Serial Entrepreneur | ๐ CEO, AI Startup Accelerator | ๐ฅ Medical AI & Healthcare | ๐ฐ๐ท๐จ๐ฆ First Korean Permanent Resident via Canadaโs Startup Visa(SUV)
Finally, the X Cheat Sheet is ready to launch! ๐ฅ And I decided to share this for free!!
I poured 6 hours into this and am excited to share it. It contains all the strategies I used to gain 1,000 followers in just 15 days.
When I started, I opened a $20 waitlist and 82 people signed up! If everyone bought it, that wouldโve been $1,600, but Iโve decided to give it away for free. Iโve grown this fast thanks to your amazing support, and I want to pay it forward.
No need to pay $20. Just follow, repost, and comment. Iโll DM you the Notion page link!
Thank you for being on this journey with me. Letโs keep growing together! ๐
๐ We just open-sourced the Mejurix Medical-Legal Embedder โ designed for the real-world needs of North American healthcare litigation and insurance.
Until now, legal-medical AI teams had to rely on domain-specific models like BioClinicalBERT or LegalBERT, or general-purpose embeddings like OpenAIโs text-embedding-ada-002. But these models lack true understanding across both domains โ which is critical in personal injury cases, where medical findings and legal implications go hand in hand.
Our fine-tuned model bridges this gap, delivering high-precision embeddings across 10+ medical-legal relationship types (e.g., severityโcompensation, symptomโdisease, outcomeโsettlement). It outperforms baseline models in cross-domain similarity and is optimized for QnA, summarization, and similarity search in legal-medical workflows.
๐ข Weโll also be releasing the model on Hugging Face Transformers very soon โ making it even easier to integrate into your applications!
๐ GitHub: https://t.co/RxsHomEYad
Vertical AI at the intersection of healthcare and law.
https://t.co/f8PP6GIlow
ํด๋จธ๋ ธ์ด๋ ํผํฉํฐ๊ฐ ๋ค๋ฅธ ํน์๋ชฉ์ ๋ก๋ด๋ค๋ณด๋ค ํ์คํ ์ฐ์๊ฐ ์๋ ๋ถ๋ถ์ด ์๋ค๋ฉด, ๋๋ถ๋ถ์ ๋๊ตฌ๋ ํ๊ฒฝ, ๊ณ๋จ์ด๋ ๊ฑด๋ฌผ๋ค์ ํํ๊น์ง ๋ชจ๋๊ฐ ์ธ๊ฐ์ ์ ์ฒด์ ์ต์ ํ๋์ด ์๊ธฐ ๋๋ฌธ์ ์๋ฌด๋ฐ ๋ณ๊ฒฝ(?)์์ด ๋ฐ๋ก ๊ธฐ์กด์ ์์ฐ์ธํ๋ผ์ ํฌ์ ํ ์ ๏ฟฝ๏ฟฝ๋ค๋ ๊ฒ์ด๋ค. ๋ชจ์๋ ๋ถ๋ถ์ ์ํํธ์จ์ด๊ฐ ๋ด๋นํด์ ๊ฐ์ ํด ๋๊ฐ ๊ฒ์ด๊ณ ...
Humanoid robots have a clear edge over specialized form factor robotsโmost labor environments are already designed for the human body. From tools to workspaces, stairs to doorways, our world fits the humanoid shape. Building robots that mimic this form factor means they can seamlessly step into roles without needing costly redesigns of existing infrastructure. Adaptability is key, and humanoids nail it.
Cursor is down, how can I do this without you @cursor_ai, I keep getting "Connection failed. If the problem persists, please check your internet connection or VPN"
Are You Quick-Witted? You Will Survive in the AI Era.
According to Sam Altman, GPT-4.5, the largest and most human-like model, was developed with an enormous investment of resources, resulting in a high price tag.
To put it simply, AI has finally learned to "read the room," and what was once considered an exclusive human abilityโquick wit and social awarenessโturns out to be an extremely valuable resource.
How Can "Quick Wit" Be Defined Technically?
The intelligence we call "quick wit", often learned through life experiences and hearing phrases like "Be more aware of your surroundings" from adults, is actually a highly sophisticated cognitive function.
At its core, it involves social intelligence, which allows one to recognize social boundaries, and emotional intelligence, which helps one consider others' feelings when communicating or acting.
For AI to develop this capability, it must master contextual reasoning, affect recognition, and social nuance detectionโskills that humans acquire naturally through social interactions.
The Evolution of GPT-4.5: Has AI Started "Reading the Room"?
Looking at the GPT-4.5 demo presented at launch, the difference from previous versions is striking.
For example, in earlier models, if you asked AI to draft a message to a friend who was late for an appointment, the response might have been so blunt that it sounded like an attempt to sever the friendship.
In contrast, GPT-4.5 now handles the situation naturally, merely pointing out the tardiness and asking for a reasonโjust as a human would. (For people, this isn't such a difficult task.)
In essence, AI has begun to understand undefined contexts and situations.
This is an ability humans develop through years of personal interactions, experiences, and social touchpoints.
It isnโt something found in books, laws, or even moral principles.
It is a metaphysical aspect of human natureโone that varies from person to person, shaped by individual life experiences. Some react aggressively, others compliantly, some gently, and others appropriately.
For AI to acquire this skill, it requires significantly more training time and computational power. (After all, even humans take over 20 years to fully develop this ability.)
As revealed in the demo, GPT-4.5 was trained using expanded reasoning capabilities and unsupervised learning techniques.
This shift in methodology dramatically increased the computational resources required,
leading to a 30-fold increase in usage costs compared to previous GPT models.
Whatโs Next?
Now that we know AI can acquire a sense of "quick wit" and social sensitivity, the next phase is another battle.
If we can enhance the DeepSeek base model with an effective persona reasoning dataset
and region-specific social norms (or rather, subtle social nuances),
we could build a second fast followerโone that shocks the world without the massive resource investments that OpenAI has made.
Has MVP + Agile Reached Its Limit?
There was a time when starting a business meant everyone kept talking about BM, BM. Back then, just asking whether there was an MVP seemed like a sophisticated question. If an investor asked, "How much revenue do you have?" they could even be seen as someone who didnโt understand tech trends.
But now, the situation has completely changed.
After analyzing YC W25 companies, I couldnโt find a single one that got accepted with just a basic MVP.
It was honestly surprising how well-built their services were.
From UI to design and user experience, nothing was lacking.
The traditional definition of an MVP, a "minimum viable product that customers want," is no longer valid.
Now, it needs to be a minimum payable product (MPV)โa product compelling enough to trigger a customerโs purchasing desire.
Why Was MVP + Agile the Go-To Model for Startups?
The reason MVP + Agile strategies were effective in the past was simple.
1. Development resources were the most expensive part of the process.
2. Market analysis could never be perfect, so startups had to keep experimenting to find the right direction.
But now, itโs a completely different era.
- Thanks to open source, there is no need to reinvent the wheel.
- Even development resources are becoming cheaper with tools like Cursor and cloud-based development platforms.
- Market research? No longer a bottleneck. By leveraging social media, news sources, and LLM-powered research services like Deepsearch, startups can generate a fully structured report within hours.
Just yesterday, I saw that DeepSeek released a bias-mitigated model on Hugging Face.
At this point, LLMs have become a public utilityโjust like electricity, computers, and storage.
Q. What if an investor still talks about MVP and Agile?
A. Walk away.
Instead, startups should also avoid the mistake of trying to convince investors by saying, "We have a technical barrier."
Honestly, investors donโt believe it, and they wonโt be convinced.
(Sometimes, founders from top U.S. schools manage to persuade them, but even they donโt have a real solution.)
In the past, there was a famous saying in the startup scene:
"Someone else already has your idea."
This was meant to encourage founders not to ask for NDAs but instead to share their ideas as widely as possible and gather momentum.
But today, the reality is:
"The product you're building can already be assembled like LEGO blocks and launched by someone else tomorrow."
The old mindset that "execution is everything" is no longer enough for startups.
If you move fast in the wrong direction, it will only take longer to get back on track.
Research like a needle, and build broadly using open source and AI. (It doesnโt take as long as you think.)
If you research broadly and develop like a needle, you will fail.