Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-07-16 · high
Yiming Ma and Xinyu Chen say the fix generalizes past finance to any sensor or telemetry model forced to bin continuous signals into tokens.
Why now: No benchmark comparisons against tokenized baselines exist yet, so the claim rests on the paper alone.
Yiming Ma and Xinyu Chen built VAIOM, a transformer that keeps financial data continuous instead of tokenizing it first, per a preprint on arXiv.
Every team feeding continuous signals, market prices, sensor readings, telemetry, into a decoder-only model has been binning that data into discrete tokens first. That binning throws away information the model never gets back.
Financial data makes the mismatch obvious: continuous, heterogeneous, and noisy all at once, per the paper.
VAIOM's fix is architectural. It takes continuous inputs directly. Discrete values only show up on the output side, skipping the tokenization step most financial transformer work depends on.
The paper frames this as a finance problem. But the fix isn't finance-specific. Any team feeding sensor or telemetry data into a next-token model has been carrying the same workaround, forced to discretize because the architecture demanded it. VAIOM is a case for dropping that constraint entirely.
Tokenization in time-series transformers was never a technical requirement, just a workaround borrowed from language modeling, and VAIOM is the argument for dropping it. No one has published benchmark results pitting VAIOM against tokenized baselines yet. Whether it beats the standard approach on real market data is the open question worth tracking.
Each link below shares sources, entities, or timing with this story.
Same source domain / Shared topic / Downstream implication
Reported by the same outlet (arxiv.org); overlapping topics (assume, between); traces where this leads (implication).
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (been, model); pushes against this story (but).
Reported by the same outlet (arxiv.org); overlapping topics (data, model); pushes against this story (against).
Reported by the same outlet (arxiv.org); overlapping topics (between, model); pushes against this story (vs).
Reported by the same outlet (arxiv.org); overlapping topics (architecture, data); pushes against this story (but).
Reported by the same outlet (arxiv.org); overlapping topics (architecture, model); pushes against this story (versus).
Reported by the same outlet (arxiv.org); overlapping topics (between, data); pushes against this story (but).
Same source domain / Shared topic
Reported by the same outlet (arxiv.org); overlapping topics (architecture, between, model).