2023-07-26T08:00:26.000Z info Created fine-tune: ft-pYZvA9NP5ddBbe6ud6Mexjne
2023-07-26T11:48:44.000Z info Fine-tune costs $0.01
2023-07-26T11:48:44.000Z info Fine-tune enqueued. Queue number: 0
2023-07-26T12:00:40.000Z info Fine-tune is in the queue. Queue number: 3
2023-07-26T12:12:46.000Z info Fine-tune is in the queue. Queue number: 2
2023-07-26T12:13:18.000Z info Fine-tune is in the queue. Queue number: 1
2023-07-26T12:13:31.000Z info Fine-tune is in the queue. Queue number: 0
2023-07-26T12:14:44.000Z info Fine-tune started
2023-07-26T12:15:43.000Z info Completed epoch 1/4
2023-07-26T12:15:44.000Z info Completed epoch 2/4
2023-07-26T12:15:45.000Z info Completed epoch 3/4
2023-07-26T12:15:46.000Z info Completed epoch 4/4
2023-07-26T12:16:05.000Z info Uploaded model: curie:ft-questetra-2023-07-26-12-16-05
2023-07-26T12:16:06.000Z info Uploaded result file: file-526WjgJtypVc9A2gcR1BWh9m
2023-07-26T12:16:06.000Z info Fine-tune succeeded
▼16行JSONLの場合
2023-07-26T06:57:15.000Z info Created fine-tune: ft-lN1WPLOp8rtoalbMG5Nctc0T
2023-07-26T10:19:09.000Z info Fine-tune costs $0.04
2023-07-26T10:19:09.000Z info Fine-tune enqueued. Queue number: 1
2023-07-26T10:21:50.000Z info Fine-tune is in the queue. Queue number: 0
2023-07-26T10:22:11.000Z info Fine-tune started
2023-07-26T10:23:16.000Z info Completed epoch 1/4
2023-07-26T10:23:19.000Z info Completed epoch 2/4
2023-07-26T10:23:23.000Z info Completed epoch 3/4
2023-07-26T10:23:26.000Z info Completed epoch 4/4
2023-07-26T10:23:43.000Z info Uploaded model: curie:ft-questetra-2023-07-26-10-23-43
2023-07-26T10:23:44.000Z info Uploaded result file: file-tOsEJP0WH3BTcLxyBMz1aLX2
2023-07-26T10:23:44.000Z info Fine-tune succeeded
On July 6, 2023, we announced the deprecation of ada, babbage, curie and davinci models. These models, including fine-tuned versions, will be turned off on January 4, 2024. We are actively working on enabling fine-tuning for upgraded base GPT-3 models as well as GPT-3.5 Turbo and GPT-4, we recommend waiting for those new options to be available rather than fine-tuning based off of the soon to be deprecated models.
GPT-4 is our most capable model. Millions of developers have requested access to the GPT-4 API since March, and the range of innovative products leveraging GPT-4 is growing every day. Today all existing API developers with a history of successful payments can access the GPT-4 API with 8K context. We plan to open up access to new developers by the end of this month, and then start raising rate-limits after that depending on compute availability. Based on the stability and readiness of these models for production-scale use, we are also making the GPT-3.5 Turbo, DALL・E and Whisper APIs generally available. We are working on safely enabling fine-tuning for GPT-4 and GPT-3.5 Turbo and expect this feature to be available later this year.
GPT-4 は当社の最も高性能なモデルです。 3 月以来、何百万もの開発者が GPT-4 API へのアクセスをリクエストしており、GPT-4 を活用した革新的な製品の範囲は日々増加しています。 現在、成功した支払い履歴を持つすべての既存の API 開発者は、8K コンテキストを備えた GPT-4 API にアクセスできます。 今月末までに新しい開発者にアクセスを開放し、その後はコンピューティングの可用性に応じてレート制限の引き上げを開始する予定です。これらのモデルの安定性と実稼働規模での使用の準備に基づいて、GPT-3.5 Turbo、DALL・E、および Whisper API も一般公開しています。 私たちは GPT-4 および GPT-3.5 Turbo の微調整を安全に有効にすることに取り組んでおり、この機能は今年後半に利用可能になる予定です。
Used for Company-wide Approval Flows such as general affairs applications, training reception, and company car usage. Achieved streamlining and efficiency in operations through digital transformation (DX)
Honda is promoting internal digital transformation (‘DX’) to provide better products and services by utilizing data and digital technology. By reviewing internal business processes, including operating and manufacturing processes, and promoting DX, we aim to improve the efficiency of existing operations and expand the creation of new value.
We are reforming our administrative work with the aim of using the time saved by streamlining through DX for creative work that should be done by people. The first thing we did was to organize our work and clarify roles and tasks. By handling tasks that both people and the system are good at, work efficiency and quality can be further improved. We also aim to utilize artificial intelligence in the future. To utilize artificial intelligence, it is necessary to organize the tasks that both people and artificial intelligence are good at.
Visualization of business processes is the most important factor in organizing increasingly complex operations. There are several systems for digitizing forms as workflow tools, but there are few services that organize tasks (work processes) by swimlane (person in charge of processing) and visualize them in workflow diagrams. We consider that the most important items for business reform are workflows that follow standardized rules and have good visibility, and that the workflow diagrams created can be operated as a system as they are. We chose Questetra BPM Suite (“Questetra”), which enables workflow diagrams to be created using the international standard (ISO 19510) Business Process Model and Notation (BPMN).
Workflow diagrams created with Questetra can be operated as workflow apps (systems) as they are. Systemization construction managers and checkers can see the problem points of the business before systemization just by looking at the workflow diagram. Therefore, by organizing operations and rationalizing them through systemization at the same time, they can get closer to fundamental reform. It is also possible to build up small improvements while operating the system, and agile development can be achieved.
What are the selection criteria?
It can be used to establish workflows for all departments and employees, and to reform complex administrative processes. The problem for administrative departments is that once a system has been created by the IT department, modification of the system is a major undertaking, which means that changes cannot be made and the system can be difficult to use. There were also basic problems such as not being able to get out of the paper application process because it was not the right time to develop the system in the first place. Therefore, everyone needs to be able to develop their own workflow apps.
Questetra not only offers good process visibility, but also no-code system development. This low development learning cost was a key selection point. In addition to the digital transformation of simple paper approval tasks, the transferring of increasingly complex business processes is also a target of DX. To organise complex workflows it is important to be able to draw workflow diagrams in accordance with BPMN, and to have visibility and common understanding: with Questetra, complex workflow apps that mix multiple departments, personnel and system processes can be easily created, and task and business reviews can be achieved while building Questetra apps.
Furthermore, Questetra is capable of API integration with external cloud services. Honda uses services such as Microsoft products, etc. Integration between Questetra and Microsoft products such as Excel and Sharepoint was an important factor in the selection process, as it reduces the man-hours required for saving and transcribing. This was an important point in the selection process. Another selection point was the ability to convert entered data items to PDF and easily send them as email attachments, which is a frequent task.
What type of work is it used for?
It is used for a variety of tasks, including internal applications and tasks requiring review.
As for application work, we have started using the system for all administrative processes at Honda, including software use, training reception, use of company vehicles and applications for general matters.
An example of this work is the construction of FAQs for internal use. In this task, the secretariat sets frequently asked questions and allocates the task of preparing a ‘draft answer’ to the person in charge of the relevant area. Review and proofreading tasks are then sequentially assigned to reviewers and proofreaders, and the completed FAQs are automatically posted on the groupware and shared as the latest information with all concerned. In the future, we aim to replace proofreading and translation tasks with AI to further improve efficiency.
What have been the effects of introducing the system?
By using Questetra, the visualization of business processes has progressed, and efficiency has been improved through the elimination of tasks, rational process reviews and the sharing of duties. For example, in the case of the aforementioned FAQ management work, it was originally necessary for the drafters and reviewers of to get together and proceed with the work through meetings. However, by using Questetra, the tasks between the people in charge have been clearly defined, and the passing of tasks has been automated, allowing the people involved in the work to proceed efficiently at their own pace. This eliminates the hassle of writing emails, which is a common problem with email-based handover, and allows the participants to concentrate on the tasks assigned to them. The common problem of unread emails is also eliminated and tasks can now be carried out without fail. In addition, the administration office can now check where tasks are stalled, so it is possible to monitor progress.
Other examples of this include General Affairs applications that involve work requests to contractors. The application forms are automatically stored in a folder built within Microsoft’s cloud, and we have started sharing that folder with the contractor. Until now, we had to transcribe the application forms into an Excel form, upload and store the attached documents submitted by email into the folder, and send them to the consignee by email. By automating these complicated tasks with Questetra, centralized data management and increased efficiency can be promoted.
The digital transformation of these administrative tasks is also being promoted with the aim of laying the groundwork for essential reforms, such as data utilization and the integration of people and AI, in anticipation of next-generation reforms.
What is your future direction?
We promote the use of Questetra in all administrative tasks within the company. We have established an in-house training system so that all employees can create their own workflow applications and respond flexibly to changing business requirements. To this end, we have established a variety of support systems, such as workshops according to skill level, support and consultation by experts, and a community of users.
The system provides an environment in which anyone who applies can build workflow applications. On the other hand, because the environment can be used by anyone, confirmation meetings are held to teach construction know-how while minimizing unforeseen obstacles arising from misconfiguration. The confirmation meetings also have another aspect: by having DX experts check the processes, we are able to propose more rational processes and build an easy-to-understand UI. In this way, we encourage operational departments to simultaneously reform processes and evolve DX to accelerate reform.
In the future, Questetra will continue to be used as the core foundation for linking APIs with external microservices and replacing tasks with artificial intelligence, in order to respond to Honda’s diverse operations in a short period of time.
Honda values a corporate culture of originality and challenge. We want to spread the culture of youthfulness and ingenuity to the frontline once again by reviewing the wasteful and routine work that we take for granted, and by building up reforms that are familiar to us as DX, and we hope to pass this culture on to our products. We consider process reform by Questetra as one of the important services that will play a role in Honda’s reforms.
クエステトラは、SaaS形態での高品質なサービスを実現しうるクラウドインフラを選定し、その運用を行います。現状の多くのサービスは、Amazon Web Services社のインフラを活用していますが、個々のシステム要素に関する詳細についてはセキュリティ上の理由により非公開としています。各通信は同一資源共有度(マルチテナンシーレベル)の異なる複数のアプリケーションによって処理され、お客様のデータは、お客様のシステム操作によってのみ追加・参照・更新・削除が行われます。お客様は、必要に応じてCSV形式(ファイル型データ等は含まれない)等にて一括ダウンロードする事も可能です。なお、お客様のデータは、お客様毎に、日次以上の頻度にて自動バックアップされます。最終のバックアップデータは、極めて耐久性の高い(オブジェクト耐久性イレブンナイン：100億分の1)ストレージに格納されます。
最近は生成 AI に関するニュースを聞かない日はありませんね。テキスト系では Google Vertex AI PaLM API 等も気になりますが、まだまだ ChatGPT の話題が一番多いでしょうか。
弊社サービス Questetra BPM Suite のお客様からも生成 AI の話題をよく聞きます。その中で Azure OpenAI Service の ChatGPT と Questetra BPM Suite との API 連携についての相談があります。
※Questetra BPM Suite について： クラウド型の業務プロセス管理システム (SaaS BPMS) で、業務フローの設計から自動化、進捗の追跡まで行うことができます。視覚的にわかりやすい業務プロセスと API の連携により、業務の効率アップを実現します。
本記事では、本家 OpenAI 社が提供する ChatGPT と Microsoft 社が提供する Azure OpenAI Service の ChatGPT との違いに触れつつ、Azure OpenAI Service の ChatGPT と Questetra BPM Suite との API 連携方法について説明します。 （Azure OpenAI Service の ChatGPT はモデルデプロイまでできているものとし、Azure OpenAI Service の操作方法等については触れません。ご了承ください）
※本記事では、以下の略称表記を使います。 Azure OpenAI Service → 「AOAI」 Azure OpenAI Service の ChatGPT → 「AOAI ChatGPT」 本家 OpenAI 社が提供する ChatGPT → 「本家ChatGPT」 Questetra BPM Suite → 「Questetra」
※OpenAI platform Introduction（OpenAI社サイト） （Overview より）「At OpenAI, protecting user data is fundamental to our mission. We do not train our models on inputs and outputs through our API. Learn more on our API data privacy page.」 （訳）OpenAI では、ユーザー データを保護することが私たちの使命の基本です。 API を介した入力と出力に関してモデルをトレーニングすることはありません。 詳細については、API データのプライバシー ページをご覧ください。
上記の「AOAI ChatGPT」向け「アドオンXML」を使うことで、手強い Microsoft 社の API リファレンスと格闘しなくても「AOAI ChatGPT」との連携が実現できます！（「AOAI ChatGPT」向け「アドオンXML」の設定については、上記リンク先の Notes をご確認ください）
複雑化する業務の整理には業務プロセスの可視化が最も重要です。ワークフローツールとして帳票の電子化を行うシステムは複数存在しますが、スイムレーン（処理担当者）ごとにタスク（作業工程）が整理されており、ワークフロー図で可視化されているサービスはあまり存在しません。「標準化されたルールに則ったワークフローであり視認性が良い」「作成したワークフロー図をそのままシステムとして稼働させられる」などが、業務改革の最重要項目と考え、国際標準（ISO19510）であるビジネスプロセスモデル表記法（BPMN：Business Process Model and Notation）でワークフロー図が作成できるQuestetra BPM Suite（以下、Questetra）を選びました。
さらに Questetra は、外部クラウドサービスとのAPI連携が可能です。Honda では Microsoft 製品などのサービスを利用しています。Questetra と Microsoft 製品のエクセルやシェアポイントとの連携は保存や転記の工数を削減できるため、選定の重要なポイントでした。また、入力したデータ項目をPDF化し、簡単にメールに添付して送付できることも、頻度の高い業務として選定ポイントの一つとなっています。
これらの機能により、生成 AI を利用した回答の精度が向上しています。正しい問いが正しい回答を導くという点で、人と同様の効果が得られています。
今後の方向性をお教えください
利用業務の拡大と生成 AI 活用
オペレーション作業（事務作業）の Questetra 化（ワークフローアプリ化）は大きく進展しました。今後は、製品開発領域への活用と生成 AI の組み込みが重要と考えています。製品開発領域のさらなる合理化は必須です。全社レベルの業務は基盤開発として多くのツールが導入・運用されていますが、開発現場レベルのプロセスやアイデア創造など、さらなる取り組みが求められます。
また、Questetraの利点である「担当者ごとに生成 AI を設定できる」価値を活かし、定型業務のAI化を推進することが重要です。現在はチャット形式でのAI利用が進んでいますが、今後はQuestetraのプロセスに組み込み、効果の定量化や品質向上を目指していく必要があります。
Kyoto, Japan, Aug. 14th, 2023, Questetra, Inc., the global SaaS provider of Business Process Management (BPM), today announced that they have published the new version 15.1 of the No-Code development platform Questetra BPM Suite. The new version 15.1 allows controlling mass email delivery from the workflow by linking with the cloud-based email delivery service Twilio SendGrid.
With the growing need for corporate marketing automation in recent years, there is also an increasing need for automatic delivery of personalized emails. Questetra BPM Suite has been traditionally providing items that can send emails from workflows, but it was unsuitable for mass distribution.
The new version 15.1 adds a built-in item that can automatically handle Twilio SendGrid. Therefore, you’ll be able to create full-fledged email marketing processes like nurturing and newsletters. In addition, by linking the workflow and Twilio SendGrid with an API, it will be possible to automatically insert workflow data into emails, preventing erroneous input of delivery data.
【What is Questetra BPM Suite】
Questetra BPM Suite is a cloud-based business process management system (SaaS BPMS) that allows you to design, automate, and track the progress of business flows. By linking visually easy-to-understand business processes and APIs, you can improve the efficiency of your business. Furthermore, due to its ease of customization, it is possible to implement process management that matches the characteristics of the organization. Since it is cloud-based, it can be used in a remote work environment and reduces operating costs, contributing to improved communication and cooperation within the company. (Examples of Business Flow Diagrams: https://questetra.zendesk.com/hc/en-us/articles/360012492211 ）
【Twilio SendGrid Auto-processing Items】
By using the following automatic processing items you can build a mechanism, for example, to automatically send seminar information emails periodically to the email address entered on the material request page.
Twilio SendGrid: Send Bulk Email
This item sends bulk email to the contacts in the specified lists and/or segments on SendGrid.
Twilio SendGrid: Add or Update Contact
This item adds or updates a contact on SendGrid.
Twilio SendGrid: Add or Update Contact (Custom Fields)
This item adds or updates a contact on SendGrid with the specified custom field values.
なお「機会」と「強み」が混同されるケースは少なくない。「機会」は社会変化や競合会社変化に起因するチャンスだ。政治の動向、法律や規制の動向、経済や景気状況、海外や国内の社会的動向、技術革新動向、などが該当する。（つまり社内発生イベントは「強み」だ）。ちなみに、機会と脅威の両面を持つ外部環境変化も少なくない。 eg: New Normal After the COVID-19 Pandemic (コロナ後のNewノーマル), Enhancing Economic Security (経済安全保障の強化), Automation with AI Technology (人工知能技術による自動化)
In this second part, we will explain how humans can coexist with AI and still get the job done, along with examples of workflows.
At the Moment, Does AI Need Humans?
Now, in Part I, we examined the possibility that the rise of AI, including ChatGPT, will take away writers’ jobs.
For example, one task that AI is very good at is:
Writing sentences based on given themes and keywords
and its accuracy is improving.
On the other hand, AI has no will or empathy, making it difficult for it to decide the theme or direction of an article, or to get close to the subject of the interview to discuss it in depth. The final review of AI-generated articles is also a human task at this point.
The task-based approach introduced in the previous part of this report suggests that even if AI takes over the responsibility of proposing ideas and drafting drafts, it is only a part of the work process, and the writer’s job itself will not disappear.
So How Can We Coexist with AI?
Now, following the task-based approach described above, let’s consider AI as:
An entity that handles specific tasks within a business process
Then, in order to effectively utilize AI, we must understand:
Which part of the work should be assigned to AI?
How will the work be transferred between humans and AI?
This is the same as when introducing RPA tools.
Incidentally, BPM (Business Process Management) is very useful when considering which part of the business to assign to AI. BPM is a business management method that emphasizes business modeling (visualization), and software called BPM tools support the implementation of BPM.
BPM tools make it easy to create workflow diagrams on a PC, and can also facilitate the smooth transfer of work between humans and AI.
Questetra BPM Suite provides a number of workflow templates that can be used free of charge, and in the following section, we will take the Response Creation Process template as an example and explain in detail how to facilitate the collaboration between humans and AI.
Response Creation Process Drafted by AI
Now, the Response Creation Process presented here is a workflow template based on the motif of a response to the Cabinet at a Diet session.
＜ Workflow diagram of the “Response Creation Process” ＞
Although the diagram above may appear complicated, the basic flow is simple:
Questions are received, and the designated staff member prepares a written response.
The most important point of this workflow template is that AI (ChatGPT) performs the process of drafting the response text. When the author of the response clicks [x1. AI Draft] on the input screen, the work is handed over to the AI. Then, AI (ChatGPT) performs a pinpoint drafting process, and a human finalizes the answer based on the draft.
＜ Sample of “Response Creation Process” Entry Screen ＞
Click here to download the Response Creation Process.
In fact, there are currently attempts to eliminate the need for humans to check and revise AI-generated manuscripts, as in the case of Nihon Keizai Shimbun’s financial summaries. However, excluding these special cases, there are still many cases where AI-generated manuscripts need to be reviewed and revised, so a process like this workflow App, in which humans and AI work together, is more versatile and practical.
Add-ons that can Work with Open AI API
In addition, we have prepared several add-ons (as of April 2023) to smoothly integrate Open AI APIs such as Chat GPT and Whisper-1 with Questetra BPM Suite. By utilizing these add-ons, processes such as dialogue with AI, transcription of voice data, and translation of voice data into English text can be performed automatically.
Below is a list of add-ons that can work with the Open AI API.
★Addon: OpenAI: Chat, Start Starts a conversation with the OpenAI API (ChatGPT). The model used is gpt-4 by default (though it can be changed). Automatically obtains a response (COMPLETION) to the instruction (PROMPT).
★Addon: OpenAI: Chat, Interact with Parameters Communicates with OpenAI API (ChatGPT). Default MODEL is gpt-4 (can be changed). It supports almost all Chat features, such as re-asking questions based on conversation history and assigning advanced parameters. Specifically, sampling temperature, upper % sampling, reappearance prohibition, frequency prohibition, logit bias, etc. can be set.
★Addon: OpenAI: Audio, Transcribe Transcribes text from audio files, using the OpenAI API’s whisper-1 model to convert audio data into text data. Abbreviations and technical terms can be set as summary sentences (PROMPT) for more accurate transcription.
★Addon: OpenAI: Audio, Transcribe to WebVTT Automatically generates a subtitle file WebVTT from a sound file. Audio recorded in an audio or video file is converted to text data by the whisper-1 engine of OpenAI API or other engines. The AI engine Model can also be changed. If abbreviations and technical terms are set as summary sentences (PROMPT), more accurate transcription can be achieved.
★Addon: OpenAI: Audio, Translate into English Translates audio data into English text. English text data is generated from sound or video files containing Japanese or other audio, using the whisper-1 model of the OpenAI API. If you set abbreviations or technical terms as the summary text (PROMPT) in advance, you can achieve more accurate translation.
Finally
Recently, due in part to the rise of ChatGPT, many have voiced concerns about the progress and rise of AI. Some of these opinions have voiced fears that reliance on AI will diminish human capabilities.
For example, if we rely solely on cars and do not walk, the muscles in our legs will naturally weaken and our ability to walk will deteriorate. AI is convenient, but too much reliance on it may diminish our ability to think, choose, and make decisions on our own.
Incidentally, Questetra BPM Suite is software that allows you to easily visualize the flow of business operations on your PC and automatically progress operations along the workflow diagrams you have created. However, it is the job of humans to think about business processes. Therefore, it can be a meaningful and rewarding task to think about how to place AI in the process.
Why don’t you take this opportunity to think concretely about the business utilization of AI tools such as ChatGPT with Questetra BPM Suite? You can start using Questetra BPM Suite immediately after applying for it, and you can also try it for free.
