Supercharging communications biz with ChatGPT
Though the “ChatGPT proper” at OpenAI (openai.com) is dismal when it comes to sensible responses in the Sinhala language (See previous article, the Big Breakthrough in Colombo with ChatGPT), another implementation by Sri Lanka’s Kapruka.com “ChatGPT in Sinhala” located at https://www.kapruka.com/tools/chatgpt_sinhala.jsp appears to be way ahead as it intelligently responds in the Sinhala language…and “pretty fast” too!
To be clear, it’s uncertain if this is a ChatGPT implementation in its original, full form similar to the native ChatGPT. It is not clear if a dedicated Sinhala language-based LLM is present or if even a Sinhala language-based data set is integrated. Still, the bot measured up considerably well, as if a biological GPT acting naturally in Sinhala, when Sinhala prompts (almost at the level of standard ChatGPT prompts) were issued.
To start off, I asked (in the Sinhala language): “How many Provinces are in Sri Lanka and what are their names?”
Within seconds, it flashed back: “9 provinces, and they are …” (it started to name them one by one, correctly).
Somewhat surprised, I decided to level up.
I asked in Sinhala: “Describe the situation on the spread of cancer in Sri Lanka with data. Using the (same) data, formulate a brief Facebook ad targeting the youth, to create awareness among them on this issue.”
In under 3 seconds, it responded in Sinhala!
“According to WHO, Sri Lanka’s cancer prevalence rate is 79.2 persons for every 100,000 adults. It has been identified that breast cancer (22.9%), cervical cancer (9.4%), and cancer in the large intestine (8.4%) are the most prevalent types of cancer in Sri Lanka. Even though cancer is not as prevalent in Sri Lanka on [1] par with other countries, it is important to be aware of its symptoms. Speak with your physician if you are worried [2] / [3] concerned.”
It was now clear that Kapruka’s “ChatGPT in Sinhala” was levelling up better than I expected, especially for a language bot that could be of “non-English implementation.”
Can it respond to a prompt fed in the negative?
Next prompt: “For which countries a democratic system would not match.”
Response: “There is no single answer to this question since different countries are at different levels in their development, have varied colonial pasts and have[4] different [5] religious traditions. However, some experts have argued democracy may not be suitable to countries with higher levels of poverty, inequality and violence in comparison to countries whose societies are more stable and prosperous.”
Then I posed a question that would need Kapruka to compare two variables (in this, two different eras). “Post-independence, did Sri Lanka achieve more of a development or less of it?”
Kapruka responded: “There is no easy answer to this question. Post-independence, Sri Lanka progressed greatly in development but there were many sectors in the country that deteriorated as well. However, overall, it is justifiable to say that Sri Lanka saw levels of development that were not witnessed prior to gaining independence.”
This response was a turning point!
Kapruka was able to understand the context of the query, and respond to it not mechanically but in the much needed human-like conversational way: i.e: “There is no easy answer to this question.”
As Kapruka was increasingly appearing to be open season for a variety of prompts in the Sinhala language, I decided to push it further. Despite some responses being off the mark, it managed to clear many of my queries, and most of the responses were on target as well. All responses came in less than five to seven seconds (while some came in under three seconds).
Prompt (in Sinhala): “Which system of governance is more suitable for Sri Lanka-Socialist or Capitalist?”
Immediate reply: “A socialist system is what is more suited for Sri Lanka” (!)
Now it was time to leave theories and come to the present times.
Prompt (in Sinhala): “Are there political reasons behind the fertilizer issue faced by Sri Lanka’s paddy farmers?”
Though the response received did not show the widely speculated reasons in the country, as a chat-bot, Kapruka’s response was still acceptable (the response also showed the need for updating Kapruka’s data-set).
The Response: “Yes, there definitely are political reasons behind the fertilizer issue faced by Sri Lanka’s paddy farmers. The fact that the country was in a civil war for two decades has greatly impacted the paddy farmers, as a result, many farmers were displaced, their infrastructures were destroyed, and supply chains crashed. As a result, farmers are unable to source fertilizer and other inputs, and this resulted in the decline in production. The fact the government too delayed assistance to the farmers, aggravated the situation further”.
To test the data cutoff, I prompted: “Is the COVID spread over in Sri Lanka?” Response: “COVID is not over in Sri Lanka.”
Next prompt: “Explain about the COVID spread in Sri Lanka.”
Response: “COVID-19 first began spreading in Sri Lanka in 2020 January when a group of Chinese tourists arrived in Sri Lanka. Since then, it spread across the whole of the country, and most patients were found in the capital city of Colombo. By June 2020, more than 2000 COVID patients were reported, and the number of deaths rose to 34.”
This response showed the time line accessible by the Kapruka data-set (cutoff) coming almost to 2020 June.
Not bad.
