News Deepseek 'clearly not interested' in scaling up — 160-person team focused on developing new models

That “fleet of Nividia GPUs” is looking extremely weak compared to what US companies are using. Musk and Altman are both using over 100k H100s vs 50k random Nvidias, the best of which are significantly weaker than an H100 (H800 and A100 are both significantly weaker than a H100, heck the A100 is Ampere). It seems to me like much of the damage control trying to prove that DeepSeek didn’t train on 10% of the resources have instead proven they did indeed train it on like on 30%-40% at most.
 
As an entrepreneur myself, I would agree with their strategy. First, keep your monetary burn rate low by not making your overhead too large and too quickly. Second, make sure the market is truly what the hype is saying - which AI seems rather hollow to me. The AI market opportunities that everyone are going after already has players in those high value arenas such as IBM Watson (think: medical diagnoses, data analyses, financial market analyses, etc.). Third, once a good solid market has been identified, go after that market and don't do the usual startup goof where you chase everything with starry eyed hubris and waste your startup money.

Mature incubator programs always make sure that their startups have done a proper market study and that there is not too much competition to cause ultra high risk of competitive failure. Investment groups such as Berkshire Hathaway use these solid principles. Even on Shark tank, you will see the seasoned investors stay away from anyone who has not done their market due diligence.

For AI, It is going to be difficult to get enough of a return on $100B's of investment to pay it back in a reasonable amount time. Most investors are not going to be pleased to find a 5-10 year wait on returns. There are too many other opportunities out there with much shorter market return times, giving VCs chances to invest in multiple cycles with their money and amplify their returns.
 
  • Like
Reactions: smitz314
It's open sourced.

I'd take DeepSeek open source than OpenAI closed source anyday.

AI shouldn't be monopolized by Sam Altman and Elon Musk and Mark Zuckerberg, and billionaire tech bros.
 
  • Like
Reactions: smitz314
Well yeah. They're ultimately funded by the CCP. So they don't have to care about making money.
You do realize they have repeatedly turned down overtures to let CCP investment funds buy a chunk, right? There are actually tons of entrepreneurs in China that just want to do business without the government involved. Sure, if Deepseek gets any bigger the CCP will probably force their way in but I can assure you that most businesses aren’t working for the glory of the party.
 
  • Like
Reactions: smitz314
What happened to the buzz we built R1 for $6 million?
It was very transparent in their publication that $6 million is only training cost (variable), not total acquisition cost. That's the mainstream media not knowing the difference between variable cost vs. total costs. Even the total cost higher, it's still orders of magnitude cheaper than what GPT's total cost to train. This is why Sam Altman is calling for a ban on DeepSeek, because he can't compete.
 
  • Like
Reactions: smitz314
You do realize they have repeatedly turned down overtures to let CCP investment funds buy a chunk, right? There are actually tons of entrepreneurs in China that just want to do business without the government involved. Sure, if Deepseek gets any bigger the CCP will probably force their way in but I can assure you that most businesses aren’t working for the glory of the party.
But all companies only operate with the permittance of the party. I'm not saying there aren't smart people in China.