How to compete when you can copy an AI startup with Whisper and ChatGPT in 3 minutes?
Hey there! So, I recently recorded a YouTube video where I shared some exciting moments from a conference I attended. One of the startups I encountered really caught my attention. It had this amazing AI-powered solution that could help our company’s hiring process. It got me thinking about how easy it is to replicate such functionality. All I had to do was translate the audio to text and chat with GPT:
Step 1: getting a text from the voice
import whisper
model = whisper.load_model("medium")
result = model.transcribe("job.m4a")Good morning, Neurons Lab team. I found your position for the data scientist, hypothesis testing and deep learning, and I want to apply for it. I live in Italy. My name is Alex Honcher and I've already worked in data science for the last five years. My education is bachelor's degree in computer science and master's degree in mathematics, both in Politecnico di Milano. And first I was working as the intern in Banca Popolare for the one year. And I was working on term prediction for the retail banking part and with the team of four. In this job I was working mainly with SecondLearn and Pandas. And after this I moved and worked for the last four years in a startup. The startup was also FinTech, but investment management related. And I was working on prediction of the markets. I was building machine learning models on different datasets, mainly time series, but I need to clean them. The data was bigger, so I was using AWS SageMaker in order to process it. And I was training models there as well and also deploying them. So I understand the whole cycle of MLops part. And I really like that you have the remote policy. And I think I fit really well with your area of responsibility. I can do modeling and testing. I also have experience in Git. I always use Git. And the skills, my English is good. I do machine learning and statistical modeling. And I really like what you offer and also possibility to work part-time. So please consider my application to the position. Thank you very much. I'll be looking for your feedback. Bye-bye.Step 2: structuring text into a CV
CV
Name: Alex Honcher
Location: Italy
Education:
Bachelor's degree in Computer Science, Politecnico di Milano
Master's degree in Mathematics, Politecnico di Milano
Work Experience:
Internship at Banca Popolare (1 year)
Worked on term prediction for the retail banking sector
Collaborated with a team of four
Utilized SecondLearn and Pandas for data analysis
Startup (4 years)
Worked in a FinTech startup focused on investment management
Developed machine learning models for market prediction
Worked extensively with time series datasets
Performed data cleaning and processing using AWS SageMaker
Trained and deployed machine learning models
Experienced in the entire MLops cycle
Skills:
Machine Learning
Statistical Modeling
Git version control
Proficient in English
Other:
Interested in the remote work policy
Ability to work part-timeStep 3: asking to match a CV and the positions step-by-step
Certainly! Here's a table comparing the candidate's CV with the job description, along with a final match rating:
| Job Description Criteria | Candidate's CV |
| ------------------------------------- | --------------------------------------------------- |
| Statistical modeling and hypothesis testing | ✓ |
| Designing, training, and validating results from machine learning algorithms | ✓ |
| Writing clean, efficient SQL | Likely, based on data science experience |
| Building and testing Python applications | ✓ |
| Building deep neural networks with modern tools (e.g., PyTorch or Tensorflow) | Likely, based on machine learning experience |
| Collaborating via Git | ✓ |
| Machine learning and engineering | ✓ |
| Statistical modeling | ✓ |
| Data mining and analysis | ✓ |
| Advanced English, good communication with business stakeholders | ✓ |
| Documentation development | Likely, based on data science experience |
| Experience with online learning, reinforcement learning, contextual bandits algorithms | Not mentioned in the CV |
| Experience in launching and validating online campaigns using A/B testing or the above-mentioned methods | Not mentioned in the CV |
| Prior experience in online gaming, e-commerce, and entertainment or similar industries | Relevant experience in FinTech and market prediction |
| Remote work from any location | Expressed interest in remote work policy |
| Flexible working hours | Expressed interest in part-time work |
Based on the comparison, the candidate seems to meet most of the criteria mentioned in the job description, demonstrating relevant experience and skills. However, there are a couple of specific experiences not explicitly mentioned in the candidate's CV. Considering the overall alignment, the candidate could be rated as 8/10 for their fit with the job description.The structure is excellent, and you can even score it and package it as JSON or any other format for visualization on the website. But it made me wonder, where is the defense? Where is the uniqueness? How can AI startups differentiate themselves from one another and even from non-AI startups or software companies?
AI startups are in a trouble
AI startups earn less and are harder to build
This is a classic price from Andreessen & Horowitz written in 2019. AI programs are not the same as SaaS programs “just” running in the cloud and have multiple margin hits:
- infrastructure involves ad-hoc human work (data labeling and edge cases handling)
- edge cases are not “edge” per se — most of your business will come from them and you will never have enough data “just to train AI”
- “data effects” are a myth (most of the data collected will be essentially the same) and R&D is not a moat anymore
Algorithms are not a moat anymore (at least not for you)
The graphs, I believe, are already self-explainable. Everyone can run the same open AI models using the same cloud providers the same as you do. I often hear a point about your unique data that no one else has access to. Well, if your data is really that unique, it means that most probably it won’t be enough to train a really good state-of-the-art AI model out of it. You still can build a very decent and unique automated decision-making machine and business out of it, but it won’t be AI business as we see it today.
Back to the roots?
If computation, AI models, data, and prompt engineering are not a moat, then what is?
“Classic” business
This amazing article by NFX raises fundamental questions about your business. I especially like the question “If you remove the word “AI” from your pitch — is it still a winning pitch”? Or this one “Does it take >60sec with ChatGPT to complete the same task? Our voice and recruitment example can be reduced to less than 60 seconds easily. We are back to the roots — understanding real customer pains, analyzing the value chain, and injecting automation where it really matters.
“Classic” operations
When everyone has the same LLMs and cloud centers, the competition narrows down to the “golden” triangle of speed, costs, and quality. The new important questions are: Build your own LLM? Run on your own cloud? Optimize the cloud architecture? Own your own hardware? Find a cheaper data labeling team? Speed up the progress of your AI team? Those are not creative prompt engineering challenges for “Chief Prompt Engineering Officers” but rather tedious and complex management tasks.
Conclusions
In an era where AI models and cloud infrastructure are becoming increasingly accessible to all, AI startups face the challenge of differentiation. However, by embracing classic business principles, focusing on domain expertise, and prioritizing operational efficiency, AI startups can carve out their unique position in the market. While algorithms alone may not be a moat, the combination of unique data, customization, continuous innovation, and a customer-centric approach can set them apart from the competition.
