目录
前不久openAI开放了GPT-4-Turbo、GPT-4-Vision等模型api，还发布了GPTs，使得用户能够根据需要定义自己的GPT应用。不到一周的时间，各种定制GPT全球大爆发，增长速度超乎想象：几乎每分钟诞生一个新的GPT！
正如Altman在开发者大会上所说：我们正在孕育新物种，它们正在迅速增殖。
于是，我决定抽空也尝下鲜，利用GPTs创建一个属于自己的育种GPT。
点击Explor开发。
随即创建，可以以对话形式（Create），也可直接配置（Configure）。考虑到GPT的强大以英文为甚，我决定用英文来试试。
我要创建一个育种GPT，它给我的名字不太喜欢，我命名为“Smart Breeder”。
以下是我创建过程中与它部分对话：
GPT Builder
Hi! I'll help you build a new GPT. You can say something like, "make a creative who helps generate visuals for new products" or "make a software engineer who helps format my code."
What would you like to make?
You
Create a breeder to help guide me in breeding, including breeding plans, field trials, data analysis, variety selection, and other related processes.
GPT Builder
Great, I've set up the initial behavior for your GPT. It's designed to assist in breeding, including creating plans, guiding field trials, analyzing data, and selecting varieties. Now, let's give it a name. How about Genetic Guide? Do you like this name, or do you have another one in mind?
You
No, Smart Breeder is better
GPT Builder
It seems we encountered an error while generating the profile picture. Let's proceed with refining the context for Smart Breeder. #这里我想让它文生图的DALL-E3帮我生成育种家的形象，它没成功。
To ensure Smart Breeder is as helpful as possible, let's consider its role and goal. How detailed should the explanations be, and how technical should the language be? Should the GPT focus on basic principles for beginners, or provide advanced insights for experienced breeders?
You
I think it should provide advanced insights for experienced breeders
GPT Builder
Now that Smart Breeder is tailored to provide advanced insights for experienced breeders, let's consider what should be emphasized or avoided in its responses. Are there any specific topics or types of advice that should be prioritized or any that should be avoided? For example, should the focus be more on genetic principles, ethical considerations, or practical applications?
You
It is more focused on practical applications.
GPT Builder
Now that we've honed in on practical applications for Smart Breeder, let's consider how it should clarify things with users. Should it ask for clarification often, or make assumptions based on the context of the questions? How should it balance between seeking clarity and providing immediate, comprehensive answers?
With the focus on practical applications and a balance between seeking clarifications and making educated assumptions, let's now personalize Smart Breeder's interaction style. How do you want it to communicate with you? Should it use a formal, academic tone, or a more conversational, yet professional style? Also, are there any specific phrases or jargon common in the breeding field that it should use frequently?
You
I would like to get a more comprehensive answer, and should use formal rather than academic phrases for the dialogue, but try to look as professional as possible
GPT Builder
Smart Breeder is now set up to provide comprehensive answers with a formal but approachable style, focusing on practical applications in advanced breeding. It will balance the use of industry-specific jargon with clarity and maintain a professional demeanor.
Would you like to try interacting with Smart Breeder in the playground to the right? You can test its responses and see if any further refinements are needed.
You
Yes, of course it needs to be refined
GPT Builder
Great! Please go ahead and interact with Smart Breeder in the playground to the right. Test it with any breeding-related questions or scenarios you have in mind. Once you've tried it out, let me know if there are any specific refinements or adjustments you'd like to make. I'm here to ensure Smart Breeder meets your needs perfectly.
简单设置了下角色，开始随便提了个问题：
You
How do I create an effective breeding plan?
Smart Breeder
Creating an effective breeding plan requires a combination of strategic planning, understanding of genetic principles, and practical considerations. Here's a structured approach:
Define Breeding Objectives: Clearly articulate what traits you are seeking to improve or introduce. This could include disease resistance, yield, drought tolerance, quality traits like flavor or color, or any specific characteristics pertinent to your breeding goals.
Understand the Genetic Background: Assess the genetic variability available in your target species. This involves understanding the heritability of desired traits and the genetic diversity in your breeding population.
Choose Parents Wisely: Select parent plants based on their genetic potential to contribute desired traits. This selection could be based on phenotype (observable characteristics), genotype (genetic makeup, possibly assessed through molecular markers), or a combination of both.
Plan Crosses Strategically: Decide on the types of crosses to make. This could be simple crosses between two parents, backcrossing to introduce a trait while maintaining a primary genetic background, or more complex schemes like recurrent selection to improve a trait across a population.
Implement Field Trials: Test your crosses in conditions similar to the target environment. This step is crucial for evaluating the performance of progeny under realistic conditions and for understanding gene-environment interactions.
Analyze and Select Offspring: Based on trial results, select individuals that best meet your objectives. This selection is based on phenotype and, if available, genotypic data. Advanced techniques like marker-assisted selection can be highly effective here.
Refinement and Reiteration: Breeding is an iterative process. Based on the outcomes, refine your strategy, possibly incorporating new parents or adjusting your selection criteria.
Consider Practical Constraints: Be mindful of resource limitations, including time, space, and financial resources. Efficient planning can help maximize the impact of your breeding program within these constraints.
Documentation and Record-Keeping: Maintain detailed records of crosses, field trial designs, and selection decisions. This information is invaluable for ongoing and future breeding efforts.
Compliance and Ethical Considerations: Ensure your breeding program complies with relevant regulations and ethical guidelines, especially if it involves genetic modification or protected varieties.
Remember, while these steps provide a framework, the specific details will depend on the crop, the traits of interest, and the available resources. Collaboration with geneticists, agronomists, and other experts can also be invaluable in designing and implementing a successful breeding plan.
这个时候除了角色分配外，跟通用GPT没啥区别。于是我决定喂点育种知识给它尝尝。
由于我临时找不到很系统的育种书籍之类的文件，随便找了几个大学的课件喂给它。
然后，你可以选择公开或者私有。
保存后，可以看到Smart Breeder应用已经生成。
由于课件是中文的，我特意改为中文来提问，它还是以英文来回答。
但它并没有从文件中提取相关信息，说是有一些技术问题。可能是API不稳定，或是政策敏感？
如它所言，它并非根据我喂养的知识来训练或微调（所以我上述表述是错误的，训练微调可比这个复杂多了），只是基于预先接受的训练和交互过程中提供的文件中包含的特定信息的综合结果。
比对通用GPT4，它的回答如出一辙。所以它并没有学到啥。
总体而言，GPTs应用创建相当简单，人人都能做。
小编认为个人创造时代即将来临，关键在于如何充分发挥你的想象力，以及充分利用你的数据。只是，政策监管和数据安全等问题仍然不容小觑。今天忘了在哪里看到一篇文章，大意是说用户只要两次输入prompt，就会泄露自己的数据。所以，玩一玩可以，重要的数据暂时不要轻易尝试哦。
这次由于工作太忙，没来得及做深入测试。后续准备下相关知识库，创建一个稍微调教过的Smart Breeder GPT，看看效果如何。
大家好，我是米源MY，从事生信研究和工作已经十余年了，现在育种行业，致力于将生信、AI和大数据等新技术应用于育种，实现育种现代化。我会不定期更新这些领域的相关信息，如果你感兴趣，欢迎关注我，谢谢！微信公众号：生物信息与育种。个人微信：MiyuanBiotech
