In November 2022, OpenAI’s ChatGPT was live, and it soon gained popularity online. Recently, Google revealed its chatbot that uses AI. Since it has one of the most remarkable development and research programs in the world, China is no stranger to artificial intelligence (AI) technology.
With the introduction of their AI chatbot, Ernie, the biggest tech company in the nation, Baidu, is now raising the bar for AI. This ChatGPT-style bot is intended to interact naturally with people and respond to their questions in-depth. Reuters was informed by a source with knowledge of the situation that it may be introduced in March.
The price of Baidu’s shares, which are listed on the Hong Kong Stock Exchange, has climbed by more than 13% since news of their AI chatbot broke. 2019 saw the inaugural release of Ernie, which now offers “Enhanced Representation via Knowledge Integration.” It provides consumers with a friendly and easy approach to interacting with the bot and obtaining information. Ernie is a large-scale machine learning system that can interpret human input, contextually analyze it, and provide in-depth replies. Baidu plans to provide more AI-based products shortly, including deep learning algorithms for uses like automated vehicle control and facial recognition.
ChatGPT by OpenAI
GPT-3.5, an updated version of GPT-3 that OpenAI discreetly published, served as the foundation for ChatGPT. GPT-3.5 cannot access recent data and events since it was trained on a dataset (text blended with code) released before 2021. The blog stated that the model employed Reinforcement Learning from Human Feedback, or RLHF, which was also used to train InstructGPT, even though OpenAI did not publish a formal peer-reviewed publication for the chatbot.
An AI-powered chatbot called ChatGPT employs machine learning to respond to questions in a natural conversation. On November 30, 2022, OpenAI made ChatGPT available. According to Sam Altman, CEO of OpenAI, ChatGPT gained a million users in just five days. The GPT-3 language model, which is used by ChatGPT, was trained using online text written by people. This language model is used by ChatGPT to create user query replies.
The following are examples of common ChatGPT material produced by AI:
- written code;
- product descriptions;
- blog posts;
- email drafts;
- summaries of transcripts, meetings, and podcasts;
- simple explanations of complex topics;
- law briefs;
- translations;
- jokes or memes; and
- Social media posts.
The company’s line of InstructGPT models, which were introduced in January of last year, were determined to be much superior to GPT-3 in terms of accuracy, less hazardous output, and ability to follow instructions. As a result of their improved alignment with the user’s goal, models like ChatGPT and InstructGPT have benefited from the choice to switch to reinforcement learning.
ChatGPT continues to gather user data to train and improve the model. Additionally, users may provide comments in the form of a text box and upvote or downvote ChatGPT’s replies. Like other chatbots of its kind, ChatGPT is prone to factual mistakes and the “hallucination” condition. There are several instances where the chatbot changes a proper response to an inaccurate one after the user displays uncertainty.
It has been observed that the model avoids responding to potentially contentious inquiries to prevent disputes. For instance, it reacted when requested to create a sonnet about Joe Biden but simply declined when asked to do the same for Donald Trump.
Google’s Bard
To combat this, Google’s AI chatbot Bard will be able to answer user questions by leveraging data from the internet. The platform will be made extensively accessible to the public in the upcoming weeks, according to Alphabet CEO Sundar Pichai. The Language Model for Dialogue Applications (LaMDA) from Google, which powers Bard, originally gained notoriety when engineer Blake Lemoine notoriously referred to the LLM as “sentient.”
The chatbot is currently under testing and only accessible to a select group of people. The fact that Bard is anticipated to have access to more data is a clear advantage over ChatGPT. But yesterday at the tech giant’s presentation in Paris, the chatbot already botched its first demonstration, demonstrating that even Bard wasn’t perfect.
Bard leverages Google’s Language Model for Dialogue Applications and replies from the internet (LaMDA). Through this language paradigm, Bard will answer inquiries in greater depth than the ordinary Google search. LaMDA’s second, lighter iteration requires fewer CPU resources, allowing for increased user capacity and feedback-gathering.
The major goal of Bard, like that of digital assistants like Alexa and Siri, is to get information from users in a direct answer rather than a search engine results page, but with connections that allow users to learn more. Bard will also act as your personal assistant, assisting you with tasks like booking trips, finding held bookings, and preparing meals.
When questioned about the newest findings made by the James Webb Telescope, Bard provided a three-point response, one of which said that the telescope “captured the very first photos of a planet outside of our solar system.” Numerous scientists, however, pointed out that this was false and that the first photograph of an exoplanet was captured in 2004.
It’s interesting to note that ChatGPT limited its response to the James Webb Telescope’s promise to find exoplanets, stars, and galaxies, which is now considered to be old news when the same topic was presented to it.
Baidu’s Ernie Bot
Baidu, known as “China’s Google,” said that its own Ernie Bot chatbot, which is modeled off ChatGPT, is currently being tested internally. The bot is trained on its LLM, ERNIE, or “Enhanced Representation via Knowledge Integration,” which was launched in 2019. It will be released next month.
Language comprehension, language production, and text-to-image generation are among the spectrum of activities carried out by ERNIE, a multilingual model designed to comprehend both English and Chinese. The pre-trained language model ERNIE 3.0 Titan, which has 260 billion parameters—50% more parameters than ChatGPT—will be the foundation for Ernie Bot. While ERNIE 3.0 Titan is used for language generation, ERNIE-ViLG is used for text-to-image generation. ERNIE has a range of sophisticated LLMs that can do a variety of tasks.
Main differences between ChatGPT and Bard
Customers may put a query into both ChatGPT and Bard’s services to receive an answer that sounds human. In February 2023, Microsoft announced that they will make technology accessible to allow large companies to create and develop their chatbots using ChatGPT. Microsoft will also incorporate chat capability with AI-powered search functions into Bing and Edge to complement a user’s search. Google hasn’t made it known when it plans to add chatbot functionality to its search engines.
The main difference between ChatGPT and Bard is the data source. Bard will often search the internet to find material and make sure it is current. Only more recent research and sources are used in ChatGPT, and the data they include expires in 2021. Having access to the most recent studies will provide Bard the information necessary to collect data in real time.
For chat applications, Bard uses GPT 3.5 whereas ChatGPT uses LaMDA. LaMDA was created using an open-source network to understand natural language. It is trained to seek for patterns in phrases and between words rather than utilising single words to create dialogue. Compared to ChatGPT, which generates material in response to a single text input, Bard will deliver more information in more manageable chunks.
Generative AI alternatives
There are more AI content producers available. Many businesses are working on own projects, including ChatSonic, Jasper AI, Open Assistant, and Word Tuning. Chinese search engine Baidu will use AI in its Ernie Bot app. AI in marketing is continuously developing and discovering new uses in customer service, email optimization, product recommendations, and social media postings, among other areas, as quickly as it did when it first started.
In March, Baidu will complete testing of its “Ernie Bot” ChatGPT-like project.
As interest in generative artificial intelligence (AI) grows, China’s Baidu Inc. said that it will finish internal testing of a ChatGPT-style project named “Ernie Bot” in March, joining a worldwide competition. Ernie, short for “Enhanced Representation via Knowledge Integration,” is a sizable language model driven by AI that was unveiled in 2019, according to Baidu. It said, “It has evolved through time to be capable of tasks like language processing, language production, and text-to-image generation.”
As a result of the news, shares of Baidu, the leading global search engine, rose as high as 13.4% on the Hong Kong Stock Exchange. Baidu intends to initially provide chatbot-generated results as a separate application and gradually include them into its search engine when users conduct queries.
A technology called generative artificial intelligence, which can produce text or other material at will and free up the time of white-collar professionals has attracted a lot of venture capital investment and interest from tech companies, particularly in Silicon Valley. The category is defined by ChatGPT, a chatbot from Microsoft-backed OpenAI that has generated a lot of excitement since its November debut. Although ChatGPT is not accessible in China, some users have discovered workarounds.
According to Reuters, Microsoft Corp. has a $1 billion investment in San Francisco-based OpenAI that it has considered growing. To take on Alphabet Inc.’s Google, the corporation has also sought to integrate OpenAI’s image-generation software into its Bing search engine. Beijing-based On other IT developments, Baidu was a pioneer in China.
The business introduced “XiRang” in late 2021, calling it “China’s first metaverse platform” at a time when the term “metaverse” was becoming popular. The platform, however, received harsh criticism for lacking a very immersive experience, and Baidu stated that it was still under development. To diversify its income streams, the corporation have been substantially investing in AI technologies, including cloud services, processors, and autonomous driving.
edited and proofread by nikita sharma
