見出し画像

週報 2023-04-23

トピック

写経: 銀行の経営破綻連鎖の潜在的な構造

銀行の経営破綻連鎖の潜在的な構造

シリコンバレー銀行などの破綻について、前々から一度やろうと思っていた潜在的な構造の把握。こちらの記事中にある図を写経してみた。

様々な事象が複雑に絡み合って起きている構造を、ループ図で表現すると全体像を把握しやすい。
前提知識を得るためにはThe HEADLINEの記事が役に立つ。

ピックアップ記事

Google Bard

言わずと知れたGoogleが開発しているLLMであるBardが日本でもβ版が提供された。ウェイティングリストに登録すると、数分で使えるようになる。
評判としてはかなり酷評である。でもそんな言われるほど酷いものなのかはよくわからない。個人的な用途としてはクリティカルな判断材料としては活用していないし、知識を問うようなこともしていない。
実際にBardとChatGPTに同じpromptで入力してみた。user researchという言葉は漠然としすぎていて確かにChatGPTの方が「実施の調査を行うにあたって気をつけるべきポイント」という意味を汲み取ってくれている感じもするが、両者の間にそこまでクオリティの差はないようにも思える。質問次第でどうにでもなりそう。そしてChatGPT無課金勢としてはBardのレスポンスはむしろ早い。

追記: しばらく使ってみるとやっぱりChatGPTの方がより適切な回答を返してくれる感じがしてきた。

-- prompt
What points should we pay attention to when conducting user research from the product, marketing, business, and compliance perspectives?

-- response
Here are some points to consider when conducting user research from the product, marketing, business, and compliance perspectives:

Product perspective:Understand the user's needs and pain points.
- Identify the user's goals and motivations.
- Determine the user's current experience with the product.
- Identify opportunities to improve the product.
- Gather feedback on new product features and designs.

Marketing perspective:Understand the target audience's demographics, - - psychographics, and behavior.
- Identify the target audience's needs and pain points.
- Determine the target audience's goals and motivations.
- Develop marketing messages that resonate with the target audience.
- Create marketing campaigns that drive awareness, interest, and action.

Business perspective:Understand the company's goals and objectives.
- Identify the company's target market.
- Assess the competitive landscape.
- Develop a business strategy that aligns with the company's goals and objectives.
- Identify opportunities to grow the business.

Compliance perspective:Understand the applicable laws and regulations.
- Identify potential compliance risks.
- Develop and implement a compliance program to mitigate risks.
- Monitor and report on compliance performance.

It is important to consider all of these perspectives when conducting user research in order to develop a product, marketing, business, and compliance strategy that is aligned with the company's goals and objectives.

Here are some additional tips for conducting user research:
- Use a variety of research methods, such as surveys, interviews, focus groups, and usability testing.
- Be sure to collect qualitative and quantitative data.
- Analyze the data carefully and draw meaningful conclusions.
- Communicate the findings of the research to stakeholders in a clear and concise way.

User research is an essential part of any product development process. By taking the time to understand your users, you can create products that are more user-friendly, effective, and successful.

Bard

-- prompt
What points should we pay attention to when conducting user research from the product, marketing, business, and compliance perspectives?

-- response
When conducting user research from various perspectives, there are several key points that should be considered:

1. Product perspective:
- Define clear research objectives that align with product goals and user needs.
- Select the right research methods and tools to gather relevant data, such as surveys, usability tests, or A/B testing.
- Ensure that the research is conducted in a user-friendly and non-biased manner.
- Analyze and interpret the data accurately to inform product decisions and improvements.

2. Marketing perspective:
- Identify the target audience and their needs and preferences.
- Use appropriate research methods to gather insights into user behavior, such as focus groups or customer interviews.
- Understand the competitive landscape and the market trends.
- Use the research findings to inform marketing strategies and tactics that resonate with the target audience.

3. Business perspective:
- Align research objectives with business goals and objectives.
- Use research to understand the market demand and competitive landscape.
- Analyze user behavior and needs to identify opportunities for new products or services.
- Use research to inform business decisions, such as pricing or distribution strategies.

4. Compliance perspective:
- Ensure that the research is conducted in compliance with relevant laws and regulations, such as data privacy laws.
- Obtain user consent and handle user data securely.
- Ensure that the research findings are communicated in a clear and transparent manner to all stakeholders.
- Consider ethical considerations when conducting research, such as protecting vulnerable user groups.

Overall, conducting user research from multiple perspectives requires careful planning, execution, and analysis to ensure that the insights gained are relevant, accurate, and useful for informing decisions and improving products or services.

ChatGPT(GPT-3.5)

とはいえ、そもそも使い込んでいるわけでもないので(ChatGPT含め)、併用していると見えてくるものもあるのかも。

読んだ本

隷属なき道 AIとの競争に勝つ ベーシックインカムと一日三時間労働
5章~終章

著者は「Humankind 希望の歴史」を書いた人であり、こちらの方が有名だと思われる。前から「隷属なき道」は持っていて積ん読していたのだが、購入した理由はよく覚えていない。

本書の原題は「Utopia for Realists: And How We Can Get There」であり、リアリストである著者が単なる絵空事ではなく、実際にどうすれば理想的なユートピアを目指せるのか書いた作品である。
AIや機械との競争についてはそこまで言及されているわけではない。第8章で現代では機械化の時代を乗り切るためには抜本的な対策が必要として労働時間の短縮とユニバーサル・ベーシックインカムが必要だと述べている。
ChatGPTの登場により更に現実的なものになったと誰もが感じている。

本書の内容に関して、各章であらゆるデータや論文を引用しながら語られていることは面白く、興味深いエピソードがたくさんあった。そしてユートピアを実現するためには再分配、具体的にはベーシックインカムなどが必要となると主張している。
それはそれでいいとして、個人的にはユートピアとは何なのか具体的なビジョンはあまり語られていないように思えた。昔の人や貧困地域に暮らしている人から見れば、先進国でホームレス生活をしている人でもユートピアを生きているようなものらしい。仮にその傲慢かもしれないがその進歩をマイナスから0にすることだとすると、ある程度の生活水準を保って0を生きる我々には新しいユートピアが必要だという。そのユートピアは「働きすぎないでいい」であったり「差別や過剰な格差がない」であったりと、もちろん大事ではあるがマイナスを消すための延長のように感じた。

とはいえ決められたユートピアを謳歌することが重要ではないため、この本を読むことで結局自分はどうありたいのか考え直すきっかけとなった。

統計学

基礎から学ぶ統計学 2章

WMW(Wicloxon-Mann-Whiteny)検定を使って検定を行う際の作業の流れに慣れる。WMW検定は3人の統計学者の名前からなる名前であるが、MannとWhitenyによる検定は、U検定とも呼ばれる。

WMW検定はノンパラメトリック統計であり、これは母集団がどんな確率分布に従っていようと関係なく使える手法である。
検定の手順は以下の通りで二項検定と流れは同じである。

  1. 帰無仮説$${H_0}$$と対立仮説$${H_A}$$を立てる

  2. 検定統計量$${U}$$を計算する

    1. $${n_A, n_B }$$は標本Aと標本Bの標本サイズ。$${R_A, R_B }$$は標本Aと標本Bの順位和。

    2. $${U_1 = n_A n_B + \frac1 2 n_A(n_A + 1) - R_A}$$
      $${U_2 = n_A n_B + \frac1 2 n_A(n_B + 1) - R_B}$$

    3. U = Min($${U_1, U_2}$$)

  3. $${U}$$の臨界値$${U_{0.05}}$$を数表から見つける

  4. $${U}$$と$${U_{0.05}}$$から有意差の有無を判断する

同時に2章は感謝の章でもある。WMW検定の検定統計量$${U}$$の計算は標本サイズが増えれば増えるほど厄介な計算が必要となるが、過去の偉人たちが検定統計量を計算するマニュアルが用意されている。
また、臨界値$${U_{0.05}}$$に至っては、本来なら帰無分布を作成して棄却域を設定してとこれまた厄介な計算が必要となるが、標本サイズに応じた棄却域における臨界値を予め数表として残してくれている。
これらによって統計を使って何をやろうとするユーザーはかなりの省力化が実現されている。昔の偉人たちマジ感謝である。

練習問題Dは計算することだけに夢中になって、「統計的に優位な差が認められた」と結論するだけになってしまった。結局のところ目的は何だったのか、2つの大小関係を明確にする必要があった。

基礎から学ぶ統計学 3章

統計学を実際の実務で利用するにあたって重要なことが書いてある。
検定には2種類の過ちが存在する。

  • A:有意差なし

    • 1. 帰無仮説$${H_0}$$は正しい

    • 2. 帰無仮説$${H_0}$$は間違い

  • B: 有意差あり

    • 1. 帰無仮説$${H_0}$$は正しい

    • 2. 帰無仮説$${H_0}$$は間違い

A-1は有意差なしと判断して実際に帰無仮説$${H_0}$$(比べるもの同士が等しい)は正しい場合であり、これは正しい判断である。
B-2は有意差ありと判断して実際に帰無仮説$${H_0}$$は間違いである場合であり、これも正しい判断である。1章や2章では主にこの結果を期待していることになる。

B-1は有意差ありと判断したが実際に帰無仮説$${H_0}$$は正しい場合であり、これは第1種の過誤と呼ばれる。これはすんなりと理解できる。もともと検定の過程で棄却域に設定されることが多い5%という数値は、「偶然にしては珍しい数値」として設定されたものだからである。なので第1種の過誤を犯す確率である$${α}$$エラーが有意水準と同じであることは直感的にわかりやすい。

A-2は有意差なしと判断したが実際に帰無仮説$${H_0}$$は間違いである場合であり、これは第2種の過誤と呼ばれる。四象限で捉えた時にこのようなパターンが存在することはわかるが、どのような存在なのかが直感的でない。比較するものの母集団の確率分布が似ていれば第2種の過誤を犯す確率$${β}$$エラーは容易に起こりうるが、その確率は有意水準や標本サイズなどにも影響される。$${β}$$が小さければ小さいほど、有意差なしと判断した場合にそれが真実である可能性が高くなるため、$${1-β}$$は検出力や検定力と呼ばれる。標本サイズが十分に大きい場合は第2種の過誤はほとんど起きないため、標本サイズ$${n}$$を見積もるために検出力分析検定力分析が存在する。

とりあえず肝に銘じておくべきことは、有意差ありとなった場合に帰無仮説$${H_0}$$は間違いであったと結論することは概ね問題ないが、有意差なしとなった場合に言えることは「結局何もわからなかった」ということである。
個人的にぼんやりと思っていたが、やはりそういう解釈が適切らしい。

英語

Speak

当初からなのだけども、文章の中で反応しづらい単語があるのがややネガティブな体験。もちろん完璧な発音ができているわけではないが、単語の発音練習機能で試してみると概ね処理されることも多い。(音節の多い単語は難しい。)

Speak 56日連続

この記事が気に入ったらサポートをしてみませんか?