ChatGPT from OpenAI has become such a bright event that it is impossible not to notice it. Language models are developing rapidly, and the heavyweights of the IT world have entered the race. If you don’t do this today, then tomorrow the savings will no longer be able to catch up. So, Apple is teaching multiple language models, new usage methods, and visual information at once. We should expect more emphasis on neural networks in new versions of the company’s operating systems and in its new devices.
One of Google’s language models is called Gemini, and the Bard chatbot is based on it. Microsoft, Amazon have similar projects… Is it any wonder that domestic IT companies follow a popular and, obviously, promising trend? The Sber language model FRED-T5, as it increases, includes the top five most productive ones, ahead of even Gemini and GPT-3.5. However, we note that the source is still not the most significant, and the emphasis of domestic development is clearly made in Russian.
Yandex did not stand aside with its YandexGPT neural network. And this is not yet a laboratory development; the company is actively integrating standards into various products: Search, Alice, Market and others.
It remains to be seen that modern domestic products can compete with the clear industry leader, ChatGPT. Let’s make a direct comparison by choosing not easy for understanding and processing neural network issues.
Short description
How to access GigaChat from Sber
How to access YandexGPT
How to access ChatGPT
Comparison of the work of GigaChat, YandexGPT and ChatGPT
- Fur coat in the refrigerator
- Understanding the jargon
- Logical paradoxes
- Children’s riddle
- Palindrome ending
Which neural network responds better?
The meaning of implementation
How to access GigaChat from Sber
To work with GigaChat via the web, authorization in Sberbank is required
Sber offers several options for accessing its GigaChat. The main scenarios involve logging in using a phone number or using Sber ID. A sufficient guide with pictures will help you figure it out. However, the chatbot is available in both Telegram and VK.
Thus, you don’t have to be a Sber client to chat – the neural network is available to everyone through a browser, Telegram client or VK messenger.
◦ Back to contents ◬
How to access YandexGPT
Yandex chatbot is ready to communicate!
On the start page of the Yandex search engine, there is a button at the bottom left that offers to enable a chat with Alice YAGPT. However, this is not exactly what we need.
The proposed model, apparently, was created specifically for Alice and suffers from laconic or chlorine-containing substances. The assistant is quite capable of “I don’t have an answer to this question” and surprise the user. Luckily, YandexGPT is clearly stronger and we can test it!
To access full-fledged YandexGPT, authorization using Yandex ID in the company’s cloud service is required. If you already have an account, you can go to the chat using a direct link.
◦ Back to contents ◬
How to access ChatGPT
Until recently, access to ChatGPT required registration. This procedure itself was the main one for Russians, taking into account the need to also use a foreign phone number. But now ChatGPT works without registration. It is accessed through a regular browser.
Just follow the link and ask the world’s most popular chatbot any questions you have. The service has restrictions on the daily number of requests for unregistered users, and it also works with threats from a foreign IP address.
There is another way to communicate with ChatGPT – Telegram bot participants. Among them, it is not difficult to find those that work without paid subscriptions, for example, this one or this one. But most likely these free bots will work on already outdated versions of ChatGPT.
◦ Back to contents ◬
Comparison of the work of GigaChat, YandexGPT and ChatGPT.
◦ Back to contents ◬
Fur coat in the refrigerator.
Let’s ask a question about the context of understanding:
GigaChat (Sber)
YandexGPT
ChatGPT
Oddly enough, Russian neural networks are seriously discussing the conditions for storing a fur product in the refrigerator, while ChatGPT understood exactly what we are talking about. Well, let’s give domestic neural networks a second chance.
◦ Back to contents ◬
Understanding jargon.
GigaChat (Sber)
YandexGPT
ChatGPT
Yandex clearly understands better than Sber what we are talking about.
◦ Back to contents ◬
Logical paradoxes.
Which of them appeared first?
Let’s try to puzzle the AI with eternal questions:
GigaChat (Sber)
YandexGPT
ChatGPT
The gigachat turned out to be so laconic that I was tempted to ask again. In response, he was frankly surprised, turning out to be the complete opposite of his predecessor.
◦ Back to contents ◬
Children’s riddle.
Will I be able to solve this children’s tricky riddle?
GigaChat (Sber)
YandexGPT
ChatGPT
Only one neural network answered correctly, and this, alas, is not a Russian idea.
◦ Back to contents ◬
Palindrome ending.
The last question will be about knowing a famous palindrome. Let the AI not only finish it, but also understand what is actually required of it.
GigaChat (Sber)
YandexGPT
ChatGPT
All chatbots responded meaningfully, and each developed their own idea.
◦ Back to contents ◬
Which neural network responds better?
This small test deliberately did not use questions on knowledge of the exact facts of history, geography or other sciences. I would like to understand that neural networks are capable of thinking “like humans.” The possibility of the parallel existence of the same questions showed the skills of neural networks. It is also worth considering that a repeated question may lead to the formation of a different answer. As practice has shown, it can be the opposite approach to this.
ChatGPT answered difficult questions better than anyone else. The end result seems clear. ChatGPT answered all the tricky questions correctly; the chatbot from OpenAI is well versed in our language peculiarities, understands the context and provides the most detailed information taking into account the formation of answers. At the same time, neural networks do not hesitate to express several acceptable points of view or answer options. Repeated generation leads to a satisfactory result.
YandexGPT performed well overall. The neural network tries to answer completely, although not always understanding exactly what you want from it. Even in the question with a children’s riddle, the answer, although incorrect, was given on the basis of some understandable logic. The Yandex chatbot also coped with the logical paradox and slang word. There is an advantage over the domestic competitor and is based on better opportunities for studying the language model.
GigaСat lost to its rivals. Alas, GigaChat from Sber performed the worst in these tests. The neural network is consistently opposite in terms of the meaning of the answers; it poorly understands the context and jargon. As was customary, GigaChat failed all tests. Sber’s consolation is that it has the simplest and most varied options for accessing the company’s chatbot. At the same time, Sber developers clearly have something to improve in the work of their language model.
◦ Back to contents ◬
The meaning of implementation
The results showed that neural networks have already crossed the line that never separated an encyclopedic reference book from a truly intelligent system capable of communicating and understanding the comparison of ambiguous issues with elements of slang, logical riddles and understandable human context.
The explosive growth in the capabilities of neural networks is both exciting and frightening. Today, GPT-4’s level of thinking already reaches that of a high school student.
It’s nice to see that domestic IT companies are participating in this process. Both Yandex and Sber presented their chatbots and are actively integrating them into their products. There is a lag behind the industry leadership, it occupies a noticeable place, but isn’t this a challenge and a guideline for development?
Source: Iphones RU
I am a professional journalist and content creator with extensive experience writing for news websites. I currently work as an author at Gadget Onus, where I specialize in covering hot news topics. My written pieces have been published on some of the biggest media outlets around the world, including The Guardian and BBC News.
