DEV Community: Fannie Lin

Open-Sourced GPTZero

Fannie Lin — Sun, 29 Jan 2023 00:00:45 +0000

We've open-sourced the implementation of GPTZero!

How does it work?

It uses GPTv2 Perplexity Model from Huggingface to find total perplexity of the sentence.
Later, we break it into individual sentences based on some delimiters, and find individual perplexity per line and find total average perplexities.
The average perplexities is then passed through certain threshold, and if it crosses it, it becomes human-centric, otherwise the text is generated by an AI.

You can find the implementation on GitHub here.
Play around with it with a website here.
How we implement, watch here.
Join our Discord community for discussion and feedback.

See you there~

Powered by Ailiverse

Why did I give up my academia career after Ph.D. and build a start-up?

Fannie Lin — Mon, 23 Jan 2023 11:30:39 +0000

I just got my Ph.D. in Electrical Engineering in 2022 from Nanyang Technological University and Denmark Technical University, with 22 research papers published in 4 years.

I also received a couple of faculty position offers from universities worldwide. But I was hesitant then.

My Ph.D. journey was tough, like a truly "permanent head damage". It was hard because:

- The knowledge gap in AI was huge for me.

My Ph.D. topic was to research How Artificial Intelligence can be applied in the electric power industry. I was trained to be an electrical engineer with no computer science or artificial intelligence background. Thus, learning from scratch about AI and adopting the methodology to real use cases in a brand new problem, was hard. I spent the entire 2 years finishing my first research work and publishing the 1st journey paper.

- The data collection procedure was painful.

As you know, AI models require data for training. To build a surrogate model for the circuit, if using the conventional approach, I needed to collect about 10k data points while 1 data point means 1 set of circuit experiments (taking about 30min). This was a very time-consuming process.

As a Ph.D. candidate, what I have was TIME. Thus I had 4 years on this.

Imagine this happens to a company, like a startup or an MSE which is looking for high growth. What if they want to develop their custom AI model to empower their business,

Do they have enough talents to bridge the knowledge gap?
Do they have enough resources for data collection and model training?**

As reported, to build the ChatGPT model, more than 3 billion US dollars have been spent on it.

Should AI development be always such a luxury?

I wanted to make a change. I believe entrepreneurship was the bridge to bring technology from the lab to real products and people's real lives.

With Ailiverse, we are holding a vision to make deep learning accessible to ALL. With our award-winning breakthrough in domain adaptation along with few shot learning, Ailiverse NeuCore can make computer vision development easier than ever:

Only 10% of the training data is required.
Training can speed up to 10X.
No expertise in AI is required to build a custom model.

With this, you can build your computer vision models in Minutes!
We are just launching our Beta product, looking forward to any kind of feedback!

PS: Ailiverse has recently been featured by Sequoia India and Southeastern Asia Spark Program, which is an important recognition of our vision!
_

SaaS for SME or mid- size enterprises?

Fannie Lin — Sun, 22 Jan 2023 13:04:26 +0000

discuss

This is defenitely a funding winter. But as founders, we need to stay strong and learn from all the mistakes.

Are you a founder building a SaaS solution？What do you think will be the difference of building and selling your SaaS to SMEs (headcounts 1-99) versus to Mid-size enterprises (headcounts 100-999)?

Share your thoughts and learning with us below.

Any startup here is competing with big boy?

Fannie Lin — Mon, 09 Jan 2023 04:22:03 +0000

Wondering any startups here are competing with super big boys? Like google / amazon / datarobot, etc? Share your experience, thoughts and lessons below. Let's chat.

discuss

We created a new Image Segmentation model, seeking your feedback

Fannie Lin — Thu, 05 Jan 2023 13:45:58 +0000

We are building an API solution for computer vision to make model development easier ~

Even though we have been aware that a lot of repetitive work can be replaced by computer vision much more efficiently, ❌not all companies have started their transformation journey.

👉This could be due to the following reasons:

Model development is expensive in general.
Talent crunch.
Lack of training data, etc.

💪We want to change this scenario, to make it more accessible：

with no need for professional AI knowledge,
with less training data
and less training time, than ever.

Thus, we built this API product, which enables users to build their computer vision models in Minutes. For example, to segment ships with only 4 data points and 52 seconds of training like this: https://www.youtube.com/watch?v=2SFMvUk9WYk

And another example is to build a 🧢PPE model (people protection equipment detection), here's how:
① Annotate 5 data points;
② 🚀Wait for 46 seconds for training.
③💣 Your model is ready to go, integrable, and scalable.

The inference result is shown below:

🚀We are about to launch our beta product and thus are seeking early users to help us iterate and improve! The whole building process is usually less than 10 Minutes per model including data annotation.

😊If you are willing to help us, please leave your info here (https://www.ailiverse.com/waiting-list) and we will provide you with access to our documentation and GitHub repo soon! Thank you in advance! ♥