Skip to content

Category

small language model

2,897 papers

#small language model Open access Sep 2026

Round Numbers Do Not Attract Price: Levels Merely Shifted by 25 Pips Are Touched More Often in All Eight Cells, and the Close Sits 0.2186 Pips Farther from a Round Number Than Chance Would Put It [F011]

「キリのいい数字は意識される」「大台には注文が溜まる」という言い方がある。だが「キリ番に触れた」という観察だけでは何も言えない。キリ番は 50 pips ごとに必ずあるので、必ずどれかが近くにあり、近ければ触れる。決着させるには、同じ間隔で意味の無い水準を置けばよい。キリ番を 25 pips ずらすと、間隔は 50 pips のまま、キリのよさ...

Yuuki Yamagishi · 0 citations
#small language model Open access Sep 2026

The Distance to the Level Ahead Moves the Time-Outs, Not the Win Rate — The Unconditional Win Rate Moves 9.97 Points, of Which the Time-Outs Account for 27.17 While the Win Rate Among Decided Trades Moves Only 2.46 [F028]

「利確の手前に抵抗があるから、そこまで届きにくい」という言い方がある。だが「届きにくい」には二つの行き先がある。反対の柵へ行ったか、どちらにも届かなかったか(時間切れ)である。前者なら勝率が落ち、後者なら決着しないだけである。そして柵が対称でコストが無ければ、決着した中での勝率は任意停止定理により 0.5 である。水準がそこ...

Yuuki Yamagishi · 0 citations
#small language model Open access Sep 2026

A Naive t Value Inflates with the Number of Observations per Day: Changing the Bar Size Moves the Count by a Factor of 11.9751, and Even a Synthetic Series with No Edge Exceeds 2 in 0.5437 of Five-Minute Cells [F014]

手法の成績に t 値が添えられていることがある。その多くは、標準偏差を件数の平方根で割ったものを分母にした素朴な t である。だが同じ日の観測は独立でない。同じ相場の同じ動きを、何度も数えているからである。一日あたり m 件、日の中の相関を ρ とすると、標準誤差の比は √(1 + (m − 1)ρ) になる。ρ が 0 なら膨らまず、1 なら √m 倍にな...

Yuuki Yamagishi · 0 citations
#small language model Open access Sep 2026

A Stopping Rule Changes the Shape of the Profit and Loss, Not the Direction — Across Twenty Rules the Win Rate Runs from 0.0562 to 0.9443, and a Synthetic Random Walk with Zero Directional Information Moves Just as Much (Correlation +0.9922) [F026]

「損小利大にする」「建値に逃がす」「トレールで伸ばす」。どれもいつ切るかの規則であって、どちらへ建てるかの規則ではない。だから方向の情報は原理的に増えない。では停止則を変えると何が動くのか。本稿の分離子はいつ切るかであり、入口・データ・期間・銘柄をすべて固定したまま、停止則だけを二十通りに変える。方向とそれ以外を分ける...

Yuuki Yamagishi · 0 citations
#small language model Open access Sep 2026

Aligning Timeframes Only Reduces the Count by as Much as You Aligned: The Count Falls to 0.3683 While the Directional Information Does Not Rise, and the Apparent Significance Is the Overlap — Removing It Drops the Median Daily t from 2.4643 to 0.5923 [F010]

上位の足の向きにそろえてから入る、三本の移動平均が順に並ぶのを待つ、という作法がある。条件を足すほど精度が上がると言われる。だが条件を足せば件数は必ず減る。これは市場について何も言っておらず、掛け算の性質である。そして件数が減れば勝率は動きやすくなる。問うべきは、増えるのは情報なのか、減るのは件数だけなのか、である。...

Yuuki Yamagishi · 0 citations
#small language model Open access Sep 2026

A general convergence theorem for nearest-element recurrences

Let b(n) be a strictly increasing sequence of positive integers satisfyinglim_{n→∞} b(n+1)/b(n) = 1. Define a recurrence by taking a_n to be the element of b nearest to a_{n-1} + a_{n-2}, with ties broken by taking the smaller element. We prove that for any initial values a_1, a_2 in b, lim_{n→∞} a_n / a_{n-1} = φ := (...

Yixiang Wang · 0 citations
#small language model Open access Sep 2026

NS-Filtered Fine-Tuning: 86.8% Compute Reduction with Equal or Better Benchmark Performance

Modern language model training applies a uniform gradient update to every token, regardless of its information content. We introduce NS-weighted training, which filters training examples by their Shannon information. On a 50MB sample of the C4 web corpus, unigram NS-data filtering matches a 1000-step standard training...

Orders Of Magnitude LLC · 0 citations
#small language model Open access Sep 2026

Scalable Oversight Across Generative Visual AI: Toward Visual Storytelling for Everyone

Generative visual AI has advanced by scaling data and compute, but its next bottleneck is oversight: the expert signals that evaluate, reward, and teach models what ``good'' looks like. Providing such oversight is increasingly difficult because foundation vision-language models now match or surpass most humans at the s...

Zhiqiu Lin · 0 citations
#small language model Open access Sep 2026

Inherited Weights: Measuring the Tokenization Premium and the Sovereignty Cost of Language Models for the Global South

Most language models deployed outside the United States and China are fine-tuned descendants of a small number of pretrained families. We argue that fine-tuning leaves the decisive properties of a model inherited from its base: vocabulary, knowledge, cultural alignment, memorized data and license. We contribute (i) an...

Paulo Câmara · 0 citations
#large language models Dataset Open access Sep 2026

Research Gap Discovery Dataset: LLM-Extracted Limitations and Research Gaps from 3,326 Scientific Papers

This dataset contains 3,326 recent arXiv computer science papers (first versions dated 16 March to 17 July 2026). For each paper, a large language model read the title and abstract and wrote one sentence each on the paper's main limitation, an open research gap for future work, and why that gap matters. It was built to...

Mounata Saha · 0 citations
#small language model Open access Sep 2026

The Model Is a Dependency: Provenance and Error Detection in Financial Data Extraction

This work was produced through an AI-assisted workflow directed and reviewed by the author. This study separates identifying a model run from checking whether its numeric answer is correct. It evaluates a membership-only source check and top-20 truncated token entropy on a small water-table task, then examines extracti...

N Milton · 0 citations
#small language model Preprint Sep 2026

Tipping Points in LLM-Based Multi-Agent Systems: Stance on Climate Change Action

The findings to date suggest that significant abrupt changes in climate-change stance do occur in this simple model, and the need to prevent LLM biases from interfering with the conversational dynamics and, more generally, to maintain agent personality and episodic memories of interactions in the face of such biases is...

A. Altunyan, Shimon Edelman · 0 citations

From tech blogs

See all →

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.