Skip to content
Log in
Sponsored Artificial Intelligence: A Modern Approach, Global Edition Artificial Intelligence: A Modern Approach, Global Edition

Category

small language model

3,130 papers

#small language model Open access Sep 2026

What Makes It Judder Is Not the Size of the Friction but the Gap Between Two ── With static 30 N, kinetic 20 N and a spring of 1000 N/m, each slip jumps 20.0000 mm ── make the two equal and the amplitude is exactly 0 ── stiffen the spring tenfold and it only falls to a tenth; it does not go ── [Paper 439]

Heavy friction is expected to make motion jerky. This paper shows that the cause is not the size but the gap──make the two frictions equal and the motion is smooth however large they are. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed──the distinction between static...

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

The NEMS Framework: No External Model Selection as a Principle of Foundational Self-Containment

This paper provides a high-level narrative summary of the physics arc (Papers 01–25) of the NEMS suite: semantic closure, self-containment, and their consequences in physics and quantum gravity. The suite introduces a formal "no external model selection" principle, classifies realized systems into a small set of univer...

Nova Spivack · 0 citations
#small language model Open access Sep 2026

The Metal That Reacts Most Readily Is the One That Does Not Corrode ── Aluminium sits at -1.66 V, baser than iron, yet its film is 4 nm, 1/125 of iron's 500 nm ── the ordering is reversed ── what separates them is not the potential but whether the film can cover what lies beneath ── [Paper 441]

A metal that reacts more readily is expected to corrode sooner. This paper shows that the ordering is reversed──the basest metal is protected by the thinnest film. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed──standard electrode potentials, passivity and the Pilli...

Yuuki Yamagishi · 0 citations
#diffusion models Open access Sep 2026

Whether Being Advantageous Matters Is Decided by the Size of the Population ── A mutation with a selection coefficient of 0.01 fixes 19.8013 times as often as a neutral one at N=500, but only 1.2012 times as often at N=10 ── what decides is not s but Ns ── and at s=0 the probability is exactly the neutral value ── [Paper 478]

An advantageous mutation is said to spread. This paper shows that what decides whether the advantage matters is not its size but its product with the population size──where Ns is small, being advantageous does no good. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed─...

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

A Question Worth Answering: What Would a Neural Network Preserve?

The report asks a simple but focused question: if a numerical mechanism attached to a trained neural network had to select exactly one component to preserve, what would it choose, and would that choice predict real functional consequences better than simple baselines? We ran an exploratory series of experiments on smal...

Sohan Poudel · 0 citations
#small language model Open access Sep 2026

The Water Load Grows as the Square, and the Dam Only as the First Power ── Ten times the depth multiplies the resultant thrust by 100 and the overturning moment by 1000 ── yet the required width only doubles ── 18.6339 m at H=50, 37.2678 at H=100, with b/H constant at 0.3727 ── [Paper 454]

A deeper dam is thought to need a disproportionately thicker body. This paper shows that although the force grows as the square and the moment as the cube, the required width grows only as the first power──because the resisting side grows as the cube too. No new theorem or law is claimed. Scope of this paper (scope not...

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

A Question Worth Answering: What Would a Neural Network Preserve?

The report asks a simple but focused question: if a numerical mechanism attached to a trained neural network had to select exactly one component to preserve, what would it choose, and would that choice predict real functional consequences better than simple baselines? We ran an exploratory series of experiments on smal...

Sohan Poudel · 0 citations
#small language model Open access Sep 2026

The Same Proportional Increase Buys Less at the Top ── Raising the compression ratio by 1.3333 gains +0.0531 in efficiency starting from 6 but only +0.0359 starting from 16 ── the ratio is 1.480429, matching the closed form (r_2/r_1)^gamma-1 exactly ── an efficiency of 0.70 needs a ratio of 20.2860 ── [Paper 468]

Raising the compression ratio raises efficiency──so it is said. This paper shows that the same proportional increase buys 1.480429 times less at the top──and that the ratio has a closed form. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed──the theoretical efficiency...

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

Halving the Size Beats Changing the Material Forty Times Over ── Spheres of radius 5 mm and 20 cm have cooling time constants of 348.8333 and 13953.3333 seconds, a ratio of exactly 40.000000000 ── changing the material from water to metal at one size gives only 1.3081 ── what sets the cooling is not specific heat but the surface-to-volume ratio ── [Paper 465]

How fast a thing cools is thought to be set by its material. This paper shows that size acts 40 times harder──changing from water to metal gives only 1.3081. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed──Newton’s law of cooling, the lumped capacitance method and t...

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

LLM Parkinsonism: Executive-Control Failure, Token-Inefficient Persistence, and a Global Executive Control Architecture for Autonomous Language-Model Agents

Large language models (LLMs) can plan, write code, call tools, manipulate software, and execute long-horizon workflows. Yet strong local competence does not guarantee project-level executive control. An agent can make large, useful early advances and then shift toward progressively smaller, lower-value actions—addition...

Dongsheng Xiao · 0 citations
#small language model Open access Sep 2026

An Edge-AI-Driven Crystal Structure Exploration Tool for Secondary School Chemistry: Design and Implementation

Crystal structure is one of the most abstract topics in secondary school chemistry. Textbookspresent three-dimensional lattices as flat diagrams, and physical models cannot be modified, sostudents often memorize conclusions such as "NaCl has a coordination number of 6" withoutunderstanding how structure determines prop...

Kerong He · 0 citations
#diffusion models Open access Sep 2026

What the Waiting Is For Is Diffusion, Not Reaction ── One millimetre takes 2.7778 hours and a hundred takes 3.17 years ── ten times the distance costs 100 times the time ── and ten kelvin multiplies the reaction by 4.0553 while multiplying the diffusion by only 1.5220 ── [Paper 445]

Ageing is taken to be slow because the reaction is slow. This paper shows that sometimes what one waits for is transport──and that transport barely speeds up when heated. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed──the diffusion equation, the square law for diff...

Yuuki Yamagishi · 0 citations

From tech blogs

See all →
MIT News · Artificial Intelligence Oct 15, 2026

AI model decodes the language cells use to communicate

A new computational approach reveals how cells communicate to assemble into tissues. This could accelerate stem cell engineering, improve disease models, and inform therapies.

Microsoft Research Blog Oct 15, 2026

Proof-based general-purpose computing with Magnetar

Zero-knowledge proofs can verify computing results without rerunning workloads or exposing sensitive data. Magnetar helps bring them into the mainstream with support for standard programming languages, major speed gains, and quantum-resistant security. The post Proof-based general-purpose computing with Magnetar appeared first on Microsoft Research.

Microsoft Research Blog Oct 12, 2026

Simulating student learning for better AI tutoring

StudentSim is a research system for evaluating and improving AI tutoring tools before they are tested with learners. It models how individuals may engage with tutors, giving feedback to improve tutor performance across subjects like chess, writing, and mathematics. The post Simulating student learning for better AI tutoring appeared first on Microsoft Research.

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