Creative LLM use cases - Dan Loeb, Baupost, Altimeter, YCombinator, Ralph Wiggum
Plus: 8 new workflows, 3 new tools and every AI trading arena
Flat Circle tracks creative use cases of LLMs in hedge funds. If you haven’t already, join hundreds of PMs, analysts and engineers reading each week:
Investor LLM workflows
Dan Loeb asked Claude which public companies it might disrupt (Third Point 4Q25 LP Letter)
A simple query into Claude’s chatbot: “Which companies is Anthropic capable of dislocating or disrupting?” yields some fascinating results and was in our view a fruitful source of hedges for our firm.
My take: the subtext here is that models are evolving into the source of conventional wisdom.
AI driven short seller Abelian Analysis analyzed hundreds of Youtube transcripts to assess the pricing environment for CVNA’s used vehicles (Short Report, Github)
Each transcript was analyzed by Claude Sonnet 4 using a prompt designed to separate market conditions from creator mood. This distinction is critical. A dealer complaining about thin margins is telling you margins are compressed — a bearish market signal. A flipper excited about “deals everywhere” is telling you inventory is high and prices are soft — also bearish. The LLM was instructed to ignore emotional spin and extract the underlying market reality across three categorical signals (inventory direction, demand strength, repossession activity), two continuous scores (bullish 0-100, bearish 0-100), and a sensationalism rating (1-10) that we use for quality control.
Investor from Altimeter Capital outlines two LLM workflows, including a PDF example from a “council of LLMs that rigorously debate topics with access to web search” (@_clarktang, thank you @realLigerCub)
An RIA PM shares his deep research prompt (@TedHZhang)
AI-native hedge fund, Minotaur Capital, used “Ralph Wiggum” style iterative research loop to determine that gaming stocks were oversold following the Genie 3 release (Minotaur January Letter)
We immediately spun up a research process using the iterative techniques we described in our December Quarterly. From a 127-word prompt asking for implications on the games industry, our AI system iteratively chose what to explore: value chain analysis, five-year scenarios with falsifiable signposts, unit economics ($/minute cost models), IP and licensing questions, and a winners/losers matrix across engines, platforms, and publishers. Over 50 iterations it built out each section, cited sources, and stress-tested its own conclusions.
Former Baupost investor Dave Plon shares an AI workflow around killing ideas faster: develop a list of non-negotiables (e.g., CEO compensation structure, guidance track record), and have the system eliminate every name in your coverage failing those non-negotiables. (Business Breakdowns, 12m 50s)
Former Maverick / DE Shaw / Citadel PM shares his prompt to analyze the technical setup (@FundamentalEdge, thank you @realLigerCub)
New paper analyzes the impact of LLM tools like ChatGPT on price reactions during earnings calls. Stocks don’t react faster, but at a greater magnitude after a delay due to model latency and transcription availability (Price Discovery Within Earnings Calls, thank you Justina Lee)
Cons to LLM investor workflows
An investor argues LLMs make it harder to build conviction (@evrgn11112231)
I view investment research as akin to the slow LLM training process. It’s not supposed to be fast. The goal is to ingest raw data over long periods of time to train your brain (the ultimate LLM) for instant recall and pattern matching later.
Former credit investor warns on using LLMs for initiation style reports, as they often miss key events that would materially alter the narrative (@BuysideAIReview). My take: the quality of an LLM workflow is only as good as its eval. Before asking for a deep dive, first build a list of key events, transactions, players, etc - then run deep research as a loop until it hits all the required items. See the “Ralph Wiggum” discussion above.
Interesting LLM tools
Former mega fund and credit hedge fund investor reviewing every buyside AI tool (Buyside AI Reviews)
Former hedge fund PM previews an insights feed built from earnings call transcripts (@atelicinvest)
New tool benchmarks the stock impact of short reports / short selling firms (ShortReportImpact)
YCombinator Request for Startup: AI-Native Hedge Funds
YCombinator just published its latest Request for Startups:
…the next Renaissance, Bridgewater, and D.E. Shaw's are going to be built on AI. The biggest funds in the world have been slow to adapt. I worked as a quant researcher at one of these funds, and when I asked compliance to let us use ChatGPT, I didn't even get a response. It made it clear to me that the hedge funds of the future won't just bolt AI onto their existing strategies. They'll use it to come up with entirely new ones. That's where the alpha is.
Now tracking every AI trading arena
AI trading arenas are public experiments where LLMs perform research and trade in a live environment. They are one way to track LLM progress in making investment decisions.
Our new page tracking every public arena is here: AI Trading Arenas
Key takeaways include: (i) the median model always loses money, (ii) newer frontier models outperform the older models, (iii) soon-to-be-released Grok 4.2 is undefeated, (iv) Claude has not yet won any trading arena.
Follow for more investor LLM workflows
If you would like to discuss incorporating LLMs into your research process, reply to this email or reach out via X or LinkedIn.