Within the generative artificial intelligence (AI) war, Google is fighting back to defend its lead within the search engine market.

AI chatbot ‘Bard’ based on the next-generation large-scale language model ‘PaLM 2’ was unveiled. By remodeling Google’s seek for the primary time in 25 years, it has made it clear that it is going to not be pushed back by allies similar to Open AI and Microsoft (MS).

MS also fired back by completely opening ‘Bing’, a search engine equipped with ‘GPT-4’, without waiting. The AI ​​search engine Bing was fully opened inside three months of its launch.

Bing’s day by day lively users have surpassed 100 million, and day by day installs of the Bing mobile app have quadrupled since launch. As well as, the market share of the Edge browser has grown for eight consecutive quarters, and it is claimed that Bing’s accessibility continues to extend through the Windows taskbar, which is utilized by greater than 500 million users every month.

As Google introduces a latest search engine with AI chatbot function added, evidently it’s launching a full-fledged counterattack against Microsoft’s swift move to regain the throne within the search market with open AI on its back.

We compared ‘Bard’, an AI chatbot based on Palm 2 for Google’s search service, and ‘ChatGPT’, which is installed in MS search engine Bing.

◆ Language model

Google announced that ‘Palm 2’ will likely be released in 4 sub-models: Gecko, Otter, Bison and Unicorn. Gecko is claimed to be very lightweight and might operate on a mobile device and run interactive applications on the device even when offline.

The precise size is unknown, but we estimate that the PaLM 2 may have between 100 and 300 billion parameters, much smaller than the ‘PaLM’ model with 540 billion parameters.

If this number is near or below 100 billion, the Palm 2 will most probably be smaller than the 175 billion GPT-3.5 when it comes to parameters.

GPT-4 also didn’t disclose the precise variety of parameters. GPT-4 has 12 sub-models ranging in size from 125 million to 1 trillion parameters.

Each models use the ‘Transformer’ architecture, a neural network design that allows parallel processing and long-range dependencies.

◆ Data set

Farm2 has taught over 100 languages ​​and a wide range of areas similar to math, science, programming, literature and more. It uses curated datasets that filter out low-quality or harmful text, similar to spam, hate speech, or misinformation.

Farm 2 may learn from multiple sources of knowledge and mix them in a coherent way, using a method called path learning.

GPT-4 has learned a wide range of data, including nearly every domain available on the Web and 26 languages.

It uses a knowledge set called “Pile,” which consists of 825 terabytes of text scraped from various sources similar to Wikipedia, Reddit, books, news articles, web pages, and more. We don’t use filtering or curation on our data. That’s, it may possibly learn from any text it encounters, but potentially inherit biases or errors.

Google and OpenAI didn’t disclose details concerning the size of their training data sets, but Farm 2’s data set comprises 1.56 trillion words, while GPT-4’s data set comprises 1.37 trillion words. This permits the Palm 2 to provide more accurate and richer text, providing a more extensive knowledge base than the GPT-4.

◆ usage fee

‘ChatGPT’ chatbot service based on OpenAI’s GPT-4 is barely available to paid subscribers of ChatGPT Plus for $20 monthly. Anyone can access GPT-4 free of charge using ‘Bing Chat’, a chatbot of Microsoft’s search engine ‘Bing’, but only accepts a limited variety of questions every day.

Then again, the ‘Bard’ chatbot service based on Google’s language model Palm 2 is free to make use of in 180 countries.

◆ Web Information

One in all the important differences between chatGPT and bard is web access. Bards can fetch responses from the web in real time, while ChatGPT4 relies on data sets by the tip of 2021.

This limitation could make it difficult for chatGPT to offer up-to-date information, while bards can provide the newest answers to questions and up-to-date research findings.

ChatGPT may require access to web browsing features to make use of the Web.

◆ Accessibility

Bard can easily integrate with Google search and other web services, while ChatGPT is restricted to an open AI platform. ChatGPT doesn’t provide a mobile app version, so accessibility is restricted.

Nevertheless, Bard integrates into many Google apps and services to make them accessible to smartphone users.

◆ response

ChatGPT is efficient at generating very comprehensive text responses, but falls short at displaying images in responses.

This will likely be because OpenAI has other models similar to ‘Dal-E’ that may generate images.

Nevertheless, bards can provide images in text responses.

◆ Image Prompt

Bard users can use images as prompts or scan images through ‘Google Lens’ to make use of as prompts. For instance, a user who’s stuck with crafting materials and doesn’t know the best way to use them can click on an image and ask the Bard what to do with it.

GPT-4 may use images as input prompts together with text. In case you command ‘tell me what you’ll be able to cook with the ingredients shown on this picture’ together with an image of the inside the refrigerator, it is going to recommend some dishes.

◆ Voice Prompts

ChatGPT cannot use voice prompts. Bards, then again, can accept the user’s voice as input. Using your voice to offer prompts generally is a convenient technique to get a response while multitasking.

Bards can use text, image, and voice all as input prompts.

◆ Export text

Bards have an export choice to share their conversations. If the user chooses to share, the resulting text of the bard could be exported to Gmail in a format similar to PDF, Word, etc. With these export features, email creation will likely be hassle-free.

ChatGPT comes with various extensions for similar functionality, but bard comes with export as a default feature.

◆ Search option

Bard provides a search option for users to validate the data suggested by the bard. All of Bard’s responses have a small Google search icon ‘Google it’. Clicking it is going to open a search box containing the relevant search term and take you to a different page displaying search ends in the same old way.

◆ Multiple response support

Bard provides multiple response drafts and allows users to pick out essentially the most appropriate response, while ChatGPT can only provide one response at a time.

Bards also generate latest content every time they reply to a prompt. The identical query could be addressed in many alternative ways.

Usually, if many users of ChatGPT ask the identical query, all of them get the identical response. The wording could also be barely different, but the fundamentals are the identical.

◆ Provide plug-ins

Google has announced that Bard will offer various plugins similar to Walmart, Spotify, Uber Eats, Adobe Firefly and plenty of other Google apps.

After introducing a plugin for ChatGPT in March, openAI now provides plugins for over 70 apps.

While ChatGPT’s plugins are only accessible to those that have signed up for ChatGPT Plus, a premium service that costs $20 monthly, Bard’s plugins are free for all users.

Reporter Park Chan cpark@aitimes.com