AI rapid development

With the recent development of OpenAI ChatGPT-3 and GPT-3.5 (and with GPT-4.0 coming up this year), a large amount of use cases have become available to the general public. This includes writing SEO-optimized blog posts, extracting certain data from text or summarizing it, or simply having a human-like chat bot. Fortunately, we have been one of the early adopters of GPT-3 since we started working with it 6 months ago.

One of the key innovations of GPT-3 is its use of unsupervised learning. Unlike many previous language processing models that required large amounts of labeled data to train, GPT-3 can be trained on a massive dataset of unstructured text scraped from the internet. This allows GPT-3 to learn the patterns and structure of language from a wide range of sources, giving it a more robust understanding of language than models trained on smaller, more restricted datasets.

Most importantly, OpenAI has given API access to its state of the art model, and even the ability to create fine tuned versions of it (which we use as well).

Our Use Of GPT-3 And Its Impact

For us, one of the most important results of GPT-3 is the ability to read out specific data from unstructured text. Specifically, CVs of contractors contain the wildest content - however, we need specific knowledge such as skills or certain industry knowledge of the freelancer in order to find the best project for him. Even just a list of structured experiences (i.e. the job title and the start/end date) is non-trivial to extract using conventional methods. By using GPT-3 we are able to extract the data, which is necessary to match projects and freelancers without human effort involved - even when we are speaking about specific edge cases.

With this automatic and advanced matching, we are able to quickly match a large amount of contractors to projects in an extremely scalable and fast way - this ensures that ANY project that FRATCH receives, gets a suitable match in a short amount of time. Further, this does not require as much manpower (e.g. staffing managers) as in the past while keeping a high-quality standard all the time.

We now go even a step further - ChatGPT has shown an amazing ability to chat with people. This can be used to write business mails as well. Finding a good match is one thing - but actually communicating in this trilateral setup of us, the freelancer and the client takes up a large amount of time. We are currently seeking to automate this as well, therewith FRATCH on its own is able to match thousands of projects to contractors without limitations through human involvement.

To summarize, GPT-3 enables us to supercharge contractor-to-project matching and recruitment, we:

• structurize CVs
• match contractors to projects automatically
• automate the trilateral communication process between us, the contractor, and the client.

The Tech

From a technical perspective, this involves two main techniques: Collecting our own data and fine tuning GPT for specific tasks, and prompt engineering. Prompt engineering is becoming more important with these impactful models - and prompts are becoming an integral part of a company´s IP, which is why we will not delve too deep into them for now.

Overall, GPT-3 represents a major advance in language processing AI and has the potential to revolutionize a wide range of industries. Its ability to learn from unstructured data and generate human-like text and code makes it a valuable tool for businesses, researchers, and developers. And we are on the forefront of that!