Research Journey

Notes, findings, and honest reflections from my DBAI at PolyU Hong Kong.

Research
People trust AI differently when losses are on the table
A story-first look at why people resist AI assistance more when decisions feel like avoiding losses instead of chasing gains.
Research
The ABCs of who actually benefits from working with AI
A story-first look at why AI does not help everyone equally, and why the people who benefit most may not be the ones we expect.
Research
Trusting an AI teammate depends on who else trusts it
A story-first look at why trust in AI teammates is not formed alone. People watch how other humans treat the AI too.
Research
From AGI to ASI, the scary question is what happens after the finish line
A story-first look at why reaching AGI is not the end of the AI story. It may be the start of a much harder one.
Research
Customers react differently when they know a human is behind the AI
A story-first look at hybrid service agents and why disclosing human involvement can change how customers treat AI-powered support.
Research
Whether people follow algorithmic advice depends on incentives and framing
A story-first look at why people do not respond to algorithmic advice in a vacuum. Incentives and framing shape whether advice is used.
Research
Bot gender matters most when the service moment is tense
A story-first look at why bot gender can shape customer reactions in high-tension service contexts.
Research
AI trust is built through networks, not one-on-one magic
A story-first look at how domain experts help build trust in AI-based counterparts through triadic relationships.
Research
Managing human-AI teams can quietly bend leaders' ethics
A story-first look at the dark side of managing human-AI collaboration, where leaders may become more morally flexible than they realise.
Research
AI can boost employee creativity, but only under the right conditions
A story-first look at when and how AI augments employee creativity at work.
Research
Middle managers are becoming translators for AI service teams
A story-first look at how AI changes the work characteristics of middle managers leading AI service teams.
Research
People distrust AI managers because they do not feel benevolent
A story-first look at why employees may trust AI management less than human management, even when the AI seems competent.
Research
AI-augmented leadership works when leaders still do the human part
A story-first look at how leaders can work with AI to improve team performance without outsourcing leadership itself.
Research
Rideshare drivers show algorithm aversion is not irrational stubbornness
A story-first look at algorithm aversion among ridesharing drivers and why ignoring the algorithm can sometimes make sense.
Research
One plus one can be bigger than two when humans and machines share information
A story-first look at why human-machine collaboration depends on information flow, not just model accuracy.
Research
AI empathy can be good, but people still care who they think is listening
A story-first look at why people value empathy differently when they believe it comes from a human versus AI.
Research
Generative AI boosts creativity most for people who know how to think about thinking
A story-first look at why generative AI affects creativity differently depending on employees' metacognitive strategies.
Research
AI helps experienced workers differently from senior workers
A story-first look at why task experience and seniority do not produce the same benefits when workers team with AI.
Research
Perfume creation shows why human-AI collaboration needs translation
A story-first look at cognitive asymmetry in human-AI collaboration through the surprisingly useful world of perfume creation.
Research
AI chat can engage customers, but the handoff still matters
A story-first look at AI in online customer chats and why engagement depends on more than fast automated replies.
Research
ChatGPT made professional writing faster — and slightly more equal
A story-first look at what happens when professionals use ChatGPT for everyday writing work: faster drafts, better output, and a surprising reduction in performance gaps.
Research
In creative work, AI helps most when it critiques instead of performs
A story-first look at why the same LLM can help or hurt creative work depending on whether it acts like a ghostwriter or a sounding board.
Research
The future of work is not automation or augmentation — it is knowing which task gets which
A story-first look at why AI should automate some tasks, assist humans on others, and leave the hardest work to people who still know what they are doing.
Research
Social bots can make people engage — but not always in the way platforms hope
A story-first look at social bots on microblogging platforms: they can increase engagement, but may also train users to post for the bot instead of for each other.
Research
ChatGPT is becoming a shopping channel, but not yet a shopping monster
A story-first look at ChatGPT referrals to e-commerce sites: LLM traffic is real, growing, and useful for complex products, but it is still a niche channel.
Research
The best human-algorithm teams know when to disagree
A story-first look at loan evaluation experiments showing that human-algorithm collaboration works best when humans disagree with algorithms at the right time.
Research
Ghostwriter or sounding board — the collaboration mode matters more than the AI itself
The thing I like about this paper is that it complicates the easy story. It is not simply “AI good” or “AI bad.” It is more like: AI changes the work, and then the real...
Research
When AI makes up financial numbers — the hallucination problem in accounting applications
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
When AI handles the routine work, analysts shift toward the work that actually matters
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
What happens when AI writes accounting research — and when it hallucinates results
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
AI is already writing parts of corporate financial disclosures — and we can detect it
The part that bothers me is how calm the language sounds. It makes the risk feel distant, when the real problem is already sitting inside today's systems and decisions.
Research
Analysts are optimistic — and machine learning can now quantify exactly how much
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath...
Research
Do investors judge analysts by how they look? The uncomfortable evidence says partly yes.
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
The obfuscation hypothesis — why bad news comes wrapped in difficult prose
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
Detecting CEO depression through voice — and why it predicts corporate outcomes
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
Measuring CEO empathy with AI — and what it predicts about firm outcomes
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
The CEO's face during an interview tells analysts something — even if they can't explain why
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
Italy banned ChatGPT. Here's what happened to financial analysts.
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
When ChatGPT goes down, trading volume drops — a natural experiment in AI dependence
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
ChatGPT is bullish on Chinese stocks. DeepSeek isn't. Both are wrong.
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
When companies talk about AI — signals or just noise?
The part that bothers me is how calm the language sounds. It makes the risk feel distant, when the real problem is already sitting inside today's systems and decisions.
Research
Whose view of corporate culture is right — employees, analysts, or managers?
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
Measuring corporate culture at scale — and finding it predicts more than you'd expect
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
How companies talked about ChatGPT in the months after it launched — and what it signals
The thing I like about this paper is that it complicates the easy story. It is not simply “AI good” or “AI bad.” It is more like: AI changes the work, and then the real...
Research
The case for detailed financial data — why aggregated numbers leave prediction on the table
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
FinBERT and the problem with general-purpose NLP in financial text
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath...
Research
What executives say that ChatGPT wouldn't — a new way to measure information in earnings calls
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
What do investors actually ask AI? Evidence from 1.7 million stock queries
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath...
Research
The look-ahead bias problem — why AI financial predictions may be more impressive than they are
The part that bothers me is how calm the language sounds. It makes the risk feel distant, when the real problem is already sitting inside today's systems and decisions.
Research
Why "liability" isn't a negative word — the problem with generic sentiment dictionaries in finance
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath...
Research
Man + Machine beats Man vs Machine — the collaborative model for stock analysis
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
The misleading chart problem — how visual disclosure shapes and distorts investor understanding
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
Using machine learning to measure what accounting has always struggled to quantify
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
Machine learning is most useful for earnings forecasts precisely where human analysts are worst
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath...
Research
Machine learning as a noise filter — what non-linear accounting relationships reveal
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
The number and its story — why context transforms what accounting data means
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath...
Research
A picture is worth a thousand words — measuring investor sentiment from news photos
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
Predicting fraud before it happens — machine learning for material misstatement forecasting
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
Shipping containers from orbit — how satellite data predicts stock returns in 27 countries
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
Satellites over shopping centres — what alternative data does to information asymmetry
This is one of those risks that sounds abstract until you imagine explaining it after the fact. Then it suddenly becomes very concrete, very expensive, and very difficult...
Research
AI articles on Seeking Alpha — more coverage, but less insight per article
At first this sounds like a technical finance question. Then you look closer and realise it is also a question about attention, incentives, and whether people are using...
Research
How you say it matters — vocal quality, bad news, and market reactions
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath...
Research
Companies are quietly rewriting their disclosures for AI readers
The part that bothers me is how calm the language sounds. It makes the risk feel distant, when the real problem is already sitting inside today's systems and decisions.
Research
The missing layer — why AI agents need an operating system
I have learned to be suspicious of agent demos that look too smooth. The real question is not whether an agent can do one impressive thing once. It is whether the system...
Research
The deployment dilemma — choosing between speed and intelligence in AI agents
AI agent research always sounds clean in diagrams. Then you remember real agents have to plan, coordinate, fail, recover, and not confidently walk into walls while calling...
Research
Can you distill multi-agent intelligence into a single model? AgentArk says yes.
I have learned to be suspicious of agent demos that look too smooth. The real question is not whether an agent can do one impressive thing once. It is whether the system...
Research
We finally have a rigorous definition of AGI. Here's what it reveals.
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
GPT-4.5 was judged more human than actual humans. And most researchers still said AGI wasn't close.
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...
Research
The coordination curse — why AI teams get worse, not better
AI agent research always sounds clean in diagrams. Then you remember real agents have to plan, coordinate, fail, recover, and not confidently walk into walls while calling...
Research
The bottleneck theory of AI — why every breakthrough follows the same pattern
This is one of those studies where the headline is only the doorway. The useful part is what happens after you walk through it and realise the practical question is...
Research
A complete history of AI — from logic machines to the GPT revolution
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
A 397 billion parameter model on a consumer laptop — the memory wall has been broken
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...
Research
Two failure modes that break every AI agent — and how to fix them
AI agent research always sounds clean in diagrams. Then you remember real agents have to plan, coordinate, fail, recover, and not confidently walk into walls while calling...
Research
Anthropic built their most powerful AI — then decided not to release it
The part that bothers me is how calm the language sounds. It makes the risk feel distant, when the real problem is already sitting inside today's systems and decisions.
Research
Context engineering — the skill that actually makes AI work
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
I thought more AI agents meant better results. The research says otherwise.
I have learned to be suspicious of agent demos that look too smooth. The real question is not whether an agent can do one impressive thing once. It is whether the system...
Research
DeepMind's AGI bet — what it means to go all-in on general intelligence
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
The architecture of 1-million-token AI — a DeepSeek V4 deep dive
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...
Research
DeepSeek V4 — 1 million tokens at 10% of the cost
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
March 7, 2026 — the day a virtual fly changed the conversation
I enjoy these papers because they pull AI out of the browser window. They remind me that intelligence has always been physical, adaptive, and annoyingly hard to reduce to...
Research
Zero-shot wetware — what happens when you boot a biological brain in silicon
This is the kind of AI research that makes the field feel bigger than chatbots. It starts with brains, bodies, movement, or biology, and suddenly the question becomes:...
Research
The first complete brain simulation — 125,000 neurons, 50 million synapses
I enjoy these papers because they pull AI out of the browser window. They remind me that intelligence has always been physical, adaptive, and annoyingly hard to reduce to...
Research
DeepSeek's Engram — the architecture that changes how LLMs remember
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
The Engram Paradigm — why current LLMs waste their own intelligence
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...
Research
Google's prompting framework — from a single instruction to autonomous agents
AI agent research always sounds clean in diagrams. Then you remember real agents have to plan, coordinate, fail, recover, and not confidently walk into walls while calling...
Research
Are we at the intelligence explosion? What the inflection point actually means
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...
Research
Intelligence is not a volume knob — a 10-tier framework for thinking about AI
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
Apple's breakthrough — running large language models 20x faster from flash memory
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...
Research
Nested Learning — a new theory of how AI actually learns
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
What a fruit fly's brain taught researchers about building better AI
I enjoy these papers because they pull AI out of the browser window. They remind me that intelligence has always been physical, adaptive, and annoyingly hard to reduce to...
Research
OpenClaw — the autonomous AI operating system that goes beyond chatbots
AI agent research always sounds clean in diagrams. Then you remember real agents have to plan, coordinate, fail, recover, and not confidently walk into walls while calling...
Research
Why the future of AI compute might be in space
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...
Research
Embodied AI's GPT-3 moment is coming. Open source will be there first.
This is the kind of AI research that makes the field feel bigger than chatbots. It starts with brains, bodies, movement, or biology, and suddenly the question becomes:...
Research
Strategic chaos — how adversarial actors could destabilise AI agent ecosystems
The part that bothers me is how calm the language sounds. It makes the risk feel distant, when the real problem is already sitting inside today's systems and decisions.
Research
System 2 AI — we've moved from prediction to deliberation
Technical AI papers can look intimidating from the outside. Lots of architecture names, benchmark tables, and acronyms behaving like they pay rent. But the useful question...
Research
Why LLMs are stuck — the case for world models
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward,...