Thank you, everyone.
In this post, I will analyse Appen from a relatively basic technical perspective. As I am not of Appen, my information may not be professionally accurate. If there are any errors, I welcome corrections from everyone.
Before delving into what Appen specifically does, let's first familiarise ourselves with some technical terms.
AI (Artificial Intelligence): This broad term encompasses technologies that enable machines to carry out tasks typically requiring human intelligence. These capabilities span problem-solving, learning, language understanding, image and sound recognition, and beyond. AI is often categorised into two types: narrow AI and general AI. Narrow AI systems demonstrate intelligence in specific tasks (such as autonomous vehicles or recommendation systems) but lack broad, generalised intelligence.
AGI (Artificial General Intelligence): Also referred to as strong AI, AGI represents a theoretical AI capable of performing any intellectual task that a human can. It would excel in specialised tasks and also learn, understand, and apply knowledge across various disciplines, theoretically capable of completing any cognitive task humans can undertake.
Gen-AI (General AI): This specific term refers to AI technologies capable of generating new content, including text (like chatbots), images (such as image generation algorithms), music, video, and more. Generative AI employs machine learning models (like GANs, Variational Auto-encoders, etc.) to produce novel, unseen outputs useful for applications like artistic creation, media generation, and educational content production.
LLM (Large Language Model):Large Language Models are advanced AI models designed to understand and generate human-like text based on given inputs. Trained on vast amounts of text data, they can perform various language-based tasks such as answering questions, summarising documents, translating languages, and more. These models leverage deep learning techniques, especially transformers, to process and produce language in a contextually relevant manner.
RAG (Retrieval-Augmented Generation): This natural language processing technique involves a language model using external information to enhance its responses. The RAG system merges a retrieval component, which fetches relevant documents or data from a knowledge base, with a generative component (like a language model) that uses this information to generate more accurate and informed responses. This method enables the model to provide answers that are not solely based on its training data but also enriched with current or specific information from external sources.
Agent: In AI and computing, an "agent" is a system or entity that autonomously acts within an environment to achieve specific goals. Agents can be software programmes, robots, or any systems that make decisions or perform tasks on behalf of a user, often learning from their interactions and experiences to enhance their performance over time.
Manual Annotation: This process involves adding labels or annotations to data, aiding machine learning models in understanding the data and making accurate predictions, such as annotating objects in images to train models in recognising various items.
Crowdsourcing: A distributed model for data collection and processing, crowdsourcing gathers information or input from a vast number of internet users (crowd workers) via the web.
Appen is a company that specialises in providing data solutions aimed at developing artificial intelligence and machine learning projects. Its services include data collection and annotation, speech and image recognition, and search engine evaluation, all designed to enhance and optimise AI system performance.
The main products of Appen include the following:
https://crowd.appen.com/
https://datasets.appen.com
https://www.appen.com.cn/platform-overview/
They also launched a new LLM platform (demo video):
https://www.youtube.com/watch?v=kxYtIlhpFyM&t=865s
Competitors:
Through simple online research, it becomes apparent that Appen’s only true competitor is Scale AI. Here’s a comparative look at the strengths and weaknesses of each:
From my personal perspective, the advantages and disadvantages of these technological operations are not absolute; in theory, both entities could easily cross into each other's domains. Scale AI’s valuation could reach $14 billion, largely because they are one of the Y Combinator companies. Sam Altman, one of the founders of OpenAI, was formerly the chairman of Y Combinator. Scale AI provides annotation services for ChatGPT. In terms of financing environment and public attention, Scale AI holds more advantages than Appen.
Annotation is Appen’s core business, and this has been the case since its IPO in January 2015. The success of ChatGPT in 2023 has significantly impacted various industries. Fortunately, Appen has proactively transformed, launching an LLM platform earlier this year. Their demo videos also keep up with the latest industry technologies. However, the extent of ChatGPT’s impact and whether it will completely disrupt the annotation industry remains uncertain.
On 1st September 2023, the Google team published a paper titled "RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback." Google can now annotate and train LLMs using AI. I believe this is the main reason why Appen lost Google as a major client.
source:
https://arxiv.org/pdf/2309.00267Annotation is Appen's main business, involving diverse sectors such as autonomous driving, thief movement recognition, and medical diagnosis. However, its technology does not delve deeply into specific AI applications. Established in 1996, Appen could not have foreseen the impact of technologies like LLMs, including ChatGPT. Like any other company, Appen faces the possibility of obsolescence due to emerging technologies. Appen remains committed to its identity; it cannot simply switch to producing GPUs just because Nvidia is profitable. Instead, it continues to deepen its expertise in its field and seek new opportunities.
Looking at the period from the beginning of 2015 to the beginning of 2022, Appen indeed turned in an impressive report card. If you had invested at the IPO, within just seven years, you would have received dividends equal to 100% of your principal, and the share price at the beginning of 2022 was higher than at the IPO.
What Appen now demonstrates is its ability to leverage its existing platform and client base to actively undergo transformation. It is time to view Appen not for its past achievements but as a newly listed company with a monthly turnover of $15 million. If you are seeking a stock that could increase tenfold in a short period or a company riding on trending topics, then Appen might not be your choice.
(20min delay)