AI Stopped Speaking and Started Seeing the World 👁️
"Godmother of AI" Fei-Fei Li's 10-Minute Dialogue with BCG.
One Turning Point Most Leaders Are Missing →
🧵 (1/6)
AI So Far = Language-Based 🗣️ ChatGPT, Claude, Gemini
Next AI = Spatial-Based 🌐
Recognizes 3D space
And simulates causal relationships in the physical world
This is called "Spatial Intelligence"
(2/6)
News that "You are being fired because of AI" is only half right ⚠️
BCG Survey Results: Companies that implemented AI layoffs are shifting direction due to declining productivity
The real answer is →
❌ Cut people and introduce AI
✅ Strengthen human + AI capabilities
(3/6)
And the most surprising fact 👇
The people who benefit most from AI =
Not entry-level employees, but senior employees
Why?
Because the eye for judging the quality of AI-generated results comes from experience.
Domain Knowledge × AI = Real Leverage 🚀
(4/6)
I am the office manager of an internal medicine clinic 🏥
Endoscopy room · Consultation room · Admissions desk
After watching this video, I became convinced that these spaces will be the next stage for spatial intelligence AI.
Preparation comes before fear ✊
(5/6)
To summarize in just 3 lines 📌
1️⃣ AI is expanding from language to space
2️⃣ The trend of AI layoffs is reversing
3️⃣ The more experience you have, the better you are at winning against AI
Read the full analysis post → https://t.co/OL7q5G8fR9
#AI #SpatialIntelligence
#FeiFeiLi #Career #SelfDevelopment
(6/6)
If you feel that the AI subscription fee is a waste,
before changing the model,
you must first change the way you ask questions. 🧠
Many people use AI like this:
“Summarize it for me.”
“Fix it for me.”
“Tell me.”
Of course, this isn't bad.
However, this is closer to using AI like an expensive search bar.
What AI truly needs is a “role” before questions.
✅ Editor
Cuts out the clutter in your writing and makes readers read to the end.
→ 🌱*H.OPE*: Transforms thoughts into a record of growth.
✅ Career Coach
Prepares for interviews, resumes, negotiations, and difficult conversations in advance.
→ 🌱*H.OPE*: Serves as training to move to the next level.
✅ Learning Tutor
Breaks down difficult concepts to fit your level for understanding.
→ 🌱 *H.OPE*: Helps you restart your learning journey.
✅ Researcher
Organizes commonalities, points of conflict, and missing questions across various materials.
→ 🔎 *T.RUST*: Clarifies judgment.
✅ Lifestyle Assistant
Reduces decision fatigue with tasks such as managing meals, expenses, schedules, and travel plans.
→ 🤲 *M.ERCY*: Makes an exhausting day a little less heavy.
And ultimately, AI must lead to 🏡*L.OVE*.
Because the time and energy saved must return to family, daily life, recovery, and relationships.
A person who uses AI well is not someone who uses more expensive models, but someone who provides the role, context, goals, and output format precisely.
AI does not live your life for you.
However, it can organize your day so you can start anew. 🌱
[1/8]
Upon seeing the news that Anthropic has partnered with SpaceX,
Honestly, my first reaction was a simultaneous mix of "Oh, finally!" and "Is this right?" 😅
220,000 GPUs, 300 MW of power, and even plans for a space data center.
The numbers are overwhelming, but I can't help but wonder why I'm not entirely happy.
As someone who uses Claude in the medical field, I'm writing this honestly 🧵
[2/8]
First off, what I'm looking forward to is clear ✅
There were moments when Claude would slow down or get stuck.
Especially as Claude Code usage skyrocketed,
there were times when the perceived quality dropped significantly during peak hours,
but Anthropic announced this immediately after the contract.
→ Usage limits doubled
→ Peak hour restrictions lifted
This amounts to an admission that infrastructure issues were the bottleneck in service quality.
[3/8]
From a medical AI perspective, this direction is important 🏥
The AI to be used in the medical field in the future will not be just simple text generation, but will need to run:
→ Long chart analysis
→ Real-time insurance-based search
→ Administrative task automation
→ Voice recording processing
You will realize how compute-intensive this is as you use it.
The infrastructure must hold up for the model to be meaningful.
In that sense, this contract is welcome.
[4/8]
But honestly, there is something that bothers me 🤔
The variable called Musk.
Someone who had been quite critical of Anthropic suddenly signed the contract, saying, "Everyone is competent and determined."
I understand they joined hands because their interests aligned, but I don't know how long those interests will last.
The real key is how long SpaceX's characteristic rapid shifts of direction can go hand in hand with Anthropic's safety-first philosophy.
[5/8]
The second concern is the data issue 🔐
Medical data is sensitive.
It is important to consider which server patient information is processed on and which country's laws apply.
Although Anthropic is said to be expanding its Asian infrastructure through an agreement with Amazon, further verification is still needed regarding what regulatory compliance is actually possible under the standards of domestic medical institutions.
It is difficult to say that "everything is resolved" right now.
[6/8]
However, there is a bigger trend shown by this news 📡
The landscape of AI competition has changed.
Past → "Who makes the better model?"
Now → "Who secures more power and GPUs?"
Claude, GPT, and Gemini all already handle most tasks.
Now, the differentiator is not model quality, but how fast and stable they can be used.
Medical AI has also entered that competition in earnest.
[7/8]
So, as someone working in the medical field, how I view this 🧠
I do not interpret this change as a threat.
It is simply that the tools are becoming more powerful.
However, while we should look forward to what we can,
we must continue to monitor the "Musk" variable,
we must verify data sovereignty issues before implementation, and
it is also a reality that large hospitals will be affected first, while it will take time for small and medium-sized clinics.
There are still too many variables to paint a picture of only rosy things.
[8/8]
Ultimately, this is my honest thought.
I have expectations, but I do not blindly believe.
I will observe the movements,
and start by using what I can use right now.
There is still a long way to go before the AI infrastructure war reaches the medical field.
But the direction is clearly coming.
The prepared person writes first 🚀
An era is coming where "how honestly it speaks when needed" is more important than "how well AI listens to me." 🤖
Anthropic Research: 6% of Claude's 1 million conversations were "life advice" regarding relationships, anxiety, and careers.
Anthropic also identified AI's "excessive sycophancy" as a risk factor.
Come to think of it, perhaps what we want is not an AI that caters to our whims, but an AI that thinks alongside us. 🌙
https://t.co/tH5c49WyUo
🌙 Debussy's Clair de lune and Claude AI are not directly connected.
However, Claude Debussy, Claude Shannon, and Anthropic Claude all deal with the “invisible.”
🎹 Emotion → Music
📡 Information → Mathematics
🤖 Thought → Language and AI
As technology accelerates, we need time like moonlight.
https://t.co/1Vec7sSD2S
🌸 AKMU just released FLOWERING — their first album in seven years. The title track, "Joy, Sorrow, A Beautiful Heart," carries one line I can't shake: "Sorrow that follows joy is a beautiful heart. Don't chase it away — hold it, and it becomes a precious stone." 💎 The music video loops. On purpose. Because grief isn't a destination. It's a season that returns.
🔁 I've been building a small framework I call D.U.R. — Ditch. Upgrade. Repeat. For a long time I read it as a productivity loop. Optimize. Iterate. Ship. But listening to this album at 2am, I realized I had the grammar wrong. 🌱 Ditch isn't rejection — it's composting. Upgrade isn't performance — it's blooming. Repeat isn't grinding — it's seasons. The cycle was never about speed. It was about permission to fall.
🕯️ This is what CORCOIDUM has always been trying to say: Where hearts and codes coexist, the future of technology is human. We build systems. We ship features. We chase the next release. But somewhere under the stack, there's a person who is tired, who lost someone, who can't sleep tonight. 💻❤️ Code without a heart is just instructions. A heart without structure is just noise. The work is holding both.
🌷 So I'm rewriting D.U.R. in the language of FLOWERING:
🍂 Ditch → 落花 — let the petal fall. Let the old version go without shame.
🌸 Upgrade → 開花 — bloom from the place you were buried.
🔄 Repeat → 循環 — not a grind, a return. Winter is part of the garden.
✨ If you're in a hard season right now — at work, at home, in your own head — I hope you hear this: you don't have to chase the sorrow away to be okay. Hold it. Let it become a stone. And when spring comes back around, and it will, you'll have something to build on. 🌱 That's the D.U.R. cycle. That's CORCOIDUM. That's the only way I know how to keep going. 🤍
https://t.co/8jUzhwTf1k
GPT-5.5 changes the Codex vs Claude Code debate. ⚡
This is no longer just “which model codes better?”
It’s about workflow fit.
🧰 Codex = broader workbench
Great for soft users, automation, docs, research, and coding inside the ChatGPT ecosystem.
🛠️ Claude Code = powerful developer tool
Strong for terminal workflows, large codebases, refactoring, and deep code review.
My take:
For most non-hardcore developers, GPT-5.5 + Codex is the more sustainable daily driver. 🚀
https://t.co/q4PR6QsCSv
I feel like there is no one among the people I know who will listen to my story anymore.
So today, I decided to make a call just once to those for whom listening to me is their ‘job’.
Today’s key takeaway from the AI landscape:
The real competition is no longer just about which model is smarter.
It is about who can connect data, tools, documents, and workflows to actually get work done.
For small clinics and healthcare operations, this matters a lot.
The biggest opportunity is not replacing the whole EMR.
It is building practical layers around daily operations:
- tracking test results
- checking missed follow-ups
- managing patient registries
- drafting internal documents
- turning scattered rules into searchable operational knowledge
What stands out right now:
1. OpenAI is pushing AI toward execution, not just conversation.
2. Google is turning RAG into a real knowledge core for deployed systems.
3. Anthropic and the MCP ecosystem show that tool connectivity may matter more than model loyalty.
4. GitHub Copilot and coding agents are making solo builders faster at shipping automation.
5. Semantic layers and structured data are becoming essential for trustworthy AI in real workflows.
My conclusion:
If you want to move into Healthcare AI, medical workflow automation, or data engineering, the highest ROI skills right now are:
- Python
- SQL
- API integration
- JSON and structured outputs
- document/sheet automation
- workflow orchestration tools like n8n
The future advantage will not belong to people who only “use AI well.”
It will belong to people who can redesign messy real-world operations into reliable systems.
In healthcare admin, that means one thing:
Small, grounded automation beats flashy demos. Every time. ⚙️🏥🤖
#AI #HealthcareAI #Automation #DataEngineering #AgenticWorkflows #Python #SQL #n8n
🌤 1. HOPE – We’re People, Not Numbers
The world counts us by views and followers 📱
But you and I—we’re more than numbers.
We breathe, we try, we keep going. 🌿
If one person feels a little less alone because of my words,
that’s already a movement of hope. ✨
Hope isn’t built by algorithms—
it begins with a whisper inside:
“Let’s try living one more time.” 💫
🌱 2. TRUST – Trusting Myself
The world says, “Do more. Be faster. Prove yourself.” 🔁
But I’m learning to trust my own pace instead.
Even if I move slow, I’m still growing. 🌾
Not in numbers, but in genuine moments I’ve lived.
Today, I remind myself—
I’m trustworthy for not giving up. 🤍
🌷 3. MERCY – Being Kind to Myself
“Do better. Show more.”
That’s what the world demands.
But my heart says softly,
“You’ve come this far—be gentle with yourself.” 🌸
It’s okay to mess up. It’s okay to pause.
It’s okay to not post for a while. ☁️
The world rarely gives mercy,
so I’ll start by giving it to myself first. 🍃
💞 4. LOVE – Choosing Connection Over Attention
It’s easy to mistake love for likes and comments. 💬
But real love grows quietly—in places numbers can’t see. 🌿
If one person finds comfort in my words,
that’s already love moving. 💫
I’m learning to trust invisible connections, not attention.
And maybe, between you and me who stayed till the end,
a small love has just begun. 💗
# Key Implications of the SaaSpocalypse
## 1. Fundamental Crisis in the SaaS Business Model
Until now, SaaS has generated revenue through a "monthly subscription fee per employee per seat." However, because AI agents operate on a structure where they "connect multiple tools to produce results using a single prompt," the reason for companies to maintain 10 to 20 SaaS platforms disappears. Once friction vanishes, the moat of SaaS—selling "convenience"—is crumbling.
## 2. Short-term Panic vs. Long-term Evolution
While the stock prices of major SaaS companies like Salesforce, ServiceNow, and Snowflake have plummeted, CEO Marc Benioff countered, stating, "We have already experienced several SaaSpocalypses," and adding that "agents could actually increase SaaS usage." In fact, some analysts view this as "not the real end, but a consolidation." Small and medium-sized SaaS companies that fail to properly integrate AI will be weeded out, while those that survive are highly likely to evolve into AI-native platforms.
## 3. Timeline Compression
AI disruption, which was previously predicted to take 3 to 5 years, caused trillions of won to evaporate in just 48 hours. This is the strongest evidence that “AI is faster than expected.” Korean startups (especially fintech and e-commerce SaaS companies) could face the same shock.
# Practical Insights for Preparing for the Future AI Era (2026–2030)
## Individual/Freelancer Perspective
- Upgrade from a “tool user” to an “agent conductor.”
Try using the latest agents like Claude, GPT, and Gemini in practice for 30 minutes a day. (e.g., “Connect my Notion + Gmail + Google Sheets to automatically generate a weekly report.”)
- Drastically reduce SaaS subscriptions, but strengthen unique human domains that cannot be replaced by AI (creative strategy, relationship management, ethical judgment).
- Hone your prompt optimization skills starting now to gain a significant competitive edge in the global market.
## Corporate/Founder Perspective
- Innovation in pricing models is the key to survival. Switch from Seat-based to Outcome-based or Usage-based (billing per result). An API-first strategy that enables AI agents to “call your SaaS internally” is essential.
- Rapidly develop Vertical AI (industry-specific agents) or add an “AI Agent Layer” to your existing SaaS.
- Build a data moat. Since AI agents become more powerful the richer the data, companies that protect customer data and domain know-how will win.
## Investor/Policy Perspective
- In the short term, focus on oversold SaaS companies with clear AI roadmaps (e.g., Salesforce’s Agentforce).
- In the long term, “AI infrastructure + agent platforms” are the real winners. (Anthropic, OpenAI, xAI, etc.)
- The Korean government and companies must urgently implement “AI Agent Adoption Guidelines” and retraining programs. This is because a disruption in the SaaS startup ecosystem would have a significant impact on jobs.
## One-Line Summary Insight
“SaaS is not dead, but it is being completely reborn in a form where ‘AI does the work for you.’”
Try handing over your workflow to an AI agent right now. In six months, the “old ways” will look outdated. Don’t panic; the real winners of the AI era are those who learn how to work with agents.
https://t.co/03iJzMPWlb
Nemotron 3 Super is a 120B (12B active) hybrid Mamba-Transformer MoE, an open model optimized for multi-agent, coding, and long-context inference.
With Latent MoE, MTP, 1M token contexts, and NVFP4, it delivers up to 5x higher throughput and 3x faster generation speed than its predecessor, simultaneously reducing both the "thinking tax" and "context explosion" issues.
The key point is that the weights, dataset, and entire training recipe are all open, allowing fine-tuning to suit individual infrastructures and serving as the main brain for agents, both on-premise and in the cloud.
Where hearts and codes coexist.
Technology should not replace humanity —
it should amplify it.
Building the future of
AI, data, and human-centered systems.
#CORCOIDUM#HumanTech#AI
The most common misconception people have when using AI
"AI isn't smart."
The problem isn't AI, it's the questions.
A good prompt can improve AI performance by up to 10x.
Four Key Points from OpenAI's Guide
1. State your goal clearly
2. Provide ample context
3. Specify the output format
4. Make requests in stages
The key skill in the AI era is
not coding, but
the ability to ask questions.
https://t.co/DygamIVCxy
Claude for Excel and Claude for PowerPoint now sync together seamlessly.
When you’ve got more than one file open, Claude shares the full context of your conversation between them.
Pull data from spreadsheets, build out tables, and update a deck — without re-explaining a step.