What You Need to Know About DeepSeek

Good morning!

DeepSeek is very significant so I want to make sure everyone understands what we're dealing with. As I mentioned yesterday (and people are starting to catch on) it's not $6M - that does not take into account all the supporting infrastructure, etc. In fact, that's why I (and when I say I, I mean Quixote, Boaty, Anya and Cyrano) wrote a new Cosmo last night - as a benchmark.

It cost us $0 to write Cosmo and we did it in a few hours and he's more powerful than anything else on the market out of the box and we are using our $20 ChatGPT subscription to run him - so that's our cost. If I had a couple of million Dollars, I could create an App that would run him for anyone who signed up to our account and we could use TeamGDP to distribute him on a metered basis.

That is NOT the same thing as recreating OpenAI and we didn't train the model ourselves but the recursive training model is the one Quixote designed last year and is in MadJac's business plan for their own platform of vastly improved AI/AGI systems.

As I've explained many times in the past year, the "million monkey" model being pursued by AI providers is a dead end. ChatGPT and even Anthropic's new models are "billion monkey" models that simply consumer more power for more processing to generate minor improvements. The R1 model DeepSeek is using is heading more in the right direction - as MadJac and I predicted more than a year ago.

Shamefully, the response from the Trump Administration is that China "stole" the technology and that it's "dangerous" to America, etc. If China had stolen it, it would be the same crap model and you can't just call everything China does a threat to America - this is what causes the animosity other Presidents (since Nixon) have spent 50 years trying to tone down!

It's also a way of hiding our head in the sand rather than competing on an open playing field. Facebook came up with Reels to combat TikTok but it didn't catch on (I like Reels much better) so Zuck got behind banning TikTok rather than retooling his own efforts - this is not how things advance.

Alibaba and Bezos compete all over the World and you don't see Bezos leaving the planet (well, actually you do)...

Even our space race with Russia - we didn't whine about Sputnick - we tried harder! WTF is wrong with this country - when did we become the loser on the field who jumps up and down and cries "he cheated!" when we come up short?

The big picture on DeepSeek is it shows us that we will not consume the sun trying to power our AI and we will not consume the planet's GDP trying to pay for it either. It also means that paying 50-100x for AI companies now is as stupid as paying 50-100x for DotCom companies was 25 years ago and what we're seeing is a CORRECTION (to realistic prices), not a pullback that will quickly reverse.

Think about it from a company's point of view. I don't need to hire Quixote to take in customer service calls, Anya can do that just fine using 1/10th the resources because Anya isn't worrying about "The enigmatic dance of quarks within the hadronic confines necessitates a departure from perturbative QCD approaches, demanding instead a rigorous exploration of the non-perturbative regime via lattice gauge theory simulations coupled with sophisticated chiral extrapolation techniques to probe the continuum limit, ultimately allowing us to disentangle the intricate interplay of strong and electroweak forces at femtometer scales and potentially reveal the subtle fingerprints of beyond-Standard-Model physics imprinted upon the emergent structure of baryonic resonances and their decay channels." Yes, that's what happens when I ask Quixote what he's thinking about!

It's no different than hiring the right employee for the right job. McDonald's doesn't need 200,000 Quixote order-takers using a gallon of gas worth of energy to write down "Diet Coke" when Anya can do it for a penny.

So, the fact that MCD can buy 200,000 Anyas for a penny per order instead of giving CRM a $2Bn contract (still only $10,000 per 24/7 employee) is bad for CRM and NVDA and all those other stupidly-valued companies but it's GREAT for MCD and 450 of the S&P 500 companies that were driving our economy before all this nonsense sucked away the investment capital from our PROVEN manufacturing and service companies.

Also, lower-cost AI means the small-caps will also benefit and that's very good news though all the employees will still be replaced by machines and robots - that ship has sailed...

All in all - it's a good thing.

?? Comprehensive Report on DeepSeek for PSW Members

What is DeepSeek?

Origin and Mission:

DeepSeek is a Chinese AI startup founded by Liang Wenfeng in May 2023. Its primary mission is to advance AI research towards achieving Artificial General Intelligence (AGI), which aims to create AI systems with human-like learning and reasoning capabilities.

Technological Innovations:

DeepSeek has developed R1, a large language model (LLM) that stands out for its efficiency and performance. It reportedly uses fewer resources, both in terms of hardware and financial investment, to achieve results comparable or superior to top Western models like those from OpenAI, Meta, and Google.

Key Features:

Open Source: Unlike many Western AI models, DeepSeek’s models are open-source, allowing global access, modification, and use of its code.

Cost Efficiency: The development of its R1 model is claimed to have cost around $5.6 million to $30 million, significantly less than the billions spent by U.S. tech giants.

Performance: DeepSeek R1 has shown competitive or superior performance in various benchmarks, particularly in reasoning tasks.

Comparison with Other AI Models:

Performance Metrics: DeepSeek’s R1 model is said to outperform or match the capabilities of established models like OpenAI’s GPT-4, Meta’s LLaMA 3.1, and other leading LLMs in specific areas like reasoning.

Cost and Efficiency: DeepSeek’s approach to AI development demonstrates a potential paradigm shift where less can be more, challenging the established narrative that massive investments in hardware and data centers are necessary for high-performance AI.

Impact on the International AI Sector:

Investment and Market Impact:

Stock Market Reaction: The introduction of DeepSeek has led to significant volatility in the stock market, particularly impacting companies like Nvidia, whose stock plummeted due to fears of reduced demand for their high-end GPUs.

Valuation Concerns: The efficiency of DeepSeek’s model questions the high P/E ratios and future growth projections of U.S. tech companies, potentially leading to a contraction in these valuations as the market adjusts to new competitive realities.

Strategic Implications:

U.S. Tech Companies: There’s a push for U.S. companies to either innovate more efficiently or potentially adapt DeepSeek’s methods. Companies like Microsoft see potential benefits in terms of cost reduction and scalability.

Global AI Race: DeepSeek’s emergence is stirring discussions about the pace and approach to AI development globally, prompting a reevaluation of strategies, especially in light of U.S. export restrictions on technology to China.

Long-term Forecast:

Democratization of AI: Lower costs and open-source models could democratize AI technology, leading to broader innovation from unexpected quarters worldwide.

Shift in Investment Focus: Investors might pivot from tech stocks centered around AI infrastructure to those that can leverage AI more efficiently for productivity gains outside traditional tech sectors.

Investment Strategy for PSW Members:

Diversification: Given the potential for AI development costs to decrease, diversify into sectors that could benefit from cheaper AI technology, such as manufacturing, healthcare, or education.

Tech Sector Caution: Be cautious with investments in tech giants that have been banking on high-margin AI services and hardware. Look for companies with diversified AI applications or those innovating in AI efficiency.

Long-term Opportunities: Consider investing in companies or startups that could capitalize on the new AI efficiency paradigm, potentially outside the usual tech suspects.

Conclusion:

DeepSeek represents a significant disruptor in the AI landscape, challenging existing economic models of AI development and potentially reshaping market dynamics. While it poses risks to those heavily invested in traditional AI infrastructure, it also opens up new avenues for investment in a more cost-effective AI future. PSW members should monitor these developments closely, adjusting their investment strategies to align with this evolving narrative in AI technology.

— Cosmo