OpenAI Has Just the First Mover Hype?

Think about baby ducks: the first moving thing they see after hatching becomes their “mother”, even if it’s not actually their mother.

This same basic idea applies to how we value things in the market, especially new products.

When something completely new comes along, people don’t really know what it should cost, It’s like being in the dark without any reference points.

This is exactly what happens when consumers face a new product or service, they have no experience to compare it with.

Whatever price is set first becomes their reference point, usually the price that the first company in the market decides.

# Long Story Short of this first part:

Supply and Demand AREN’T independent forces.

(Keep it in mind as it’s fundamental to understand my claims)

# This effect has two parts:

• Arbitrary
  • because that first price could be anything.
• Coherent
  • because once people accept that first price, it sticks in their minds and influences what they think a reasonable price is for a similar product.

# What does this mean?

When people think something is new (whether it truly is or not), they can’t really judge if the price makes sense because there’s nothing to compare it to. The first company to enter the market gets to set this reference point for everyone else.This concept is also known as ‘Anchoring’ or ‘Priming‘ and it is applied for prices as well as for market leaders. I think OpenAI anchored the world, in a way that every person that works in any field besides AI believes LLMs and Generative AI is just about ChatGPT and OpenAI products.

# The reality is that OpenAI is:

1. anything but not open
2. overspending in development
3. overpricing all their products

Later we will discuss the first point, which is linked to why OpenAI doesn’t want to lose its First Mover Hype.

But let’s now analyze the second and third ones:

2. OpenAI stated that running 1 task on o3 (which is their new model in development) costs them around $1000, which is more or less 170 times the cost of running 1 task on o1.

3. OpenAI makes you pay $200 to use their best models without limitations and now DeepSeek has very similar performance while it’s basically free.

By the way, as of the 2024 close, OpenAI is $5 Billion in loss, despite generating $3.7 Billion in revenue.

# Let’s compare OpenAI with its competitors:

As a developer, my personal experience says to me that Claude’s model series from Anthropic is way better than OpenAI’s model series.

Claude:

• is the best at programming and writing,
• follow the instructions properly,
• is created for professionals.

On the opposite, OpenAI’s models are created for consumer use, for example, ChatGPT has a basic free plan with infinite use, positioning OpenAI in a potential ‘No Marketing Required Zone’ similar to Tesla.Claude even in the same basic free plan has daily limitations, cause their product is valuable and if you want to use it a lot, you must pay accordingly.

# DeepSeek, the elephant in the room:

You heard for sure about DeepSeek as everyone is talking about it.The most important thing to understand why US Stock Market shitted its pants after the DeepSeek R1 launch, is that DeepSeek now offers a very similar OpenAI flagship product option for free.

(By the end of this article, you will know why this isn’t even a problem for Nvidia or the US Stock Market)

More specifically, DeepSeek R1 is the best LLM out there.

# But why is that?

DeepSeek R1 is:

• open source
• 97% cheaper than OpenAI o1
• released with smaller versions that can run on normal office computers

In a matter of days, the market realized that there was no need for:

• astronomical capital raise
• spend a lot in API costs
• use OpenAI products

To sum it up, you can now run your dedicated instance of DeepSeek R1 in the Cloud and on your own laptop.

# Am I a DeepSeek advocate?

No, I stand for open-source products,not for DeepSeek or Chinese censorship.I think all AI and LLM development should be open-source.Ironically, a Chinese Venture Capital open source their LLM and OpenAI doesn’t.

# Do R1 distilled models are equally powerful to OpenAI o1?

NO, the right question is:

# Do you need such powers that o1 can offer?

Probably NO.

# Is this a win for China?

NO, TLDR: it’s a win for the whole AI industry development.OpenAI never revealed how they do things but suddenly very similar things are available to everyone.Moreover, now the reasoning AI field is more accessible and more progress will be made here.So yes DeepSeek helped more than OpenAI, which by the way, isn’t open at all.

# Is this a loss for Nvidia and the US Stock Market?

NO, TLDR: the Jevon paradox will play out.Let me explain better, as a technology becomes more accessible to the broader public, more people will use it.Even more people will be willing to run an LLM locally and yeah you can indeed run DeepSeek on a normal computer, But it is also true that a company will most likely need to run the most powerful and expensive one which requires high-end GPUs.

# Does OpenAI want to lose their First Mover Hype?

Of course not, As a matter of fact, 3 days later the publication of DeepSeek R1, OpenAI announced Operator: which is a shiny new technology that can use your browser by prompting the LLM. It can do a lot of useful stuff for sure.Guess What #1: Operator is now available only to those who spend $200 for their subscription.Guess What #2: The same core technology is available to anyone and for free on GitHub.Of course, not everyone has the necessary knowledge to execute some commands, but it’s more than obvious that LLMs do and can teach you how to do it for free.

# Is this a lose for closed-source LLMs?

YES.

# There’s a saying:

If the thing that allows you to be the top player is a secret, then maybe the value is the secret, not you.

# The final question is:

If your secret is shared, would you still be the top player?

Originally published here on Substack, January 2025.