No matter how great the company, it has to be the case that there
is always
more talent outside the company than inside the company.
The only way to stay
competitive is to be able to attract and
retain more talent into the company.
The only way to attract great talent is by helping as many people
as possible
and being as inclusive in culture as possible.
Recently, I realized that text-content products are very ineffective
in content
recommendation compared to video-content products like
TikTok. I have been
thinking about how to improve text-content
products in general.
The ecosystem of text-content is viral with lots of creators and
consumers.
However, somehow, I feel like the text-content product
is not well done so far.
One thing I found is that text-content products' recommendation
engines are
very ineffective compared to video-content products'.
One important reason is
because users read a short thumbnail of
titles before deciding they want to go
inside and read.
But this boils down the entire content down to a few bytes of
characters, which
oftentimes is too low bandwidth to be effective.
So the entire success versus
failure of textual content rests on
if the title is appealing to users.
In contrast, video-content products like TikTok are much more
effective; the
first five seconds of a video provide very high
bandwidth, allowing users to
quickly and effectively decide if
they want to continue watching the whole thing.
It turns out that how the interface is designed determines if a
recommendation
engine would work or not. For text-content
products, because users can only
decide to read based on such
limited bandwidth, oftentimes recommendation
engines just makes a small amount of articles go viral .
As a result, most articles are not explored at all, preventing the
recommendation
engine from effectively discovering whether majority of the content is good or bad.
Being able to identify which way is better will become more important
than the
traditional coding skill of knowing how to implement a
feature. In other words,
the "what" will become more important than
the "how."
Looking at today's extremely successful entrepreneurs, the one common
thing
they shared is an extremely ambitious, to a crazy degree,
vision, which also
happens to be right at a relatively early phase
in their career.
A few examples are Chinese entrepreneur Zhang Yiming, who realized
information distribution was key to society's efficiency when he was
in college;
Google's DeepMind founder Demis Hassabis saw AI as a way
to solve biology
long before he received the Nobel Prize in protein
folding.
Elon Musk thought electric power would be central to the transition
of energy
when he was in college. These early visions made sure that
their paths weren't
taken at random later in life.
This is probably the single most important thing for a business.
The best
businesses understand their customers better than customers
understand
themselves. We can see this by using some of the best
products in our daily life.
These best products really get us and what we want. As users, we
want to be
delighted more and more by great products that truly
understand us. And we
are definitely willing to pay for those
products that really matter to us.
Thinking from customers' angles is probably the greatest
superpower.
Infra will be oriented towards maximizing leverage of LLM. An
example is the
question: "Which programming language to use?" Of
course, this depends on
the task to be solved.
But at the same time, it's more important to think in this way:
"Which language
is easiest for LLM?" At current LLM capacity, this
means some simple languages
like Python or React.
This is why the development ecosystem of these languages will become
more
viral as LLM gets more adoption among programmers. It's simply
more efficient
for programmers to work in languages that the LLM is
good at.
I previously saw a similar post saying that Swift is getting less
adoption because
it's less LLM-friendly. At some point, the business
or even society's
infrastructure will be designed to gear towards
maximizing the value generated
by LLM and AIs.
Being able to interface effectively with intelligence technology like LLM will be
the key to future world's innovation and productivity. To be able to interface with
LLM, requires the user to have a basic understanding of the technology itself.
The LLM itself of course has a great understanding about the technology. But
it's still necessary for humans to have a rough picture as well.
Therefore, I think the future of education is surrounded by learning based on this
kind of interface. I think having a good understanding of technology, rather than
knowing how to do every task in every stack of the technology,
will become far
more important in the future.
Essentially, I think
the people who are CEO-like, who have a general understanding
across multiple layers of the company, will better leverage
intelligence technology
than people who specialize in a particular
stack. This will be recursively true in each
individual stack as
well: the people who specialize in a particular stack will better
leverage technology if they also understand different lateral
aspects related to the
stack. Because AI will be so good at
executing a well-described task, the hard part
will be describing
the task itself.
To describe the task requires the person to have an understanding
of the
importance of the task itself. This is why understanding
is far more important.
The wrong question is: if I don't have a GPU right now, why would I
study CUDA? The right question is: if I
have a GPU, what would I do
with it?
This is an interview question from Starbucks for high-position
leadership roles: If you have $1 billion today,
how would you spend
it? Most times, we tend to think and act based on the current
condition. This doesn't
take into account the proactiveness of being
an agent.
In reality, getting a GPU is relatively easy; it just costs some
money. The real difficult question is: how would
I use a GPU to build
a better product? Once we find the answer to the latter question, the
former GPU
problem is easy to solve.
The same applies to startups. The fundraising problem is relatively
simple. The hard part is to have a vision
and understanding for
building the product.
This is something I used to do as a habit during high school and rarely after
college graduation. The main reason I stopped was that I realized there were
plenty of people around me who were better at being technical than I was.
Therefore, I thought it was better for me to find my relative advantage.
This was a mistake. Staying technical is important to get inspiration on the
product side. Also, I realized that long-term learning for technical understanding
is actually very important and gives product founder an unfair advantage.
So two pieces of advice on how to do long-term self-learning: First, treat
knowledge as a rabbit hole. Don't seek structure at first. Only try to structure the
knowledge learned after. Otherwise, it causes a lot of frustration. Second, stay
humble, ignorant, and curious. Because learning is essentially a process of
recognizing my ignorance, without being humble, it's hard to continue.
The most important thing about programming is about forming
the
programming model. The programming syntax is relatively easy to
pick up. The
difficult part is to understand how CUDA works with
hardware. This requires
forming an understanding of CUDA from
the hardware level. And this part is
very hard to be automated
away by AI.
I expect even in the future, having versus not having this
understanding will
cause leverage for human programmers to be
more productive and creative. But
once understanding the
programming model of a language, the rest of the work
can be
easily done by AI much more efficiently.
This lowers the bar for learning programming down to understanding the
programming model.