Hacker News Reader

Comment by openAIThrowaway | Hacker News Reader

> Meanwhile, Nadella is applying serious pressure. Microsoft didn’t pour billions into OpenAI just to lose ground

This part especially doesn’t ring true. Microsoft started hedging their bets after the Altman fiasco and are moving toward an AWS like strategy where they are hosting every model that they can get their hands on. AWS is already hosting DeepSeek on bedrock. There is no reason that Microsoft couldn’t do the same

> Some even worry that if OpenAI collapses, it could tank the entire AI sector, dragging the U.S. economy into a deep downturn given how much Big Tech and Wall Street have bet on this.

This makes no sense either. The only company with any market cap that will be affected negatively by LLM commoditization is Nvidia.

Google still has a distribution advantage and Microsoft and Amazon will be glad to host any models. This only helps Apple and Meta is already open sourcing thier model.

This entire post reads like someone who doesn’t have the slightest grasp on industry trends.

Besides, why would anyone making OpenAI comp post thus openly? I would be hoarding cash as long as possible and I could think of a couple of ways to profit off of this information legally.

Yes. And Microsoft is already hosting multiple DeepSeek options in Azure. That’s quite a hedge.

https://azure.microsoft.com/en-us/blog/deepseek-r1-is-now-av...

https://learn.microsoft.com/en-us/azure/ai-studio/how-to/dep...

If it's a true story from inside the organization then it's what I would expect from a large organization that got blind-sided by a competitor. Seen it several times before. The rapid advances in models coming out of China is not going to stop, seems like development of models around video is especially moving fast. I'm guessing the moat in the future will be the brand/device, when you have a chat/functions as part of the device it will be hard for anyone else to compete even if the models are slightly better. i.e Google search all over again. Personally i don't think the value from todays LLM are near the sunk cost, hoping that will change.

It’s a troll post. If it isn’t obvious, every reply is a retort of “you must be a Chinese spy”.

Everyone in the industry has been saying “there is no moat” around AI well before DeepSeek became public knowledge. This shouldn’t have been a surprise to anyone. All of the cloud providers have been hedging their bets

Yea the edits made after my response just made it less interesting.

I'm sure this is just fan fiction, but you're saying some people think "DeepSeek has cracked something fundamental that we missed", and other people think they just stole everything? Any actual evidence one way or another?

This reads like it was written by The Neuron posing as an insider at OpenAI. Source: I write for the Neuron and did not write this lol.

Why would I believe and why posting it here?

> Lately, there’s been a growing paranoia inside OpenAI—whispers about “Chinese spies” in the office.

Wasn't it obvious from the very beginning that it was likely that China would try to get their hands on this tech? If they're only getting paranoid about it now that seems shortsighted.

Why does this read like something written by an LLM? Down to em dashes ("—") being used, something humans online don't do. Could be real but are people really putting their posts through LLMs for editing?

1. I am human and use em dashes in both my formal and informal online writing. I see other humans doing it all the time. :-)

2. I do frequently use LLMs to improve my writing/grammar/spelling. When I do, they always removes my em dashes and replace them with words.

I don’t think any LLMs use dashes in the way OP is using them.

Some LLMs use dashes when doing lists but that’s not what OP is doing.

This is _exactly_ how ChatGPT recently uses punctuation (to the point that it can be used for detection, replacing old "as an AI language model"). Not Claude (which refuses to process this text!), not DeepSeek, not Gemini (which prefers classic quotes).

So if this would be a real OpenAI employee, it would take no time to find the account of user who generated this text. However as this is just a typical sensationalist troll slop with zero commercial secrets, so no action will be taken.

Someone needs to grep logs at OpenAI/Deepseek to dox the op.

This is a doom troll, right?

Why, cuz it makes sense?

> But let me remind you—we’re standing on the edge of World War 3. This isn’t just about OpenAI or DeepSeek; it’s about who controls the future. If China wins the AI race, or if the U.S. loses its edge, we won’t just be talking about lost jobs and struggling tech companies. This is about global power, and the consequences will be catastrophic. The outcome of this war will shape the future of you, your manager, your mother/father, your children, and everyone you know.

Either this part among others is from someone on the outside that is believing the hype or it could be someone from the inside “getting high on their own supply”.

There are a lot of risks that we have from China that doesn’t involve AI - hacks on infrastructure, distribution of supply lines that start or go through China including rare earth minerals, bombing Taiwan, etc

It's a shame my three takeaways are the following.

1. I hope those Chinese spies are really talented and have been great contributors. Maybe look at what they have contributed in tangible work since they have been hired. Talented engineers are desperate for work and OpenAI hires Chinese spies? Well, hopefully they were 10x performers when they weren't stealing secrets.

2. International collaborations? Of course everything was going to be stolen and eventually make its way to China.

3. China's primary mission objective is to steal IP.

> DeepSeek's next-gen models are already outperforming us, especially in reasoning and long-context capabilities

This is interesting - this is the first time I heard someone claim that DeepSeek may excel at long-context capabilities. All public benchmarks show that DS-V3 isn't very good at dealing with longer inputs, but maybe OAI had better evals internally?

In short, I don't believe you and I think there's some technical details that do not add up.

And yes, before you say it, I am sure I am a Chinese spy /s

Proof of domain ownership? Add a meta tag to the html served from openai.com <meta hnitem="42999774">

If this is genuine someone would have gone to great lengths to avoid detection. I doubt they would or perhaps even could do as you ask, and why would they?

exactly my point. This is so obviously NOT genuine.

No idea how I helped you reach that conclusion. It wasn't mine.

That would nullify the point of posting anonymously. Most likely that would instantly identify the employee in question to OpenAI staff.

Reply on news.ycombinator.com