文字入力 / かな漢字変換 IME / Kana-Kanji conversion 機械翻訳 Machine translation 検索 Search 誤り訂正 Error correction 文書分類 Text classification 音声認識 Speech recognition OCR（文字認識） Character recognition 質問応答 Q&A Sure! Sure! Sure! Sure! Sure! Sure! Sure! Sure! 一方、NLPに興味を持ってくれる人も増え、 ヒアリングの機会が急増しました The number of people interested in NLP has increased, and the number of the interviews has increased rapidly.
discussions with more than 40 people after the release of Chat GPT. • Increased opportunities to talk to our CXOs and external experts. GPT・生成AIに関する議論に呼ばれるように I gradually started to be invited to interviews/discussions related GPT / GenAI. • 辻CEOが生成AIに注目しはじめた • 社内のエンジニアイベントでトークした • GenAI began to attract CEO Tsuji’s attention • I spoke at an Engineer’s event in MF
discussions with more than 40 people after the release of Chat GPT. • Increased opportunities to talk to our CXOs and external experts. GPT・生成AIに関する議論に呼ばれるように I gradually started to be invited to interviews/discussions related GPT / GenAI. • 辻CEOが生成AIに注目しはじめた • 社内のエンジニアイベントでトークした • GenAI began to attract CEO Tsuji’s attention • I spoke at an Engineer’s event in MF 今日はヒアリングを通して思ったことを紹介します In today's talk, I will introduce what I thought through these interviews.
抽象的すぎる要望・無理な要望 ◦ ChatGPTで何かしたい ◦ 為替レートを予測したい ヒアリングの内容は両極端 The contents of the interviews are at both extremes. 具体的 Detailed/Concrete 抽象的 Abstract 固有表現抽出 NER ChatGPTで何かしたい Want to do with ChatGPT • Too detailed / concrete ◦ Extract names, addresses, dates etc., from contract documents (NER) ◦ Match the accounting titles • Too abstract / impossible requests ◦ Want to make something with ChatGPT ◦ To predict an exchange rate
I want to extract information such as contractors, contract dates from contracts NERであればGPT-4(V)で十分できそう I think NER can be developed using GPT-4 / GPT-4V
I want to extract information such as contractors, contract dates from contracts NERであればGPT-4(V)で十分できそう I think NER can be developed using GPT-4 / GPT-4V できました！ We released the feature!
I want to extract information such as contractors, contract dates from contracts NERであればGPT-4(V)で十分できそう I think NER can be developed using GPT-4 / GPT-4V できました！ We released the feature! （僕の仕事は…？） (Is there any work left for me?)
I want to extract information such as contractors, contract dates from contracts NERであればGPT-4(V)で十分できそう I think NER can be developed using GPT-4 / GPT-4V できました！ We released the feature! （僕の仕事は…？） (Is there any work left for me?) こういうことをたくさん経験しました I experienced many things like this.
product team 具体的 / Detailed / concrete 抽象的 / Abstract 固有表現抽出 / NER 科目分類 / Title classification 生成AIで何かしたい / Want to use GenAI 具体的な要望はエンジニアが生成 AIのAPIを使えばなんとかなる NLPerの仕事は、生成AIのAPIでできるかどうかを見極めること These can be implemented by engineers using the GenAI API. The NLPer’s job is to determine whether or not it can be done by GenAI.
requests from several products 具体的 / Detailed/ concrete 抽象的 / Abstract 複数プロダクトの要望がある・関連する / Request from several products or related to several products 1プロダクトから要望がある / Request from one product 固有表現抽出 / NER 科目分類 / Title classification 生成AIで何かしたい / Want to do use GenAI 複数プロダクトのヒアリング結果から 共通する課題を見つけて解くのが良さそう We should find and solve common questions across multiple products, from the interview results プロダクト側で実装可能 / Product team can develop it
the GenAI era, there were few requests about ML / NLP. • After the spread of GenAI, we get a lot of requests from product teams. → Issues related to several products can now be identified. 生成AIの時代になって変わったこと What has changed in the GenAI era. 60以上のプロダクトがある More than 60 products
まとめ Summary • After the GenAI era started, we are getting a lot of requests from product teams • We have more opportunities to solve common issues as opposed to individual tasks • Interviewing is very important まだまだNLPerができることはたくさんありそうです！ There are so many tasks for NLPers to tackle at MF!
trafficbrain
ssii
joisino
ymorita613
nanofi
eumesy
kentosasaki
nami_ogawa
o_ob
cassininazir
reverentgeek
thekraken
vanstee
kastner
mongodb
jakevdp
matthewcrist
myddelton
chriscoyier
zakiyama
jrom
