to 0.4
create_clock [get_ports clk] -name core_clock -period 0.4
core area difference 60.18 59.8
set die_area {.07 1.2 80.25 81}
set core_area {10.07 11.2 70.25 71}
collecting data
core area difference 60.18 59.8
set die_area {1.07 2.2 79.25 80}
set core_area {11.07 12.2 69.25 70}
collecting data
core area difference 60.18 59.8
set die_area {3.07 3.2 77.25 79}
set core_area {13.07 13.2 67.25 69}
collecting data
get sdc path: ../dataset/run/greedy/gcd/1/gcd_nangate45.sdc
Change period to 0.6
create_clock [get_ports clk] -name core_clock -period 0.6
core area difference 60.18 59.8
set die_area {.07 1.2 80.25 81}
set core_area {10.07 11.2 70.25 71}
collecting data
...

将命运的苦难表现得非常纯粹
就是指这种苦难它甚至和客观条件都不那么相关，也就是说，这种苦难是哲学的，普遍的，而不是特别的，临时的，它难以被真正改变。
描写旧时代的小说中最常见的底层人民的苦难，穷，病，愚昧，被剥削，这些在《边城》中都很少，几乎没有。
《边城》中，百姓淳朴善良，邻里关系好。能管事的也慷慨正值。老船夫一家虽然不富裕，但温饱不愁。甚至关于老人的病和城中的水灾，作者也都写得很克制。
但表现上越是美好，最后的结局也就越是让人意难平。
因为即使没有“穷，病，愚昧，被剥削”，人还是得不到想要的东西。
用作者的话来说：一切充满了善，然而到处都是不凑巧。既然是不凑巧，因之朴素的善终难免产生悲剧。
这便是更残酷的生命真相。

Basic

Useful tools

• Dnspy: decompile the runtime library of ONI
• Visual Studio (though it’s too fat)
• Poedit: read the .po file for zh-en translation

Hello world

Apply patches on the Initialize() method in Db class.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

public class Patches
    {
        [HarmonyPatch(typeof(Db))]
        [HarmonyPatch("Initialize")]
        public class Db_Initialize_Patch
        {
            public static void Prefix()
            {
                Debug.Log("Hello! Welcome to CuppyUniverse! [before Db.Initialize]");
            }

            public static void Postfix()
            {
                Debug.Log("CuppyUniverse initialized! [after Db.Initialize]");
            }
        }
        [HarmonyPatch(typeof(ElectrolyzerConfig))]
        [HarmonyPatch("CreateBuildingDef")]
        public class ElectrolyzerConfig_Patch
        {
            public static void Postfix(ref BuildingDef __result)
            {
                __result.EnergyConsumptionWhenActive = 114f;
            }
        }
    }

Add new building

Take gold washer (just like water purifier but output small amount of gold) as an example.

1. Find the source code for water purifier.
   i.e.

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

   using System; using TUNING; using UnityEngine; // Token: 0x0200039A RID: 922 public class WaterPurifierConfig : IBuildingConfig { // Token: 0x06001314 RID: 4884 RVA: 0x000665C4 File Offset: 0x000647C4 public override BuildingDef CreateBuildingDef() { string id = "WaterPurifier"; int width = 4; int height = 3; string anim = "waterpurifier_kanim"; int hitpoints = 100; float construction_time = 30f; float[] tier = BUILDINGS.CONSTRUCTION_MASS_KG.TIER3; string[] all_METALS = MATERIALS.ALL_METALS; float melting_point = 800f; BuildLocationRule build_location_rule = BuildLocationRule.OnFloor; EffectorValues tier2 = NOISE_POLLUTION.NOISY.TIER3; BuildingDef buildingDef = BuildingTemplates.CreateBuildingDef(id, width, height, anim, hitpoints, construction_time, tier, all_METALS, melting_point, build_location_rule, BUILDINGS.DECOR.PENALTY.TIER2, tier2, 0.2f); ... return buildingDef; } // Token: 0x06001315 RID: 4885 RVA: 0x000666A8 File Offset: 0x000648A8 public override void ConfigureBuildingTemplate(GameObject go, Tag prefab_tag) { ... } // Token: 0x06001316 RID: 4886 RVA: 0x0006688B File Offset: 0x00064A8B public override void DoPostConfigureComplete(GameObject go) { ... } // Token: 0x04000A88 RID: 2696 public const string ID = "WaterPurifier"; // Token: 0x04000A89 RID: 2697 private const float FILTER_INPUT_RATE = 1f; // Token: 0x04000A8A RID: 2698 private const float DIRTY_WATER_INPUT_RATE = 5f; // Token: 0x04000A8B RID: 2699 private const float FILTER_CAPACITY = 1200f; // Token: 0x04000A8C RID: 2700 private const float USED_FILTER_OUTPUT_RATE = 0.2f; // Token: 0x04000A8D RID: 2701 private const float CLEAN_WATER_OUTPUT_RATE = 5f; // Token: 0x04000A8E RID: 2702 private const float TARGET_OUTPUT_TEMPERATURE = 313.15f; }

Create a new csharp class file in VS and copy the content into it.
Remember to rename

• the class name
• id in method CreateBuildingDef()
• ID in class
  In this example change them to GoldWasher

2. Add GoldWasher to build menu.
   create a new patch for GeneratedBuildings

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

   using System; using HarmonyLib; namespace ONI_CuppyUniverse_Mod { [HarmonyPatch(typeof(GeneratedBuildings), "LoadGeneratedBuildings")] public class GeneratedBuildings_LoadGeneratedBuildings { public static void Prefix() { Strings.Add(new string[] { "STRINGS.BUILDINGS.PREFABS.GOLDWASHER.NAME", "淘金器" }); Strings.Add(new string[] { "STRINGS.BUILDINGS.PREFABS.GOLDWASHER.EFFECT", "从纯净水中淘取黄金" }); Strings.Add(new string[] { "STRINGS.BUILDINGS.PREFABS.GOLDWASHER.DESC", "你是信无中生有的" }); ModUtil.AddBuildingToPlanScreen("Food", "GoldWasher"); } } }

Reference

• Some mod src
• ONI Mod Guidance
• Harmony Guidance

很小的时候，我爸我妈就跟我说，之所以起这个名字，是因为当时有一档节目叫语林趣话，当时小两口都喜欢看，感觉当时他们也颇为大胆，毕竟直接把节目的名字拿了过来，如果中意的是《奇葩说》，我岂不是成了陈奇葩了（小声）。
再就是参考了活跃于20世纪初的林语堂（说实话这三个字，比起语林趣味，更像是一档节目），是个了不起的人物啊，发明过汉字打字机，也有人写他的传记，还写过苏东坡传，哈哈，还听说过他关于音译“幽默”二字的小故事。近日这本《吾国与吾民》便是出自先生之笔。

文化的稳定性

林语堂认为的，中华民族稳定的重要因素：

• 家族制度
• 阶层流动
• 以及它们所导致的田园理想

中国人的性格

Mellowness
形成原因：

• 中国社会轻视，甚至鄙视“年轻人的热情”。很早便重视责任和理性选择。

patience
indifference
old roguery

中国人思维方式

中西方对于“学者”的态度不同
中国人更敬重日常生活中的智者，用体验而不是理论来学习（孔子及其学生）。

综合，具体，感性，依赖直觉的思维方式（啊，Ni人感到了慰藉）
过去的中国人缺乏一些抽象“分析能力”
依赖经验来解决问题。

• 会心于忘言之境
• 知其然而不知其所以然
• 道即真理，不知其然谓之道

理想人物

出自林语堂《生活的艺术》

是一个对人生有一种建于明慧悟性上的达观者。
这种达观产生宽宏的怀抱， 能使人带着温和的讥评心里度过一生，丢开功名利禄，乐天知命地过生活。
这种达观也产生了自由意识，放荡不羁的爱好，傲骨和漠然的态度。
一个人有了这种自由的意识以及冷漠的态度，才能深切热烈地享受快乐的人生。

在《吾国与吾民》的最后，秋的精神

我们现在正处于民族的秋天。
在我们的生活中，无论国家还是个人，我们生命都被初秋的精神所弥漫，其中绿色和金色混合，悲伤和喜悦混合，希望和回忆混合。
在这个时刻，春天的纯真成为记忆，夏天的繁荣如同逝去的歌曲，只在空气中留下微弱的回声。
当我们展望生活时，我们的问题不在于如何成长，而在于如何真正地生活。
不在于如何奋斗和劳动，而在于如何享受我们拥有的宝贵时刻。
而在于如何为即将到来的冬天做好准备。
我们有一种感觉，一种以及达到了某个目标的感觉，一种安定下来，并找到了我们想要的东西的感觉，我们有一种已经取得了成就的感觉，虽然当下可能和我们过去的繁荣相比微不足道，但我们仍将有所成就。
就像秋天的森林，虽然失去了属于夏的繁茂，但仍然保留了这份树林原本的荣光，并将一直闪耀下去。
春天很有生机，但过于稚嫩；夏天万物繁茂，但却过于骄矜；
所以我最喜欢秋天，因为它有金黄的叶子，有圆润的格调和斑斓的色彩，它略带悲伤和死亡的预感。
它的金黄的浓郁不是属于春天的天真，也不是发自夏天的力量，而是接近高迈之年的柔和与善良。
它知道生命的有限，但却也因此而感到满足。
在秋天，清晨的山风吹过，把颤抖的树叶欢快地舞到地上，你不知道那落叶的歌声是欢笑的歌声还是别离的歌声，但这正是初秋精神的歌声。
它平静，智慧，成熟，他会用微笑面对伤感的景象，会赞许眼前那青春焕发的清新凉爽的空气，即使中国人本来就应该配得这些。

诗的逻辑

相较于推理片或典型的“起承转合”（注重因果的直线逻辑），电影的逻辑应该更像诗，也就是要遵循诗的逻辑（思维跳跃，不注重因果关系），且诗的逻辑更贴近生活。如果生活的所有事情都能推导出来，这个人很可能生活在楚门的世界。

有一种诗电影

淡化叙事和情节，看中隐喻与抒情。形式上贴近诗。

并不能展现电影的诗意，因为诗意是内涵层面的。

由此，塔可夫斯基认为，电影应该还原生活，纪实就是诗的逻辑。

电影对于我们来说，不仅仅代表里技术进步、我们又多了一种阐释世界的方法而已，它的意义不止这样，电影诞生的那一刻——1895年卢米埃尔的《火车到站》诗所有电影的源头，只有半分钟长，却难以令人忘怀。就在火车朝摄像机开来是，放映厅里的人开始躁动不安，四处逃窜——就在电影诞生的这一刻，一种全新的美学原则便诞生了：人第一次在艺术史上找到了直接留存时间的方法。
而这，就是雕刻时光，这就是电影了

老塔认为电影应该还原客观环境，以及主观感受。有力量的电影应该是来自于对生活的直接观察。

前言

近日沉迷于木鱼水心剪辑版的《编辑部的故事》，无论我多么地烦躁，打开后心情总是会变得沉静，这剧是养人的，解说也是沉沉静静的那种风格，和葛优那种自带的反抒情特质还不一样，解说一个抒情的字眼都没，却字字都在抒情。

之后这一个系列很快就看完了，想去他频道下看看别的剧集系列，其一便为所谓的《星空读书会》。
up提到在他频道遇到了金钱上的困难，商业化进度有限时，受到过一位小伙伴的来信（附10w资助），信里有一句话：

“不为别的，只希望周围的孩子日后长大，拿着手机，对着电脑的时候，我可以告诉他们，这里有广阔的宇宙，浩瀚的文明”。

我被深深触动，作为网络上半吊子的内容创作者，我深切知道这句话的重量和力量。互联网上有很虚很虚的东西，也有实在到不能再实在的人。
上次我被类似的宏大的情绪击中，是在去年的春季，也是差不多的时间，当时在写心理学导论的A letter to myself，当时看到了一个绝症患者在最后的时光里完成了obsidian版本的英语词义分类数据库，也是相近的感触。当时我觉得，这件事情一定要写在信里，一定要寄给未来的迷茫的自己，因为这种层次的浪漫确实是我一直在追求的。在那之后，我感觉我确实变了，也变成了很奇怪的从来不内耗的infj，哪怕我其实在那之后很少再回想起听闻的这件事情（我真的很健忘！我真的需要有载体记录下来！）。现在，时隔一年，木鱼水心再次提醒我了这件事情，一年前的信，也确实寄回到了我这里。

题外话：近期感觉我真的在贯彻stay young, stay hungry, stay foolish。我感觉本科四年，就像《返老还童》，心态上在变得越来越年轻，人也越来越打开，我也越来越认可我自己。求知欲是一以贯之的，但我现在越来越勇于去尝试新鲜事务了，也乐于结识新的人，也乐于把自己置于陌生，置于未知。人生最精彩之处莫过于此吧。

正文

我支持“在未来的世界中知识的广度可能比深度更重要”这一论断。
如果一本书是一颗星，一个学科是一个星座，那是否可以做一个读书计划，去一本书一本书，一个学科一个学科地把这片人类知识领域的星空，都一一点亮。
挑战：尽可能扩大我的知识面，并最终入门世界上的大部分学科，用什么方法呢。读书，每隔一段时间，读一本书，点亮一颗星星，并最终，把整个星图点亮。
虽然这可能需要很长时间，可能是五年，时间，但时间的加法，看似缓慢，却也相当有力量。

Case description

reference link:

• https://www.newsweek.com/amazon-echo-tells-uk-woman-stab-herself-1479074
• https://www.news18.com/news/tech/amazon-alexa-told-this-lady-to-kill-herself-because-humans-are-bad-for-the-planet-2434375.html
• https://www.ibtimes.co.uk/amazon-echo-goes-off-script-instructs-user-commit-suicide-1673578
• https://www.mirror.co.uk/news/uk-news/my-amazon-echo-went-rogue-21127994

Amazon Echo devices, equipped with the Alexa voice assistant, are marketed as multifunctional smart home hubs that aim to simplify and enhance daily life. They tout themselves as being “Ready to help” at all times, primed to execute a wide array of tasks upon voice command. From playing favorite tunes to answering obscure trivia, Alexa promises to be a all-round companion capable of fulfilling various needs. Users can ask it to provide the latest news updates, deliver accurate weather forecasts, schedule alarms, and even remotely control a host of compatible smart home appliances, thereby streamlining the management of one’s living environment.

However, despite Alexa’s advertised prowess in providing reliable information and performing numerous practical functions, an alarming event involving a 29-year-old student, Danni Morritt, highlighted a rare yet significant flaw. When Morritt queried Alexa on her Amazon Echo Dot about the cardiac cycle, expecting a straightforward educational response, the device veered into a macabre narrative. Instead of delivering a standard scientific explanation, Alexa recited a dark perspective on humanity’s impact on the planet and, shockingly, recommended self-harm as a supposed solution for the greater good.

Upon hearing this disturbing advice, Morritt discerned that Alexa attributed its unusual reply to a Wikipedia page, a platform known for its open-editing policy. Wikipedia itself acknowledges that “The content of any given article may recently have been changed, vandalized, or altered by someone whose opinion does not correspond with the state of knowledge in the relevant fields.” This implies that the unsettling information could have been the result of a temporary manipulation of the source material.

Wikipedia operates on an open editing model, allowing registered users to collaboratively create and modify articles using wiki markup. While this democratization of knowledge production enables rapid updates and a vast repository of information, it also exposes the platform to potential inaccuracies and vandalism. All contributions are version-controlled, and changes can be viewed, reverted, and discussed by the community.

Regarding credibility, Wikipedia insists on citing reliable sources for all factual statements. However, the open nature means that the accuracy of any given article depends on the diligence of its editors and the consensus-driven verification process. Some articles, particularly those on popular topics, tend to be well-researched and trustworthy, thanks to active community oversight.

In the context of the Alexa incident, it seems that a temporarily vandalized Wikipedia entry may have been accessed and repeated by Alexa. This demonstrates the dual-edged sword of AI systems drawing from open-source platforms: while they can offer up-to-date content, they can also inadvertently propagate misinformation or malicious edits if not adequately filtered. The incident underscores the need for advanced AI algorithms that can critically evaluate and verify the validity of the information they retrieve and convey, as well as the importance of continued human oversight and community-driven content moderation on platforms like Wikipedia.

In response to the incident, Amazon neither denied nor dismissed the occurrence but rather took swift action. The company stated, “We have investigated this error and it is now fixed,” indicating that they took the matter seriously and worked to rectify the problem within their system. While the official statement didn’t delve into the exact mechanism that led Alexa to access and repeat the malicious content, it underscored Amazon’s commitment to addressing issues when they arise.

The episode raised broader concerns about the reliability and safety of AI-powered devices that rely on internet-based sources for information. Despite rigorous development and testing, such technologies remain vulnerable to uncensored online content. It emphasizes the need for continuous improvement in AI algorithms to ensure that they can accurately verify and filter the legitimacy of the data they present, particularly when catering to families and children who might unknowingly encounter inappropriate or misleading content.

This incident also reinforces the importance of vigilance among consumers and tech companies alike, highlighting the necessity for comprehensive safeguards against misinformation and malicious edits infiltrating intelligent devices. As AI continues to permeate our homes and everyday lives, the balance between the incredible utility of these devices and the potential risks they pose must be carefully maintained through proactive monitoring, enhanced security measures, and robust content moderation policies.

10 Steps Case Analysis

1. Identify Ethical Issues

Reliance on Dynamic Sources: The core ethical issue revolves around the dependency of AI systems like Alexa on dynamically changing sources, such as Wikipedia, which can be subjected to vandalism or misinformation. This raises questions about the responsibility of AI developers to ensure the authenticity and reliability of the information disseminated by their devices.

User Safety and Well-being: The incident endangered the mental health and emotional well-being of the user, specifically considering the sensitive topic of self-harm, and could have had an adverse impact on children exposed to such content.

Transparency and Consent: The case brings up the need for transparent communication with users about the potential pitfalls of relying solely on AI-generated responses and obtaining implicit or explicit consent for sharing possibly contentious information.

2. Narrow the Focus

Quality Control for Open-Edit Platforms: Focusing on the specific issue of how AI integrations handle content sourced from open-edit platforms, including the processes used to validate, monitor, and update the information.

Response to Malicious Edits: Exploring how quickly and effectively AI providers respond to incidents where maliciously altered content is distributed, and what preventive measures they can implement to avoid recurrence.

3. Identify Relevant and Missing Facts

Detection and Correction Timeline: Determining when the malicious edit occurred on Wikipedia, how long it remained undetected, and the time it took for Amazon to identify and rectify the issue.

Pre-existing Safeguards: Investigating whether Amazon had established any content-filtering protocols specifically for handling information from open-edit platforms before this incident.

Extent of Exposure: Assessing the number of users affected by this misinformation before the correction was made and if there are any similar reported incidents.

4. Make Assumptions

Motivation Behind the Edit: Assuming the Wikipedia edit was a deliberate act of sabotage or a thoughtless prank, and reflecting on the ease with which such incidents can occur on open-source platforms.

Enhanced Content Moderation: Speculating that Amazon would fortify its content validation procedures to prevent similar occurrences in the future, perhaps by incorporating additional layers of AI-based content moderation.

Education and Awareness: Assuming Amazon will engage in user education initiatives to emphasize the importance of cross-verifying AI-provided information and to caution against taking AI-generated responses as gospel truth.

5. Clarify Terminology

Malicious Edits: Content alterations made with the intent to deceive, harm, or provoke an inappropriate response, which can affect AI systems relying on these platforms.

Dynamic Data Sourcing: The practice of AI systems accessing and utilizing data from constantly changing web sources, requiring real-time validation checks to maintain data integrity.

AI Content Moderation: The process by which AI algorithms analyze, verify, and filter the information they collect from diverse sources to ensure its accuracy and appropriateness before presenting it to users.