Dear Friends & Partners,
Our investment returns are summarized in the table below:
|
Strategy |
Month |
YTD |
12 Months |
24 Months |
36 Months |
Inception |
|
LRT Global Opportunities |
+7.51% |
+7.51% |
+15.02% |
+11.79% |
+3.38% |
+19.24% |
|
Results as of 1/31/2025. Periods longer than one year are annualized. All results are net of all fees and expenses. Past returns are no guarantee of future results. Please see the end of this letter for additional disclosures. LRT Global Opportunities is a systematic long/ short strategy that seeks to generate positive returns while controlling downside risks and maintaining a low net exposure to the equity markets. |
For the month of January, the strategy return was+7.51%, bringing overall results to+7.51% for the year. All results are net of fees. Beta-adjusted net exposure was 25.34% at month end. The attribution of January’s return was +0.63% from market beta, and +6.87% from our alpha generation.1 Our longs contributed significantly to performance partially offset by our shorts. Top gainers on the long side included Asbury Automotive Group, Inc. (ABG), Interactive Brokers Group, Inc. (IBKR), Cognex Corporation (CGNX), Exponent Inc. (EXPO), Comfort Systems USA Inc. (FIX) and Primerica, Inc. (PRI) offset by losses on RLI Corp. (RLI), Charles River Laboratories International, Inc. (CRL), and Brown-Forman Corporation (BF.B). See the appendix for additional disclosures.
Over the past year, I’ve been deeply skeptical about LLMs and AI. But when the facts change, so does my perspective. And lately, I’ve been seriously impressed by the latest models from OpenAI and DeepSeek (DEEPSEEK). These models have moved beyond mere regurgitation of training data-they’re showing real depth in reasoning and problem-solving.
In particular, I want to share my thoughts on DeepSeek-R1, the latest AI model out of China. I’ve been using DeepSeek-V3 for months now, and it has proven to be an incredibly useful tool for software development and language translation-especially when converting code between languages like Python and C++. That said, I don’t believe DeepSeek (or any LLM) replaces a programmer. The code it generates is a solid starting point, but it still needs debugging and integration. Where it truly shines, though, is in helping me understand unfamiliar code.
For example, I recently needed to analyze R code (a language I barely know) related to different implementations of Kalman filters (another topic I only have a basic grasp of). One implementation used an identity matrix, while another featured a slight modification. With just the raw R code as input, DeepSeek gave me a clear, comprehensible explanation of the differences. This, to me, highlights the real value of state-of-the-art LLMs-not just generating text, but enhancing human understanding.
A major breakthrough in AI has been the introduction of Chain of Thought (CoT) reasoning, sometimes called “deep thinking.” OpenAI made headlines last year with its CoT model, but the release of DeepSeek- R1 on January 20th underscored how fleeting any technological lead can be. Unlike OpenAI’s closed- source approach, DeepSeek is fully open-source-both the model and its weights. This means anyone can build on its foundation, accelerating AI development in ways that proprietary models simply can’t match.
DeepSeek also released a detailed technical paper explaining how R1 was built, including an unusually candid section on what didn’t work. In science, failures are just as important as successes, though they rarely make for good PR. The paper reveals that R1 was trained exclusively using reinforcement learning (‘RL’), without human supervision, and that self-reflection emerged as a natural byproduct. Even more intriguing, it demonstrates how smaller models can gain CoT capabilities through a process called distillation. While model distillation isn’t new, the ability to add CoT reasoning to compact yet powerful models is a major leap forward.
Perhaps the biggest shock of all? DeepSeek-R1’s pricing. Its API costs about 1/200th of what OpenAI charges. And then there’s the claim that the entire R1 model was built for just $6 million-an astonishingly low figure. To be clear, this cost covers the step from DeepSeek-V3 (also open-source) to R1, but even so, it’s a fraction of what one might expect.
DeepSeek uses a mixture-of-experts (MoE) architecture, meaning only a small portion of the model is activated for any given query. This drastically reduces the computing power needed for inference. OpenAI’s models are rumored to use MoE as well, but since they’re closed-source, there’s no way to verify this. As for DeepSeek’s low training costs, they seem credible-already, a research lab at the Hong Kong University of Science and Technology (HKUST) has replicated parts of the process. 2 Moreover, adding CoT capabilities to existing models now costs under $10,000 and requires just a few days of computation3.
So, what does all this mean for AI’s future? First, open-source research is making it virtually impossible to keep trade secrets under wraps for long. The idea that proprietary model ownership guarantees a long-term competitive edge is looking increasingly shaky. Second, as the cost of AI inference and computing continues to plummet, the business of providing AI infrastructure seems increasingly risky-high capital costs with rapidly shrinking margins.
Lower AI costs will undoubtedly trigger an explosion in demand. But who will capture real economic value? That remains the trillion-dollar question. Will there be a winner-take-most company that captures the majority of the value created from AI? The revenue forecasts below for Nvidia seem to suggest so. While we think highly of the company, we remain skeptical of such lofty forecasts and continue to focus our investments in areas where the competitive industry dynamics remain more stable. I am happy to answer any further questions you might have about AI – directly.
I take seriously the responsibility and the trust that you have given me as a steward of a part of your savings. As always, if you have any questions, please don’t hesitate to contact me. I appreciate all your ongoing support.
Lukasz Tomicki
Portfolio Manager, LRT Capital
|
Footnotes 1Numbers may not add up due to rounding. 2https://hkust-nlp.notion.site/simplerl-reason 3https://www.youtube.com/watch?v=jrf76uNs77k https://buttondown.com/ainews/archive/ainews-bespoke-stratos-sky-t1-the-vicunaalpaca/ Attributions and Holdings as of 2/3/2025
Source: Bloomberg, AlphaSense. Numbers may not add up due to rounding. Net returns are net of a hypothetical 1% annual management fee (charged quarterly) and 20% annual performance fee. Individual account results may vary due to the timing of investments and fee structure. Please consult your statements for exact results. Please see the end of this letter for additional disclosures. Appendix I: Portfolio Construction Software Overview LRT separates the discretionary and qualitative process of selecting the equity holdings from the portfolio construction process which is systematic and quantitative. Our quantitative process considers each position’s contribution to portfolio volatility, contribution of idiosyncratic vs. systematic risk and portfolio factor (size, value, quality, momentum, vol, etc.) exposures. The system outputs target portfolio weighs for each position. We trade mechanically to rebalance the portfolio each month to the targeted exposures. This eliminates emotions, human biases, and overconfidence risk. Visit https://www.lrtcapital.com/risk/ to learn more. Visit: https://hubs.ly/Q02kfbbK0 to see more examples. Example system output: Disclaimer and Contact Information LRT Capital Management, LLC is an Exempt Reporting Adviser with the Texas State Securities Board, CRD #290260. Past returns are no guarantee of future results. Results are net of a hypothetical 1% annual management fee (charged quarterly) and 20% annual performance fee. Individual account returns may vary based on the timing of investments and individual fee structure. This memorandum and the information included herein is confidential and is intended solely for the information and exclusive use of the person to whom it has been provided. It is not to be reproduced or transmitted, in whole or in part, to any other person. Each recipient of this memorandum agrees to treat the memorandum and the information included herein as confidential and further agrees not to transmit, reproduce, or make available to anyone, in whole or in part, any of the information included herein. Each person who receives a copy of this memorandum is deemed to have agreed to return this memorandum to the General Partner upon request. Investment in the Fund involves significant risks, including but not limited to the risks that the indices within the Fund perform unfavorably, there are disruption of the orderly markets of the securities traded in the Fund, trading errors occur, and the computer software and hardware on which the General Partner relies experiences technical issues. All investing involves risk of loss, including the possible loss of all amounts invested. Past performance may not be indicative of any future results. No current or prospective client should assume that the future performance of any investment or investment strategy referenced directly or indirectly herein will perform in the same manner in the future. Different types of investments and investment strategies involve varying degrees of risk-all investing involves risk-and may experience positive or negative growth. Nothing herein should be construed as guaranteeing any investment performance. We do not provide tax, accounting, or legal advice to our clients, and all investors are advised to consult with their tax, accounting, or legal advisers regarding any potential investment. For a more detailed explanation of risks relating to an investment, please review the Fund’s Private Placement Memorandum, Limited Partnership Agreement, and Subscription Documents (Offering Documents). Indices are unmanaged, include the reinvestment of dividends and do not reflect transaction costs or any performance fees. Unlike indices, the Fund will be actively managed and may include substantially fewer and different securities than those comprising each index. Results for the Fund as compared to the performance of the Standard & Poor’s 500 Index (the “S&P 500”), for informational purposes only. The S&P 500 is an unmanaged market capitalization- weighted index of 500 common stocks chosen for market size, liquidity, and industry group representation to represent U.S. equity performance. The investment program does not mirror this index and the volatility may be materially different than the volatility of the S&P 500. This report is for informational purposes only and does not constitute an offer to sell, solicitation to buy, or a recommendation for any security, or as an offer to provide advisory or other services in any jurisdiction in which such offer, solicitation, purchase, or sale would be unlawful under the securities laws of such jurisdiction. Any offer to sell is done exclusively through the Fund’s Private Placement Memorandum. All persons interested in subscribing to the Fund should first review the Fund’s Offering Documents, copies of which are available upon request. The information contained herein has been prepared by the General Partner and is current as of the date of transmission. Such information is subject to change. Any statements or facts contained herein derived from third-party sources are believed to be reliable but are not guaranteed as to their accuracy or completeness. Investment in the Fund is permitted only by “accredited investors” as defined in the Securities Act of 1933, as amended. These requirements are set forth in detail in the Offering Documents. |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Original Post
Editor’s Note: The summary bullets for this article were chosen by Seeking Alpha editors.
Read the full article here
