ChatGPT and Chat
History: Challenges
for the New Wave
Types of AI
Intelligent assistants
Siri、Cortana、Alexa、Google Assistant、Bixby
board，shallow
1-3 exchanges
Task-focused chatbots
Dom the Domino's Pizza Bot, customer service and FAQ bots, non-player characters
narrow，deep
3-10 exchanges
AGI, or "virtual companions"
Eliza、Alice、Cleverbot、Tay、Zo、Huging Face、Replika、ChatGPT、Bard、Bing AI Chat
board，deep
10-100+ exchanges
AI history
1965: Eliza was introduced as the first agent capable of maintaining a coherent conversation.
2011: Commercial systems emerged, represented by Apple's Siri.
2011, Apple launched Siri, which attracted widespread attention and became a commercial product that could answer almost any question. Subsequently, competitors such as Cortana, Alexa and Google Now were launched.
Although initial expectations were high, the functions of these assistants gradually stabilized, mainly being able to play music, provide weather forecasts, set alarms, etc.
2015: Facebook's announcement sparked a wave of task-oriented chatbots.
Conversational agents come in a variety of forms and purposes, including intelligent assistants with broad but shallow knowledge, such as Alexa, and task-oriented chatbots, which have deeper coverage on specific tasks.
As technology improves, costs fall, and specialization in AI grows, not all researchers continue to pursue human-level intelligence.
As large technology companies invested in building and maintaining large information repositories, commercial applications of intelligent assistants gradually emerged, followed by the mass creation of task-oriented chatbots.
From Eliza to modern chatbots like Tay and Zo, the development of chatbots has been full of cycles of hype and disappointment. Platforms like Microsoft’s Zo and Replika attracted millions of users but ultimately failed to maintain long-term user engagement or achieve profitability. While these technologies continue to evolve, they often face sustainability issues due to high maintenance costs and a lack of clear business models.
Current LLMs, such as ChatGPT, face limitations, especially when it comes to generating wrong or inaccurate responses. While these models have great potential, they still struggle with tasks that require human judgment or understanding. Furthermore, LLMs lack a true “user model” and cannot remember past interactions with a user or adjust responses based on the user’s characteristics, which limits their ability to personalize responses.
A major hurdle facing chatbots is finding a sustainable profit model. Many platforms have attracted a large number of users but ultimately failed to achieve profitability due to a failure to monetize effectively. Although some companies, such as OpenAI, have attracted external investors to support their platforms, most chatbot developers have failed to make money from direct user interactions.