NSTC to Roll Out ChatGPT Taiwan Version ChatGPT掀話題 國科會擬推出台灣版本
Taiwan is to develop its own version of ChatGPT by the end of the year.
In November 2022, OpenAI unveiled ChatGPT. Minister of National Science and Technology Council Wu Tsung-tsong admitted that he had used it many times and noted that OpenAI technology has advanced way too fast and shocked everyone! Confronted with the global AI arms race, Taiwan must develop its own ChatGPT.
Wu Tsung-tsong, NSTC Minister: “We are in the process of developing a generative pre-trained transformer, to foster domestically.”
Chang Liao Wan-chien, Legislator (DPP): “OK, has this plan started? When will it complete? ”
Wu Tsung-tsong, NSTC Minister: “Yes, it has. I hope by year-end (this year).”
It is estimated that at year end, Taiwan can roll out the prototype of a reliable ChatGPT. However, Microsoft spent US$10 billion while the National Science and Technology Council only has a US$4.3 billion budget; ChatGPT uses 10,000 GPU while Taiwan's Supercomputer Taiwania 2 only uses 2,048 GPU.
Huang Kuo-shu, Legislator (IND): “Not sure if our processing capacity and production capacity can successfully withstand Taiwan ChatGPT. ”
Wu Tsung-tsong, NSTC Minister: “So perhaps we should focus on one application and build on this foundation, then there should be an opportunity.”
Cheng Cheng-chien, Legislator (KMT): “What is our target? I hope this is not just a tool for this government to self-promote.”
Wu Tsung-tsong, NSTC Minister: “It shouldn't be.”
Wu said that they are still in the process of determining how much budget, software, and hardware are needed, but they can utilize cross-departmental budgets such as the science and technology development fund, combining public and private sectors to align globally. At the initial stage, it plans to focus application on a specific industry but will allow each and every industry to use it in the future.
ChatGPT去年底上路後,國科會主委吳政忠自己玩了好幾次,他說OpenAI進步太快,大家都嚇一跳。面對全球AI的軍備競賽,台灣必須要有自己的ChatGPT。
民進黨立委張廖萬堅問及,國內發展生成對話的引擎計畫,預計何時能完成?吳政忠表示,計畫已啟動,希望在今年年底完成。
國科會預計年底推出共用,可信賴的生成式對話引擎模型,但微軟砸了100億美金,國科會今年只有1327億預算。ChatGPT使用一萬片的GPU運算加速器,台灣的超級電腦「台灣杉二號」使用2048片GPU。
無黨籍立委黃國書提問:「我們的運算的能力、量能,足不足夠撐起台灣的ChatGPT?」
吳政忠回應:「我們如果先聚焦在某一個應用,然後把這個基礎先建立起來,應該是有機會的。」
國民黨立委鄭正鈐則質疑:「我們的目標在哪裡,因為我不希望只是變成政府大內宣的工具。」吳政忠則表示,應該不會。
吳政忠表示,需要多少預算和軟硬體還在盤點,但可動用跨部會署科發基金預算,而公私合力,再與國際介接,初期專注某個領域應用,未來讓各行各業都用得到。