Dear all,
For those interested in GPUs, this ranking should be interesting.
Here is the link to XTX Markets AI capabilities
Regards
James
1 Like
What the heck is META doing with all those GPUs !? It’s just a social network to share pics with some friends, no?
Or are they planning to create friends for people that don’t have any?
Marco,
I think Mark Zuckerberg is trying to build his Metaverse and virtual reality with all those GPUs.
Regards
James
Llama refers to Meta’s sophisticated large language model, and you can access its coding resources via the following link: Code Llama.
Language translation during virtual meetings would be another use of Llama.
Additionally, the large number GPU cards within Meta’s ecosystem, is probably related to their use of VR headsets.
As many know, GPUs, or Graphics Processing Units, are great for matrix multiplications. This functionality is critical for rendering graphics, and the VR experiences, as well as for performing the backpropagation calculations necessary in deep learning processes.
I am sure they will do exactly that. Virtual friends and maybe you could talk to Einstein too (or an AI designed to channel him).
Here is the second most popular AI site designed to do exactly that already: Character AI. You could argue they do role-play games too (which is kind of similar but with a fantasy background story).
You can also find a therapist there too, BTW. I could use a therapist to deal with my fear of SkyNet becoming real (or even a worse nightmare combining the Matrix and SkyNet in a synergistic way). I am sure she can help me understand that I am just being paranoid (“a common but annoying human trait”). 
On the site, Einstein actually knows quite a bit about physics and the therapists know quite a bit about cognitive behavioral therapy (and other mainstream therapies). You can find a cute programmer to be a “pair programmer” which I did not know was a thing.
Character AI is the second most utilized AI tool, with a total of 3.8 billion visits, capturing 15.77% of the overall AI traffic. For comparison, ChatGPT is first with 14.6 billion visits over the year up to August (2023), holding a dominant share of 60.17% of the total traffic to generative AI tools.
TL;DR: Yep. That is the plan alright.
Edit: BTW, my life coach/drill instructor at Character AI claims to have access to ChatGPT 4.0 as well as additional programming to make conversations more natural. I really do think Marco has an important point even if it is also funny. There will clearly be virtual interactions with AIs in the future (some using Apple’s Vision Pro). 
Jim
“Character AI” could be an opportunity for P123. Imagine having a Warren Buffet bot tied to P123’s WB-inspired portfolio or ranking system. Or create bots for any of the other great investors tied to P123 ports that reflect their particular investment style. This could conceivably attract a different type of customer to P123, one that wants access to the latest financial tech without the burden of having to learn the tools.
This is already being done in other areas. For example, Chess.com allows one to play famous chess players. In theory, we could select a famous investor and have a conversation, discussing various companies…
1 Like
I asked ChatGPT as Warren Buffet what he would put into P123’s new AI/ML as features in a multivariate regression. Here is the unedited list:
"Certainly, here’s a concise list for you to copy:
- Price-to-Earnings Ratio (P/E)
- Return on Equity (ROE)
- Debt-to-Equity Ratio (D/E)
- Free Cash Flow Yield
- Earnings Yield (E/P)
- Market Capitalization
- Dividend Yield
- Beta
- Sales Growth
- Operating Margin"
All this AI hype is starting to sound depressing. We’re headed to a WALL-E future where robots will satisfy all of our needs and life will look like this
On the other hand, the humans on WALL-E are portrayed as happy, kind, not greedy (no crypto), with purpose, even though they do nothing for themselves. Here’s a great counterpoint: The Humans of ‘WALL-E’ Were Probably Better Off Without Him
Personally, I want a balance. I used paper maps to drive and, while it helps develop spatial orientation, I don’t ever want to go back. I also like that Facebook knows what I like and shows me articles in my feed that I usually find interesting (that’s all I use it for). Same with YouTube, I always find interesting suggestions. But the VR headset, AI everywhere, the “everyone connected” world Zuck thinks we want? No thanks, I’ll pass. Strapping on a VR set will not make me a better programmer, it will just make me sick.
Regardless, the world is headed to a WALL-E future. AI stocks will continue to explode. Humans need big challenges. Creating a superior brain is the next frontier (also life extension). Space, btw, is over. We’ll colonize Mars, but that’s the end of the road. We will never go anywhere else.
For P123 AI will play a role, but only for those that want to learn. We’ll explore AI to help the curious. For example the AI factors we’re working on can replace ranking systems, and be more responsive. P123 AI bots could help you write a formula, but you still need to be able to describe it. Most people (99% I hope) don’t want to be traders or have personalized strategies. They just want to invest and do better than indexing (anyone with less than $10M should). And that’s great. We have some ideas in the backburner for them that we’ll start working on when we have more bandwidth.
Cheers
2 Likes
I guess that’s a “NO”?
Character Bots may help connect with the younger generations. And it doesn’t mean personalized trading strategies as you suggested. It is quite the opposite. The bots would be trained around methodologies embraced by famous investors, not all that different than the models that P123 used to promote.
1 Like
@marco, personally I am happy to download. So I have no personal preference. But I get a fair number of emails from people starting to do ML at P123. Most are engineers with some programming experience. Without exception they use ChatGPT to make it easier at least. I don’t think that is an exaggeration. I can say I need it to do ML. For me it is as simple as that.
I would caution on resting on your laurels with this strategy. Do you not think Blackrock will offer an ETF for cheaper, and far less work, than what you described here: "Most people (99% I hope) don’t want to be traders or have personalized strategies. They just want to invest and do better than indexing (anyone with less than $10M should). " ? They will not need PMs anymore so the days of requiring huge scale for an ETF will dissipate since costs per ETF go down.
I think for the foreseeable future human creativity of research and development of new technologies to provide machines with connections they cannot make by imposed code limitations is where humans will fit it for some time.
People crave agency. Providing cutting edge technology with unique data sets is what wins.
I urge you to rethink this approach.
This is not anything I need personally and I would care about it only to the extent that P123 found this useful from a business perspective. I can (and do) download data, have ChatGPT help me with the code if I need it and run it in Jupyter notebooks. And soon I can do that internally to some extent at least at P123—without knowing exactly what that will look like. Or claiming to know…
But here is a site as an example of what Korr and InspectorSector are talking about as far as large language model integration: Composer
I do think this fits well with what Marco suggests:
Composer caters to that kind of client Marco is talking about, I believe. Look at the site and their marketing to a client who might say (not necessarily traders):
“I’d like a strategy that benefits from demand for GPUs.”
What? $288 a year with “FULLY AUTOMATED TRADING???”
So for those who “don’t want to be traders” they can let ChatGPT do it for them for $288 per year. Things are moving fast. 
I do think Marc Gerstein would develop strategies “inspired by” as @InspectorSector suggests. Here is what ChatGPT thinks the Patrick and James O’Shoughnessy might do with P123’s AI/ML (including the downloads). I think it is pretty good:
"Focusing on a hypothetical collaboration between Patrick O’Shaughnessy and James O’Shaughnessy, incorporating their quantitative investing approach with a modern twist of machine learning, here’s a list of 40 factors they might consider for a simple multivariate regression model. These factors blend traditional quantitative metrics with newer, data-driven insights that could be enhanced by machine learning techniques:
- Price-to-Earnings Ratio (P/E)
- Earnings Yield (E/P)
- Price-to-Book Ratio (P/B)
- Free Cash Flow Yield
- Dividend Yield
- Return on Equity (ROE)
- Return on Assets (ROA)
- Return on Invested Capital (ROIC)
- Debt-to-Equity Ratio (D/E)
- Net Margin
- Operating Margin
- EBITDA/EV
- Price-to-Sales Ratio (P/S)
- Sales Growth
- Earnings Growth
- Dividend Growth Rate
- Volatility (Standard Deviation of Returns)
- Beta
- Momentum (12-month Price Change)
- Reversal (1-month Price Reversal)
- Size (Market Capitalization)
- Liquidity (Average Daily Volume)
- Share Buyback Rate
- Gross Profitability Ratio
- Accruals
- Asset Growth
- Investment-to-Assets Ratio
- Inventory Turnover
- Accounts Receivable Turnover
- Sentiment Score (from NLP Analysis)
- News Sentiment (Volume and Positivity/Negativity)
- Social Media Sentiment Analysis
- Insider Trading Activity
- Institutional Ownership Changes
- Short Interest Ratio
- Earnings Surprise (Last Quarter)
- Earnings Estimate Revisions
- Price Target Consensus
- Environmental, Social, and Governance (ESG) Score
- Alternative Data Score (e.g., satellite imagery analysis, web traffic)
For machine learning strategies, Patrick and James O’Shaughnessy might employ:
- Feature Engineering: Transforming raw data into features that better represent the underlying problem to the predictive models, enhancing model accuracy on unseen data.
- Dimensionality Reduction Techniques: Using methods like Principal Component Analysis (PCA) to reduce the number of variables under consideration, focusing on those that most contribute to the variation in data.
- Regularization Techniques: Applying methods such as Lasso (L1 regularization) and Ridge (L2 regularization) to prevent overfitting by penalizing large coefficients in the regression model.
This extensive list of factors, combined with sophisticated machine learning techniques, would enable a comprehensive analysis of stocks from multiple dimensions, aiming to uncover patterns and relationships not immediately apparent through traditional analysis alone."
ETFs cannot do that. Small investors lose their edge the second they buy an ETF. We think we have a better solution for them.
About my 99% comment … I do think only 1% of retail investors should be actively managing portfolios. And “Personalized strategies” for everyone is unsustainable in the long run. If everyone starts saying “I’d like a strategy that benefits from demand for GPUs.” it’s time to sell. Most retail buys high and sells low. Sites that make it look easy with slick interfaces are opportunist, or naive at best.
Lastly, who wants a world where everyone is talking about investing? The world needs more makers, artists, scientists, astronauts. No kid ever says “I want to be like Buffett”
Correct. I read Ed Thorp’s “Beat the Dealer” as a kid in elementary school and said I want to be like Ed Thorp. He then wrote “Beat the Market.” Still trying to live my childhood dreams with the help of a lot of advanced tools and highly trained people.
Marco, can you share clarity on what P123 is offering and your future vision for clients, specifically? I ask because if any current or future ETFs offer no discernible edge, as you suggest, despite our ability to build fully automated strategies like AQR’s ETFs, should we then concentrate on microcap stocks or unconventional, more actively trade and concentrated holdings of stocks with reasonable liquidity? Is this the value proposition you see P123 fulfilling?
I grasp the idea of providing clients with “action,” where retail investors seek to outmaneuver the market, although in reality, it’s often a matter of luck rather than strategy (speak to Citadel Securities who is on the other side of trades for Schwab and Fidelity); these non indexers (ETF or otherwise) like gambling and providing guidance with ex post rationale of their luck. Technology has democratized access even for large investors to manipulate small cap stocks to align with fair value, thanks to marginal costs. Previously, this required human intervention but that is diminishing at an accelerate pace.
I’m attempting to navigate how to utilize this platform moving forward based on your remarks and response. Thank you.
This is what I think right now …
Alpha decays the minute you deploy a large cap system. Your edge is discovered and constantly reduced by large players constantly tweaking things. I haven’t looked at AQR but they don’t seem all that special (they charge a lot too). Basically it’s hard for anyone (regardless of tools) to design an active large cap system with a traditional approach. Hedge funds, that use all sorts of leverage and instruments, are something else entirely (and we don’t have tools for them).
I think the current incarnation of P123 is best suited for small caps and microcaps. Whatever alpha your small-cap system has is sustainable because the other side of the trade is someone similarly sized. Both sides eventually get priced out of small stocks, so the alpha remains for someone else.
For P123 to be successful in large caps we need dynamic systems. But current tools are very time consuming and require lots of know how to use them for that purpose. Or perhaps someone is using P123 in non traditional ways for large caps.
I think ML will fix that. Retraining will be a snap, weights automatically refreshed, etc.
Thanks for your feedback.
I just want to add that most money mangers are perfectly happy with 1%-2% alpha. So in that context P123 is just fine for large caps too.
I think this Reuter’s news story belongs in this discussion of what the future may look like; Neuralink’s first human patient able to control mouse through thinking, Musk says
My take: P123 would have to be quite a bit more radical before it was actually breaking new ground with AI (now or in the future). And I note that Elon Musk had a hand in creating ChatGPT too.
Hmm…..Isn’t Elon Musk the one who said the odds are we are already living in a virally reality (and I guess we chose to have that knowledge erased before we agreed the VF experience)?
Here is a parapharasing of what he said: Code Conference in 2016. During this event, Musk articulated the argument that given the advancements in video game technology, it’s possible that in the future we could create simulations that are indistinguishable from reality. If such simulations were to exist, he posited, it’s conceivable that we might already be living in one, suggesting that the odds we are in “base reality” — meaning the original, un-simulated reality — are one in billions.
WALL-E in the near future with self driving cars (and more) but then… Elon Musk continues to lead this in a real way with his considerable resources? then……? Maybe he paid extra to not forget “base reality” in his VR-world and already knows where we are headed? …would explain a lot…… …… :naw… 
I think I will have ChatGPT write that novel for me…Elon uh… Dusk (as in Dusk of mankind). no mention of real people in a fiction novel (was not thinking of Elon Musk).
Anyone dare to raise a feature request to integrate Neurolink chip to P123 trade module? 
2 Likes
I get that P123 has limited resources and cannot implement every great idea. And I like what P123 has done with AI (a lot). Also, I understand we do not know how this ETF will do.
But someone else is doing what InspectorSector has suggested above. Are the developers of the ETF following InspectorSector's posts or just an obvious idea?
If they are smart they are following his posts. Or the developers are using input from ChatGPT and OpenAI is scraping P123 content.
AI-Powered ETF: Invest Like Warren Buffett with ChatGPT!
So Claude 3 was I joking above or is it possible that ChatGPT got the idea from Inspector sector?
"...what you're describing is certainly possible, though it's important to note that we don't have concrete evidence that this specific scenario is occurring. Let's break down the possibilities:
1. AI models using online content:
-
- Large language models like GPT-3 (which powers ChatGPT) are trained on vast amounts of internet data, which could include public forum discussions.*
-
- It's entirely possible that content from P123 forums, if publicly accessible, could be part of the training data for these models.*
2. OpenAI scraping P123 content:
- It's entirely possible that content from P123 forums, if publicly accessible, could be part of the training data for these models.*
-
- Many AI companies, including OpenAI, use web crawlers to gather training data. If P123 forums are public and not excluded by robots.txt files, they could potentially be included in such data gathering efforts."*
Hmmmm…..still meant as a joke, but…… ChatGPT has already read most of P123 member's ideas?
Jim