Day 1 · R0 Opening & Morning Keynote
Opening remarks and three keynote talks under the 'Intelligent Transformation' track theme: AI and sovereign AI viewed through Taiwan's trade structure, the geopolitical 'perfect storm' of globalization, and seven years of Taiwan AI Academy talent development
At a glance
| Speaker | Talk | In one sentence |
|---|---|---|
| Tung Tzu-hsien (童子賢), Chairman, Pegatron Corporation | AI-Driven Revolution: Opportunities for Taiwan's Industries | Starting from Taiwan's import/export structure, discussing AI's impact on industry, energy, and geopolitics, and advocating for building a 'sovereign AI with defensive capability' |
| Huang Chih-fang (黃志芳), Chairman, Taiwan External Trade Development Council (TAITRA) | AI and the Changes of a Century | Globalization tearing the world apart, US-China confrontation, and a 'perfect storm' of geopolitics and macroeconomics, with AI as the variable adding fuel to the fire |
| Tsai Ming-shun (蔡明順), Dean of Academic Affairs, Taiwan AI Academy | Intelligent Leap: The Transformative Journey of Taiwan AI Academy | Reviewing seven years of Taiwan AI Academy's training achievements and progress on the 2024 nationwide AI literacy initiative |
3 of 3 talks
1AI-Driven Revolution: Opportunities for Taiwan's IndustriesTung Tzu-hsien (童子賢), Chairman, Pegatron Corporation
Taiwan's economy runs on an extremely high trade dependency built on import/export trade; amid the AI wave, Taiwan's hardware advantage remains, but it must build its own 'sovereign AI with defensive capability.'
Key points
- During the 2021–2022 pandemic peak, Taiwan's total imports and exports surged to US$907.5 billion, while Taiwan's full-year GDP for the same period was only US$755.1 billion — an import/export-to-GDP ratio of over 115%. By comparison, South Korea (2023) is about 75%, Japan about 30%, and the US only about 20%; among advanced industrial nations, only Germany is higher (imports/exports about US$3.6 trillion / GDP about US$4.5 trillion, about 70%) — highlighting Taiwan's high dependence on foreign trade, which is also a relative advantage.
- Reviewing Taiwan's industrial history: Tung Tzu-hsien began with his childhood in Hualien, when Taiwan earned foreign exchange in its early years by exporting raw materials such as sugarcane, pineapple, and bananas; in 1972, then-Taiwan Provincial Governor Hsieh Tung-min (謝東閔) promoted the 'living room as factory' initiative, driving light-industry and textile OEM export; from there it gradually evolved into today's high-tech export structure centered on semiconductors and ICT.
- He cited a personal memory: before Tsai Ing-wen (蔡英文) became a presidential candidate, she had exchanged views with him, at a time when Taiwan's per-capita GDP had hovered around US$20,000 for years; from Tsai's election at the end of 2015 to before she left office in 2023, per-capita GDP grew beyond expectations, with the fastest growth occurring during the pandemic (he also noted the New Taiwan dollar strengthening against the Japanese yen, making wagyu beef in Tokyo cheaper for Taiwanese tourists than in Taipei).
- He used historical analogies to illustrate that 'industries evolve, and the standards for measuring national power/wealth change': in the early Industrial Revolution, smoke-belching steam trains were symbols of wealth and strength, but in the 21st century high carbon emissions are instead backward, while electrified high-speed rail running at 400 km/h is now the symbol of progress; in the 19th century, steel, coal, and ships symbolized national power, but today an Aegis destroyer of under 10,000 tons can defeat an old-style warship of 30,000–40,000 tons displacement (such as the Bismarck), because new technology brings the precision-strike capability to 'see far and hit accurately' — tonnage no longer equals national power.
- He compared wealth in gold versus Bitcoin: the roughly 170,000 tons of gold humanity has accumulated over 4,000 years is worth about US$13.7 trillion at US$2,500/oz; but the US's annual GDP of about US$26 trillion and China's of about US$18 trillion both exceed the entire historical stock of gold — the conclusion being that developing industry creates far more wealth than hoarding gold. Bitcoin (capped supply of 21 million coins) peaked at about US$70,000 per coin this March, has recently pulled back to about US$64,000, with total market cap still over US$1.4 trillion.
- He cited the BeiDou satellite system as an example of how critical infrastructure determines national power: after witnessing the US military's GPS-enabled precision strikes in the Gulf War, China spent 20 years from 2000 building out the BeiDou satellite system; the decision to open GPS for civilian use was made by US President Clinton the year before he left office, giving rise to today's car navigation and Uber (positioning error as small as 2–3 meters).
- Taiwan's 2024 semiconductor output value has been revised upward to about US$167 billion (about NT$5.2 trillion), further raised from the NT$4.8 trillion estimated at the end of last year to about NT$5.3 trillion; the communications industry's output value this year is estimated at about NT$4.3 trillion (about US$135 billion).
- Electric vehicles are a new opportunity for Taiwan: the global gasoline-vehicle industry is worth US$2–2.5 trillion, and most advanced nations will stop producing gasoline vehicles by 2035; Taiwan has historically struggled to enter the automotive industry via precision machinery/materials, but in electric vehicles it can leverage its accumulated advantages in PCs, ICT, and semiconductors to participate deeply.
- He compared the AI startup wave to the internet era: fewer than 20 companies worldwide have the capability to develop foundation large language models (LLMs), but just as the internet permeated all electrified and digitized fields, AI applications will in the future also permeate and integrate into every industry, including civilian life and military fields.
- The three prerequisites for AI development (big data, algorithms, compute) and the technology timeline: Google published the Transformer paper in 2017; OpenAI released GPT-1 and Google released BERT in 2018; GPT-3 debuted in 2020 but drew a lukewarm response; only in December 2022, when GPT-3.5 was combined with chat functionality to launch ChatGPT, did it ignite a nationwide craze. The 'Generative' in GPT means users can get responses using natural-language instructions without writing code, and it can handle multimodal content such as text, images, and audio.
- He cited a Google-NASA collaboration: AI analyzing NASA's discarded space photos discovered a 'mini solar system' that human interpretation had missed (the specific source of this example could not be verified); he also mentioned that Google's medical AI model, when answering long-form medical questions, matched scientific consensus 92.6% of the time, close to clinicians' 92.9% — this refers to Google's 540B-parameter Med-PaLM model, published in Nature in July 2023; the same study also showed the control group Flan-PaLM matched scientific consensus only 61.9% of the time, and the rate of harmful answers was similar between Med-PaLM (5.9%) and clinicians (5.7%), while Flan-PaLM reached 29.7%.
- On sovereign AI: he argued that every country should be able to independently control its AI infrastructure and applications, because AI responses carry subjective values and cultural tendencies; he used LINE as an example of how 'sovereign social software' is hard to promote (it was attempted but failed due to lack of user network effects), contrasting this with the sovereignty risk of governments relying on LINE, whose databases are stored overseas (Japan/South Korea).
- He introduced Taiwan's TAIDE project: built on Meta's open-source Llama 3 model, it is a trustworthy AI dialogue engine (TAIDE, Trustworthy AI Dialogue Engine) tailored to Taiwan's cultural characteristics; he said fewer than 5 companies worldwide currently have the fundamental capability to control large language models (LLMs).
- Training costs and energy needs: a single training run of a GPT-4-scale large language model (LLM) costs about US$78 million, while Gemini, the Google/OpenAI competitor, costs about US$190 million per training run — this requires massive chip compute and electricity, which is why OpenAI founder Sam Altman has invested in nuclear power plants, Microsoft has also invested in nuclear power, and Bill Gates has invested in a next-generation fast neutron reactor in Wyoming, USA.
- The 'the strong get stronger' phenomenon in AI: unlike the hierarchy reversal of the PC/internet era 30 years ago (large computer makers like IBM and Wang Laboratories declined while Apple, Microsoft, and Intel rose), this wave of AI, due to extremely high compute and electricity thresholds, instead lets existing tech giants maintain their advantage; Musk's Optimus humanoid robot strategically does not build a large language model (he jokingly calls it 'a silent AI'), while Apple, lagging in building its own AI infrastructure, turned to partner with OpenAI — and because Apple controls the mobile distribution channel, it is OpenAI that has to pay Apple (analogous to Google Maps or Gmail also having to pay to be listed on Apple's phones).
- Taiwan's hardware advantage extends into the AI era: a single server rack can sell for about NT$300 million; in the future all phones and laptops will embed an NPU (Neural Processing Unit), estimated to bring an additional US$50 billion in output value within five years; tests show that laptops with an NPU convert speech to meeting minutes 60 times faster than those without.
- Semiconductors as a new strategic resource: citing that the OpenAI founding team, with an average age of just 29, still depends on the 5nm/3nm processes founded by TSMC's 92-year-old Morris Chang (張忠謀) to run GPT-4/Gemini-class models with over 1 trillion parameters (1TRA), he emphasized that advanced chips are now no less important than oil; AI software output value was about US$75 billion in 2024, projected to reach US$200 billion by 2028, an area where Taiwan is relatively weak.
- He looked ahead to how AI training will evolve: moving toward 'central training plus local fine-tuning on end devices' (which he called Federal Training), for example a phone could fine-tune personally based on a user's accent (such as Taiwanese-accented Mandarin), then send the averaged parameters back to the central model for continuous improvement.
Tech, products & figures
- TAIDE (Trustworthy AI Dialogue Engine)
- Built on Meta's open-source Llama 3 model, it is a trustworthy Traditional Chinese dialogue engine tailored to Taiwan's cultural characteristics; the National Science and Technology Council (NSTC) released an 8-billion-parameter version in April 2024. Technical details and applications are covered in more depth in other sessions (Huang Han-hsuan (黃瀚萱) on multimodal model development; Tsai Chung-han (蔡宗翰) on task design and EduTAIDE).
- GPT series / BERT / Gemini
- Two representative large language model (LLM) families driven by the Transformer paper published by Google in 2017.
- NPU (Neural Processing Unit)
- A neural processing unit that will become standard in phones and laptops, greatly accelerating on-device AI computation.
- BeiDou Satellite System / GPS
- Global navigation satellite systems that form the underlying infrastructure for precision strikes and civilian positioning (car navigation, ride-hailing services).
Notable quotes
We need to be able to control our own AI sovereignty…… You don't invade others, but you also don't want to accept being invaded by others.
What this tells everyone is: rather than scooping up all the gold in the world, four thousand years' worth of gold, you're better off properly developing your country's industries — your output in a single year can exceed all the gold humanity has ever accumulated in history.
The new semiconductor trend represented by advanced chips is now no less important than oil — it will become the world's new strategic resource.
Q&A
- This was a stand-alone talk with no Q&A segment.
Fact-check notes
- The training costs of about US$78 million for GPT-4 and about US$190–192 million for Gemini Ultra are consistent with the estimates published in Stanford HAI's 2025 AI Index Report (GPT-4 about US$78 million, Gemini Ultra 1.0 about US$192 million).Sources：Voronoi 報導Fortune 報導
- Regarding the Google-NASA collaboration that reportedly discovered a 'mini solar system,' a search could not find a corresponding specific source or paper; this is still marked as unverified, and may be an approximate citation rather than a precise finding.
- The 92.6% accuracy figure for Google's medical AI has been verified to correspond to Google's Med-PaLM study (540B parameters, based on PaLM), published in Nature in July 2023.Sources：gol.idv.tw
- The Taiwan AI Academy's official post-conference report confirms that Tung Tzu-hsien's points on the day — 'building sovereign AI with sufficient defensive capability' and 'Taiwan's hardware is strong, software is weak, but it still holds an advantage for the next decade' — match what was said on-site.Sources：aiacademy.tw/conf2024-news
- For the additional output value NPUs will bring within five years, another recorded version says '500 million' (currency unspecified), a hundredfold difference from this note's '$50 billion.' After comparison, '$50 billion' is adopted: the sentence claims that 'all future phones and laptops will embed an NPU,' a global-market-scale claim for which '500 million' is clearly too small, whereas '$50 billion' retains a complete sentence structure and currency unit and is more credible; '500 million' more likely resulted from an omission in the other recorded version. However, a WebSearch found no media report citing this figure, and neither version has third-party source corroboration — this is a reasonable judgment, not a confirmed verification.
2AI and the Changes of a CenturyHuang Chih-fang (黃志芳), Chairman, Taiwan External Trade Development Council (TAITRA)
Globalization's fracturing of the world has spawned a cultural class war and US-China confrontation, compounded by a fragile macroeconomy; AI will only make this 'perfect storm' more complex.
Key points
- Setting the tone at the opening: besides running Computex, the world's most important AI exhibition platform, TAITRA must also continuously observe global geopolitical and trade changes; he set out to analyze 'what forces made the post-2020 world unrecognizable within just three or four years.'
- The two faces of globalization: after the Soviet Union's collapse in 1991 and China's opening-up reform joining globalization, the world enjoyed about 30 years of low inflation, low interest rates — an 'age of peace and prosperity' — essentially driven by capitalism's cross-border free movement of factors of production, sharing the same logic as the 16th–17th century Dutch/British East India Companies' Age of Discovery colonial expansion; but at the same time, it redistributed global industrial division of labor according to comparative advantage, causing job losses in some countries (Taiwan's overseas production ratio for orders received stayed above 50% for a long time, only dropping below 50% in the past year or two).
- Winners and losers of globalization: those who can 'move' (entrepreneurs, capital, multinational corporations) can profit from globalization — for example, Indonesian migrant workers coming to work in Taiwan can earn more than 10 times their original wages; but 'immobile' groups such as blue-collar workers and farmers become the losers, as seen in European farmers driving tractors to occupy capital city streets in protest.
- Financial liberalization worsened the wealth gap: the US abandoned the gold standard in the 1970s, and Reagan and Thatcher deregulated finance in the 1980s–90s, spawning derivative financial products (the root of the 2008–09 financial crisis); US financial industry profits as a share of total corporate profits climbed from 10% in 1985 to 40% by 2005. He cited a RAND Corporation study: between 1975 and 2018, US$50 trillion in wealth shifted in the US from the bottom 90% of the population to the wealthiest 1%.
- 'Hillbilly Elegy' and the culture-class war: citing J.D. Vance's (current running mate to Trump) 2016 bestseller Hillbilly Elegy (adapted into a Netflix film), which describes how globalization plunged blue-collar workers in the Appalachian Rust Belt into poverty, domestic violence, and drug abuse; this group sent Trump into the White House in November 2016, and MAGA (Make America Great Again) calls for restoring American blue-collar workers to a 1960s–70s-style middle-class life where they could afford rent and raise children, not simply military or political strength. He also asked whether Taiwan has its own 'Hillbilly Elegy' — arguing that the anxiety of Taiwan's younger generation about the future is exactly why political space exists for Taiwan's 'third force.'
- China's rise and US-China confrontation: between 1990 and 2020, China's GDP grew 35-fold while the US grew only 2.4-fold over the same period, with China's GDP once approaching more than 70% of the US's; he stressed that China becoming the world's factory was actually caused by the US's own policies (brand companies investing in China) — 'the US unwittingly created its own biggest competitor.' The US's recent tariffs, industrial policy, and tech blockades are essentially aimed at 'ending globalization,' to prevent China's economy from continuing to grow fast enough to overtake the US.
- The 'Triffin structure': the US exports dollars → East Asian countries (especially China) export goods to earn dollars → they use those dollars to buy US Treasury bonds → funds flow back into the US, and because the dollar as a reserve currency is no longer pegged to any physical asset, there is no mathematical limit; but the US wants to revive manufacturing, and China worries about its assets being sanctioned by the US (drawing lessons from Russia's foreign-exchange sanctions), so both sides want to change this structure — yet Taiwan's economic growth over the past 50 years has been built precisely on this structure.
- Geopolitical flashpoints everywhere: the Russia-Ukraine war has thoroughly changed Europe's security architecture, producing a lose-lose-lose outcome for Europe/Ukraine/Russia; he cited an Egyptian friend's prediction after the outbreak of the Israel-Hamas war that 'the dynamics of the Middle East have changed for good and can never go back, Israel will definitely escalate,' noting that the past year has fully confirmed this; the South China Sea, North Korea, and the China-India border are all potential flashpoints; quoting Kissinger, he described the current situation as 'feeling like it's back to before World War I broke out in 1914' (small regional conflicts dragging major powers into direct war, lacking the past kind of safety valve where a 'mob boss steps in to broker a deal').
- The macroeconomy is standing on a cliff edge: a mere 0.15 percentage point rate hike by the Bank of Japan (from 0.1% to 0.25%) triggered global stock market circuit breakers and Taiwan's stock market's biggest single-day drop in nearly 30 years, showing the global financial system now has 'no resilience whatsoever'; he also pointed out that NVIDIA's P/E ratio has reached a 'somewhat exaggerated' high valuation, meaning that once the market is shaken, volatility will be amplified to an unprecedented degree.
- US debt risk: in 2023, the US government ran a fiscal deficit of US$1.7 trillion and a trade deficit of US$770 billion; US public debt has reached US$35 trillion, with interest payments accounting for 15% of the federal budget, even exceeding the defense budget; but only 40% of US Treasury debt is held overseas (including Taiwan's central bank), while 60% is held by domestic US financial institutions and large corporations, creating a 'distorted prosperity' phenomenon where interest payments flow back into the US economy and prop up its apparent prosperity. Reviewing dollar history: the 1944 Bretton Woods conference established the dollar's gold standard (US$35 per ounce of gold); in 1971 the US printed more money than its gold reserves and abandoned the gold standard; in 1973 the dollar shifted to being pegged to oil, becoming the 'petrodollar,' cementing its status as the international reserve currency.
- AI and geopolitics: quoting Kissinger that 'AI is more powerful than nuclear weapons in geopolitics'; citing 9/11 in 2001, when US forces went to their highest alert level, then-National Security Advisor Condoleezza Rice calmly called Russian President Putin to explain that the US alert was in response to the terrorist attack, not aimed at Russian military exercises, successfully avoiding a mutual misjudgment escalating into war; he asked: if such decisions were handed to AI, given instructions to 'ensure American interests come first and give the adversary no opportunity,' would AI still show such restraint? He also mentioned that AI's deepfakes and its powerful persuasive ability (studies have shown it surpasses humans) could, if used in asymmetric warfare, be enough to paralyze a country's democratic institutions.
- Another pair of contrasts in AI governance frameworks: he proposed two contrasting concept pairs — 'KYC AI vs. decentralized AI' and 'closed vs. open-source AI' — and pointed out that ASICs are the next frontier of AI training, with China developing rapidly in token ASICs (presumably specialized chips optimized for LLM token generation inference).
- AI civilization experiments and future visions: he cited the startup Altera's AI civilization system built in the game Minecraft as an example — about 1,000 AI identities living, deciding, and building their own governments and social rules independently within it, a sandbox preview of a 'world where AI and humans coexist'; he also speculated that future AI agents with long-term memory and personalized traits might be seen by human society as another kind of 'immigrant,' triggering feelings of rejection (analogous to current anti-immigrant sentiment), 'adding more fuel' to the culture war he described earlier.
- Advice for Taiwanese entrepreneurs: facing a once-in-a-century upheaval of supply chain restructuring, the retreat of globalization, yet companies being forced into 'global localization' (such as TSMC's model), businesses should think about 'where is your moat, and how deep is it' (quoting Warren Buffett), and he recommended that industry build professional geo-economics research capability, while keeping abreast of the latest developments as geopolitics, the macroeconomy, and new AI technology interact with one another.
Tech, products & figures
- Altera (Project Sid)
- A US startup that in 2024 ran about 1,000 autonomous AI agents in Minecraft to build a simulated civilization with its own economy, culture, religion, and government institutions; founded by Robert Yang (formerly an assistant professor of computational neuroscience at MIT), who closed an oversubscribed US$9 million seed round in May 2024.
- Hillbilly Elegy
- A 2016 book by J.D. Vance describing the plight of blue-collar workers in America's Rust Belt, adapted into a Netflix film.
- Triffin structure
- A structure describing the cycle of trade and US Treasury debt underpinning the dollar's role as the international reserve currency.
Notable quotes
This is a war of culture and class…… I just talked about so much of this culture war — when AI arrives, it will surely add more fuel to the culture war I just described.
Kissinger said AI is more powerful than nuclear weapons in geopolitics.
Where is your moat? How deep is your moat? I think this is worth every business thinking carefully about.
Q&A
- This was a stand-alone talk with no Q&A segment.
Fact-check notes
- RAND Corporation's finding that '$50 trillion in wealth shifted from the bottom 90% to the top 1% between 1975 and 2018' is a publicly cited statistic used by the speaker, and its direction is consistent with RAND Corporation's related wealth-inequality research published in 2020 (the original report's figures were not individually cross-checked, so readers are still advised to verify the original source).
- The Altera/Project Sid case has been confirmed via WebSearch to be a real startup project; its timing (launched in September 2024) is nearly concurrent with this conference (2024/09/27), meaning the speaker's citation was quite timely.
3Intelligent Leap: The Transformative Journey of Taiwan AI AcademyTsai Ming-shun (蔡明順), Dean of Academic Affairs, Taiwan AI Academy
Borrowing a five-forces framework to review seven years of Taiwan AI Academy's training achievements, and taking stock of concrete 2024 progress on nationwide AI literacy certification and government collaboration.
Key points
- Opening by borrowing Michael Porter's framework, he summarized the capabilities needed to develop AI into five elements — 'brainpower, compute power, electric power, financial power, and data power' — echoing the context laid out by Tung Tzu-hsien and Huang Chih-fang from historical and geopolitical angles.
- In the seven years since its founding, Taiwan AI Academy has accumulated more than 11,000 alumni and over 2,200 participating companies spanning 15 major industries, echoing this conference's theme of 'AI for Every Industry.'
- The AI talent gap keeps widening: per the latest report from June 2024, Taiwan's average monthly AI talent gap is about 29,000 people, up year by year from about 17,000 in 2017–2018; just from last year to this year it rose from 24,000 to 29,000 (an increase of 5,000 openings), with an estimated annual growth rate of about 7%. On the government side, plans are in place to address the talent gap through 2031, with a goal of training 200,000 people within four years.
- Citing a chart that recently went viral online comparing the capabilities of the latest large language models with human engineers: most large models have already surpassed the level of the 'average human engineer' and are continuing to approach the level of 'top engineers'; he used this to remind the audience that engineers whose ability falls below the AI average will struggle to find good jobs, and only those who surpass the latest large models will command high salaries.
- In 2024, the Academy is promoting a nationwide AI literacy and tiered certification program, divided into three dimensions — 'using AI, building AI, and managing AI'; since January this year it has held quarterly 'AI for Every Industry' themed workshops.
- Concrete results: in March, the 18-month 'Feng Huo Lun Project' (a pre-trained model for the central Taiwan machine tool cluster) was unveiled to positive reception at the Taiwan International Machine Tool Show; a board-level AI literacy consensus camp was launched in partnership with China Medical University; an online AI literacy course was launched with the CommonWealth Learning (天下 Learning) platform, taught by four instructors plus Tsai Ming-shun himself.
- Government cooperation: since Q3 2024, the Academy has been helping the Executive Yuan run AI consensus camps for senior civil servants, aiming to complete training for 732 senior officials at grade 12 or above across 31 ministries and agencies by the end of October; courses for ministers and deputy ministers have already been completed.
- The AI literacy certification exam formally opened on July 1, 2024, covering nine major domains; the pass rate so far is about 93% (240 people passed), with a cumulative 110 companies having organized employees to take the certification. In Q2, a hands-on AI certification class was also launched, integrating the former manager track and technical track into generative AI courses with an emphasis on hands-on practice.
- In partnership with Business Next (《數位時代》), the Academy published the 'AI Real-World Adoption Guide for Every Industry,' surveying more than 300 companies; the top three technologies companies most want to adopt, in order, are generative AI, deep learning, and data analytics.
- On international cooperation, he mentioned talks with Google and Meta, and specially welcomed the more than 30 Malaysian scholars attending that day; he also honored founding CEO Chen Sheng-wei (陳昇瑋) for laying the platform and resource foundation in the Academy's early days.
- Conference participation statistics: total registrations for this year's conference exceeded 1,060 people, with tech-background and non-tech-background attendees at roughly 53% and 47% respectively, close to an even split.
Tech, products & figures
- Feng Huo Lun Project
- An 18-month project by Taiwan AI Academy building a pre-trained model for the central Taiwan machine tool cluster, with results presented at the Taiwan International Machine Tool Show in March 2024.
- Tiered AI Literacy Certification
- A nine-domain certification exam offered by Taiwan AI Academy starting July 1, 2024.
Notable quotes
Although Taiwan AI Academy is a foundation with still fairly limited resources, I want to tell everyone: cooperation can create unlimited opportunity.
Q&A
- Immediately after this talk, the morning agenda was declared concluded, with topic sessions starting at 1pm in the afternoon; there was no Q&A segment.
Fact-check notes
- Chen Sheng-wei (陳昇瑋) was the founding CEO of Taiwan AI Academy (1976–2020, PhD in Electrical Engineering from National Taiwan University, founded Taiwan AI Academy in 2018, passed away in 2020 from a cerebral hemorrhage sustained in a sports accident, at the age of 44).Sources：維基百科台灣人工智慧學校官網追悼頁
- The large model name 'Zero One … Preview' mentioned by the speaker may correspond to 01.AI's (零一萬物) Yi series models, but the exact version could not be confirmed with complete certainty, so it is marked as not precisely verified.