[17/06, 11:19 pm] ☸️ Danny 心: https://vt.tiktok.com/ZSQsxPLmY/
@Babe:Not really.
1. Eg. Government, big private enterprise use AI models, LLM and transformer engine but used internal dataset and accessible only by our internal staff.
At no time, the data set will be leaked to the public.
2. The AI architecture uses air-gapped network and the AI transformer engine and LLM run on internal servers and GPUs, internal clouds and internal datasets.
Permission are granted to internal staff based on strict access control with local content filtering on internal data access.
3. This is like using Microsoft OS, install in servers develop by US companies but network with air-gap, firewall and protected within an intranet not accessible by public and reserve for use by internal staff only.
So how does AI be controlled by few powerful overseas companies?
Our AI staff control, manage and maintain the entire AI models, architecture and infrastructure - all within our control like how we manage Microsoft OS, servers and network bought from US companies.
@Babe:Eg. DBS full AI solution using Microsoft Co-Pilot.
But the entire dataset that DBS access are DBS internal datasets.
So in which instance DBS future is held by Microsoft?
DBS has full control of its own Microsoft AI Co-Pilot, its Microsoft Azure cloud and its Microsoft OS.
Microsoft has no visibility into DBS AI architecture.
@Babe:1. DBS AI developers need to develop Agentic AI apps to access the internal datasets.
Microsoft has 0 visibility into this Agentic AI apps develop by DBS AI developers.
2. DBS AI developers are merely riding on the Microsoft AI Co-Pilot LLM.
Just like using Microsoft Azure Cloud and Microsoft OS
3. So in what ways DBS future is control by Microsoft?
Can you explain.
@Babe:1. If you are unable to understand complex tech, then let us look at simple analogy.
2. Eg. you buy a car - it could be China made, or Japan made, or SKorea made or German made.
3. Once you buy the car, the car is yours. It is your say how you want to use the car - the car is completely own by you, maintain by you and how and when to use it.
4. You only need to send to the factory for maintenance and spare parts.
5. AI is the same.
6. Singapore don't make car, but are we control by car manufacturers?
@Babe:"Southeast Asia's digital economy set to top US$300b by end-2025; Singapore attracts most AI funding: Report".
500 AI startups setup shops in Singapore AI hub.
https://www.channelnewsasia.com/singapore/southeast-asia-digital-economy-artificial-intelligence-investment-5459761#:~:text=Southeast%20Asia%27s%20digital%20economy%20set%20to%20top%20US%24300b%20by%20end%2D2025%3B%20Singapore%20attracts%20most%20AI%20funding%3A%20Report
@Babe:So indeed, Singapore AI hub is indeed rosy posy.
No bullshit.
@Babe:Hence what you say in this video is completely wrong.
It mislead the people who are viewing your video.
@Babe:Anyhow say.
Anyhow derive the wrong conclusion.
Anyhow mislead people with your limited technical knowledge.
Anyhow the blind leading the blind.
[17/06, 11:20 pm] ☸️ Danny 心:
@Babe:Quite obviously you all don't understand how AI works and how we use AI.
====
[19/06, 12:08 am] ☸️ Danny 心: https://vt.tiktok.com/ZSQnVa57E/
Stanford graduates booing Google CEO graduation speech on AI.
[19/06, 4:18 am] ☸️ Danny 心:
@Babe:1. As far as I know, AI professionals are highly sought after by business and fetching very high salary - way above median salary.
2. Non AI graduates that uses AI pervasively in their jobs - also gainfully employed.
3. So AI is not making workers worst off.
4. Only those workers who refuse to make use of AI in their work are facing the pressure.
@Babe:To summarise:-
1. AI don't make graduates worse off.
2. In fact, new job opportunities are opening up for graduates and workers like what the digital economy and e-commerce are opening up for workers.
3. The key is, do our graduates be able to leverage on AI like we leverage on digital economy, e-commerce, mobile apps in their jobs.
If we do, we thrive else we lose out.
@Babe:1. Previously, if our graduates and workers don't get trained in using internet, mobile apps, computer, smartphone - they can't secure a job.
2. Likewise, if now graduates and workers don't get themselves trained in AI, like how you get trained in digital economy, you won't get your jobs.
Simple analogy.
====
@Babe:The Global Human-to-Robot Ratios (Robot Density).
The International Federation of Robotics (IFR) tracks this automation metric as "robot density" (the number of operational robots per 10,000 manufacturing employees).
Rank Country Robots per 10,000 Workers
1. South Korea 1,220
2. Singapore 818
3. Germany 449
4. Japan 446
5. China 166 (National Statistic Basis) / Up to 470 in manufacturing hubs
@Babe:Singapore rank 2nd in density in the World in industrial robot adoption.
====
@hongchuan:Model ultimately still comes from that few companies. If they decide that the rest of the world can only access models 3-4 generation behind because of national security, what happens then?
@Babe:1. AI is similar to operating system Microsoft, open source Linux, Apple iOS, Google Android.
OS only control by a few US companies and open source.
2. Social media control by Instagram, YouTube, Tik Tok, xiaohongshu, etc - a few US and China companies.
Why no concern?
3. Do you know that Singapore have our own AI R&D researchers or AI creators that can create Al GPT (Generative Pre-Trained Transformer) and LLM (multi-languages LLM trained on many languages datasets for ASEAN countries)?
4. Singapore also got AI R&D researchers and creators working in Singapore top R&D labs creating AI models for government bodies.
5. Singapore also develop many humanoid robots for our own used.
6. Singapore is the 2nd global highest industrial robot adoption just behind South Korea.
Ahead of China, Germany and Japan.
So when are Singapore control by a few overseas companies?
@Babe:The GPT engine and LLM multi-languages for ASEAN that are produced by Singapore AI researchers and AI creators are called SEA-LION.
@Babe:1. Do you know that our 2 Universities are Ivy League grade - global top 8/10 NUS and top 12?
2. These Universities are producing AI R&D researchers and AI creators that can produce GPT (Generative Pre-Trained Transformer), LLM (Large Language Model) from scratch - such as those produced by ChatGpt, Google Gemini, Meta Llama etc.
3. They also able to produce DeepSeek like AI model through Knowledge Distillation.
4. Among the knowledge and subjects that they learn and competence in, including final year projects (that need to produce live GPT and LLM) in undergraduate and postgraduate research or coursework studies are:-
Subjects need to be learned to produce AI models like gpt, LLM, knowledge distillation - Neural networks, Generative Adversarial Network, machine learning, supervised learning, unsupervised learning, deep learning, reinforcement learning, semantics analysis, python programming, computer vision, multi - modal, NLP (Natural Language Programming) etc. etc. etc.
Also need to learn complex math like linear algebra, optimisation, statistics, probability, differentiation, integration etc etc....
5. And many Singapore AI R&D science lab are staff with scholars, graduates, postgraduates from Ivy League in addition to our homegrown produce AI R&D researchers and AI creators.
Who say our AI future are control by a few overseas companies only ????
@Babe:All these are Singapore built humanoid robots.
@Ranch sauce:you don't talkok, the models are made by microsoft. you have an incomplete understanding of how llms are made
@Babe:Can Singapore r&d researchers build LLM from scratch?
Yes, Singaporean R&D researchers possess the talent, infrastructure, and institutional backing to build Large Language Models (LLMs) from scratch. Backed by multi-billion dollar national AI investments, local teams have already proven this capability with sovereign models tailored to Southeast Asia. [1, 2, 3]
Local research institutions and tech consortiums are already leading the development of regional and multilingual models through several distinct advantages:
Demonstrated Capabilities: Researchers at AI Singapore (AISG), in collaboration with the Agency for Science, Technology and Research (A*STAR), have already built open-source foundational models from the ground up, such as Sea-Lion. This model is designed specifically to capture the linguistic nuances and cultural contexts of 11 Southeast Asian languages. [1, 2, 3, 4]
World-Class Infrastructure: Singapore supports these endeavors with massive computational power. Researchers leverage high-performance computing resources via the National Supercomputing Centre (NSCC) to train models without relying entirely on overseas cloud providers. [1]
National Multimodal LLM Programme (NMLP): The Singapore government has committed substantial funding (part of a S$1+ billion National AI R&D Plan) to advance fundamental research, train local AI engineers, and build base models grounded in regional context. [1, 2, 3]
While Singaporean researchers have the expertise to build base models from scratch, the local ecosystem primarily focuses on training highly efficient, specialized models (e.g., in the 30B to 50B parameter range) rather than attempting to compete directly with massive, trillion-parameter US or Chinese models on sheer scale. [1, 2]
If you want, let me know:
Are you looking to build your own specialized LLM for enterprise, healthcare, or finance?
Would you like to know more about accessing national compute resources or the Sea-Lion open-source datasets?
National Multimodal LLM Programme - Singapore - IMDA
25 May 2026 — The NMLP will: Build skilled AI talent in Singapore with funding support and access to high-end computing for loc
@Babe:you ignorant or I ignorant?
@Ranch sauce:this is not the situation currently, we lack the capacity to distill models like china, and we are more exposed to the risk of being cutoff from accessing the models in the first place due to trade agreements etc
all we are doing rn is putting money into MNCs pockets
@Babe:I think you very ignorant when you say we lack knowledge to do knowledge distillation - that produce DeepSeek.
@Babe:Can Singapore ai r&d researchers do knowledge distillation like DeepSeek?
+7
Yes, Singaporean AI researchers absolutely have the technical capability, infrastructure, and talent to do knowledge distillation at DeepSeek's scale. In fact, institutions like the National University of Singapore (NUS), Nanyang Technological University (NTU), and AI Singapore (AISG) are already pioneering advanced AI research. [1, 2, 3, 4, 5]
Here is a breakdown of why Singapore's R&D ecosystem is fully equipped for this:
World-Class Infrastructure: Singapore has top-tier compute resources. Through frameworks like the National Supercomputing Centre (NSCC), researchers have access to massive computing clusters. Furthermore, Singapore remains categorized under U.S. Tier 2 frameworks, allowing local facilities to legally procure high-end GPUs (like NVIDIA H100s) to train and run these advanced models. [1]
Advanced Research Capability: Model distillation relies on advanced mathematical and machine learning techniques. Singapore's researchers regularly publish at top-tier conferences (like NeurIPS and CVPR), meaning they possess the foundational AI and math skills needed to implement complex training pipelines. [1, 2, 3]
National Initiatives: Entities like AI Singapore (AISG) actively drive R&D in large language models, foundational AI, and model efficiency (e.g., the SEA-LION model series), focusing on distilling high-performing models for specialized, low-resource use cases.
The Key Difference:
While Singapore has the talent and compute to execute the technical aspects of model distillation, the controversy surrounding DeepSeek's process stems from how the "teacher" data is acquired. DeepSeek's massive success largely hinged on gathering highly refined synthetic datasets from large models via APIs, which has sparked legal and intellectual property disputes with Western frontier labs (like OpenAI) over Terms of Service and data scraping. Singaporean researchers, operating under strict international IP and cybersecurity laws, would need to navigate these ethical and legal boundaries carefully when sourcing or training off third-party models. [1, 2, 3, 4]
@Babe:@Babe:Don't know don't act smart.
@Ranch sauce:i said capacity u teatowel
@Babe:How do you know we don't have?
How do you know we have top AI researchers working in top R&D lab in many projects that are not published?
They don't build it for commercial projects, but many classified projects.
If you don't know, don't feed false information.
====
Source:- Google AI, ChatGpt
@Babe:Singapore’s ranking in the top 3 globally for Artificial Intelligence is a well-documented consensus across major tech indices. The city-state consistently secures top positions due to its progressive government policies, massive tech investments, rapid enterprise adoption, and a highly skilled workforce. [1, 2, 3]
Key Rankings and Recognitions
Singapore's leading status is backed by multiple authoritative global reports: [1]
Top Global AI City: Ranked #1 globally in Counterpoint Research's Global AI Cities Index, recognizing its mature AI ecosystem, talent pool, and infrastructure. [1, 2]
Global AI Financial Hub: Placed #3 globally (behind New York and San Francisco) in the DBS Global AI Financial Hub Index. This highlights Singapore's strength in balancing reliable AI governance with deep institutional adoption. [1]
AI Diffusion and Talent: Ranked #2 globally on Microsoft's AI Diffusion Index. It is also placed first internationally in AI Maturity, according to skill-level and research trends reported by Coursera. [1, 2]
@Babe:@Babe:Singapore rank 3rd in Global AI after US and China.
====
@Bertbrain:Actually, I did abit of research. I realised Singapore is leading an ASEAN LLM called SEA-Lion & Sailor 2. Perhaps these models are still not known to most b2c market. But my take is, in future it will be localised+cloud LLMs. I strongly believe Sg govt is moving at the right direction for now.
@Babe:LLM internals, transformers, fine-tuning, model compression and knowledge distillation → available mainly in advanced modules, graduate courses, research projects, and PhD work.
NUS.
National University of Singapore has several AI programmes and executive courses covering:
✅ Transformer architecture
✅ GPT and Llama models
✅ Fine-tuning and customization
✅ RAG (Retrieval-Augmented Generation)
✅ LangChain and embeddings
✅ LLM deployment and production
✅ Deep learning and NLP foundations
Examples include:
Master of Computing (AI)
Master of Technology in AI Systems
Large Language Models Overview and Applications
Advanced Large Language Models Approaches
Large Language Models in Production
These courses explicitly mention GPT-4, Llama, embeddings, and LLM architectures. �
@Babe:NTU
Nanyang Technological University offers:
✅ GPT API application development
✅ Fine-tuning GPT models
✅ NLP fundamentals
✅ Deep learning
✅ AI systems and machine learning
✅ Research opportunities through the Turing AI Scholars Programme
@Babe:Knowledge Distillation appears inside:
Deep learning modules.
Natural language processing courses.
Advanced machine learning electives.
Graduate research projects.
MSc and PhD research.
Students working with faculty research groups may implement:
DistilBERT.
TinyBERT.
LoRA and QLoRA.
Quantization.
Model pruning.
Knowledge distillation for LLM compression.
@Babe:To develop something similar to ChatGPT:-
The ideal knowledge path is:
Mathematics
Linear algebra
Probability
Optimization
Machine learning
Deep learning
Transformers
NLP
LLM training
Tokenization
Attention
Pretraining
SFT
RLHF
Efficient training
LoRA
QLoRA
Quantization
Knowledge distillation
Distributed training
PyTorch
DeepSpeed
Colossal-AI
RAG systems
Inference optimization and deployment
@Babe:There are many league of AI users and professionals:-
1. Some people just use prompt in AI models such as ChatGpt, Google AI, Anthropic Claude Mythos, DeepSeek, Meta llama etc to generate output - and claim to be AI users.
2. Some people are trained in Microsoft Co-pilot to churn out Agent AI - and claim to be AI professionals.
3. Some people are trained to be AI developers by coding in python programming, machine learning, data analytics etc and call themselves AI professionals.
4. Some are real AI R&D researchers and AI creators with solid math background and AI technology expertise that can build AI models from scratch be it GPT transformer engine, LLM, knowledge distillation etc - that produce AI models the like of ChatGpt, Google Gemini, Meta llama, DeepSeek knowledge distillation etc.
And our very NUS and NTU are producing such AI R&D expertise.
@Babe:In addition, our top R&D AI labs are staff with scholars from Ivy League with AI graduates and post graduates who have deep knowledge in foundational AI research.
That's why Singapore is rank globally 3rd in AI development and adoption.
====
@Babe:Several ASEAN countries are adopting and collaborating with Singapore to develop and adapt the SEA-LION (Southeast Asian Languages in One Network) Large Language Model.
While the foundational model development is anchored by AI Singapore (AISG), other ASEAN nations and organizations are participating in the following ways:
Enterprise Adoption: Companies across the region are using SEA-LION as a base to build their own AI systems. For instance, Indonesia's GoTo Group (Tokopedia) uses it to save on the high costs of training models from scratch.
Joint Fine-Tuning: Research institutions like Thailand's Vidyasirimedhi Institute of Science and Technology (VISTEC) have collaborated with AISG to fine-tune regional variants, such as WangchanLion for Thai language contexts.
Data Partnerships: Singapore has partnered with regional research initiatives like SEACrowd to source high-quality, culturally nuanced data (e.g., Thai and Indonesian slang) from across the region.
SEA-LION currently supports 13 regional languages—including Malay, Indonesian, Thai, Vietnamese, Lao, and Burmese—making it an inclusive AI tool for Southeast Asian enterprises.
@Babe:A few ASEAN countries - business and organisations across ASEAN are riding on Singapore developed SEA-LION LLM to develop their business Agent AI in their native language.
====
@Babe:NUS and NTU consistently rank among the top universities globally for Computer Science and Artificial Intelligence.
Global Rankings Overview
National University of Singapore (NUS) and Nanyang Technological University (NTU) share top-tier spots alongside US and UK tech giants.
Both universities hold elite positions, consistently landing in the top 20 across multiple global metrics.
Detailed Breakdown by Subject
Computer Science
QS World University Rankings by Subject: NUS ranks 4th globally, while NTU follows closely at 6th.
Times Higher Education (THE) Subject Rankings: NUS ranks 13th globally and NTU ranks 16th.
Artificial Intelligence & Data Science
U.S. News Best Global Universities: NTU is ranked an impressive 2nd globally for Artificial Intelligence, with NUS ranked 9th.
ShanghaiRanking (GRAS): NTU ranks 1st globally in Artificial Intelligence, ahead of Tsinghua and MIT. NUS ranks 27th.
CSRankings: NTU ranks 8th and NUS ranks 11th globally for AI, highlighting their heavy research output and conference publications.
====
@Babe:1. So naturally, our top AI resources are focus on strengthening our national economic competitiveness and strategic industries, security, cyber security etc.
2. Then the AI national council identified 4 commercial industries - financial, advanced manufacturing, healthcare, connectivity.
3. ASEAN AI LLM is another.
4. Of course commercial are leave to AI big tech AI - why replicate this commercial efforts?
@Babe:So are we taken hostage by a few overseas AI Big Tech?
Definitely not !
====