polynoamial
polynoamial publishes concise technical-skeptical analysis of frontier AI model trajectories and their tradable implications—especially the evolving compute economics of inference and test-time optimization.
Past bets that played out
Repeatedly argues that future frontier models (examples: “GPT-5.5 Pro”) could reach previously hard results through improved steering/scaffolding plus substantially more test-time compute. The practical takeaway: rising demand for inference/test-time compute and the AI infrastructure stack (GPUs, networking, memory, foundry capacity, data centers/cloud) is a key tradable theme.
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
What this channel is watching now
Focus on the shifting relationship between model capability and test-time compute (the “intelligence vs test-time compute (TTC)” curve). Top tickers of interest include NVDA, MSFT, AMZN, AVGO, ANET, TSM, MU, and ASML — companies tied to AI compute, semiconductors, and cloud/datacenter infrastructure.
Latest videos and market context
No videos available. Posts are shortform X threads and technical commentaries describing speculative model pathways and compute implications.
@wtgowers I can believable that GPT-5.5 Pro solves it with steering and/or scaffolding + tons of test-time compute. T...
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
@yoavgo @littmath I feel like this is a complicated point so I want to put together a longer post explaining my views.
The source contains no substantive market, macro, sector, or company-specific claims—only an intent to write a longer post later.
@yoavgo @littmath Can you clarify the question?
The source contains only a request to clarify a question and provides no market, macro, company, sector, or ticker-relevant information.
@yoavgo @littmath It's hard to draw a line... we're talking about log scale, so at some point it becomes completely u...
Commentary suggests that future frontier models (e.g., “GPT-5.5 Pro”) could require dramatically higher inference/training cost ("1000x"), implying AI compute intensity may rise nonlinearly and become economically unrealistic without steering/optimization. This is a qualitative, speculative point with no concrete company/news catalyst.
Proof-backed call history
polynoamial has produced 8 recommendations evaluated to date with an average return of 40.3169% and a win rate of 87.5%. Coverage centers on AI compute intensity and infrastructure exposure.
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
Post speculates that future frontier models (e.g., “GPT-5.5 Pro”) could achieve previously hard results via better steering/scaffolding and much more test-time compute, implying the “intelligence vs test-time compute (TTC)” curve shifts left (tasks become easier/cheaper to solve). Tradable implication: rising demand for inference/test-time compute and associated AI infrastructure (GPUs, networking, memory, foundry capacity, data centers/cloud).
About this channel
polynoamial is an X-based analyst who analyzes frontier language models and their economic implications. Work emphasizes qualitative scenarios—how steering, scaffolding, and large amounts of test-time compute could change which tasks are solvable and how that translates into demand for GPUs, networking, memory, foundry capacity, and data-center/cloud services.
@polynoamial
Most recognized assets
Unlock the full track record
Follow @polynoamial on X for ongoing shortform analysis of frontier model development and tradable implications across the AI infrastructure stack.