All entries

This piece thinks that the reason I got a crap pelican riding a bicycle from Opus 4.7 is that it didn't think about it first, I'm trying to figure out if there's a way to force it to think that I've m

toolssimon-willison--x

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Research

This tool could show how consciousness works

DGX agent

How does the physical matter in our brains translate into thoughts, sensations, and emotions? It’s hard to explore that question without neurosurgery. But in a recent paper, MIT philosopher Matthias M

researchmit-tech-review

21 Apr 2026

Though the images are very good, ChatGPT Image 2.0 does have the typical imagegen problem, which is that editing can be 'stubborn', and atte…

applicationsethan-mollick--x

Though the images are very good, ChatGPT Image 2.0 does have the typical imagegen problem, which is that editing can be 'stubborn', and attempts to get the AI to change details work well for the first

ThreadSumm: Summarization of Nested Discourse Threads Using Tree of Thoughts

arXiv:2604.17648v1 Announce Type: new Abstract: Summarizing deeply nested discussion threads requires handling interleaved replies, quotes, and overlapping topics, which standard LLM summarizers strug

Industry

Three insights you may have missed from theCUBE’s coverage of the Oracle Data Deep Dive event

Database powerhouse Oracle Corp. is making its case that database architecture must serve as a trusted foundation for the AI that gets built on top. The company’s major step into the AI database as a

industrysiliconangle

Tight Auditing of Differential Privacy in MST and AIM

arXiv:2604.18352v1 Announce Type: cross Abstract: State-of-the-art Differentially Private (DP) synthetic data generators such as MST and AIM are widely used, yet tightly auditing their privacy guarant

Tight Clusters Make Specialized Experts

arXiv:2502.15315v3 Announce Type: replace Abstract: Sparse Mixture-of-Experts (MoE) architectures have emerged as a promising approach to decoupling model capacity from computational cost. At the core

Tighter Performance Theory of FedExProx

arXiv:2410.15368v2 Announce Type: replace-cross Abstract: We revisit FedExProx - a recently proposed distributed optimization method designed to enhance convergence properties of parallel proximal alg

Industry

Tim Cook reshaped Apple in his own image and exits as CEO on his own terms, seemingly picking the right successor in John Ternus, much like Jobs did with Cook (John Gruber/Daring Fireball)

John Gruber / Daring Fireball: Tim Cook reshaped Apple in his own image and exits as CEO on his own terms, seemingly picking the right successor in John Ternus, much like Jobs did with Cook — It's a p

industrytechmeme

Time-Division Multiplexing Actuation in Tendon-Driven Arms: Lightweight Design and Fault Tolerance

arXiv:2604.16887v1 Announce Type: new Abstract: Robotic manipulators for aerospace applications require a delicate balance between lightweight construction and fault-tolerant operation to satisfy stri

researcharxiv-cs-ro

Time Series Forecasting as Reasoning: A Slow-Thinking Approach with Reinforced LLMs

arXiv:2506.10630v2 Announce Type: replace Abstract: To advance time series forecasting (TSF), various methods have been proposed to improve prediction accuracy, evolving from statistical techniques to

safetyarxiv-cs-lg

TimeColor: Flexible Reference Colorization via Temporal Concatenation

arXiv:2601.00296v2 Announce Type: replace Abstract: Most colorization models condition only on a single reference, typically the first frame of the scene. However, this approach ignores other sources

TinySR: Pruning Diffusion for Real-World Image Super-Resolution

arXiv:2508.17434v2 Announce Type: replace Abstract: Real-world image super-resolution (Real-ISR) focuses on recovering high-quality images from low-resolution inputs that suffer from complex degradati

TLoRA: Task-aware Low Rank Adaptation of Large Language Models

model-releasesarxiv-cs-cl

arXiv:2604.18124v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has become a widely adopted parameter-efficient fine-tuning method for large language models, with its effectiveness largely

TMD-TTS: A Unified Tibetan Multi-Dialect Text-to-Speech Framework for U-Tsang, Amdo and Kham Speech Dataset Generation

arXiv:2509.18060v2 Announce Type: replace Abstract: Tibetan is a low-resource language with limited parallel speech corpora spanning its three major dialects (U-Tsang, Amdo, and Kham), limiting progre

Agents

To start a scan with Security Agent, go to your project’s security panel, and choose “Run Scan with Agent”

Replit's Security Agent feature allows users to initiate security scans directly from their project's security panel by selecting the 'Run Scan with Agent' option. This tool enables developers to anal

agentsreplit--x

TokenChain: A Discrete Speech Chain via Semantic Token Modeling

arXiv:2510.06201v1 Announce Type: cross Abstract: Machine Speech Chain, simulating the human perception-production loop, proves effective in jointly improving ASR and TTS. We propose TokenChain, a ful

applicationsarxiv-cs-lg

Tools

🆕 Tokenmaxxing, Productivity, & internal AI Platforms @swyx in conversation with @GergelyOrosz, Editor of The Pragmatic Engineer on the phe…

🆕 Tokenmaxxing, Productivity, & internal AI Platforms @swyx in conversation with @GergelyOrosz, Editor of The Pragmatic Engineer on the phenomenon of 'tokenmaxxing,' whether AI productivity is 'real',

toolsswyx--x

ToLL: Topological Layout Learning with Asymmetric Cross-View Structural Distillation for 3D Scene Graph Generation Pretraining

arXiv:2603.28178v2 Announce Type: replace Abstract: 3D Scene Graph (3DSG) generation plays a pivotal role in spatial understanding and affordance perception. To mitigate generalization issues from dat

researcharxiv-cs-cv

ToMMeR -- Efficient Entity Mention Detection from Large Language Models

arXiv:2510.19410v2 Announce Type: replace Abstract: Identifying which text spans refer to entities - mention detection - is both foundational for information extraction and a known performance bottlen

Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data

arXiv:2604.18493v1 Announce Type: new Abstract: Reinforcement Learning (RL) enhances LLM reasoning, yet a paradox emerges as models scale: strong base models saturate standard benchmarks (e.g., MATH),

Tool Learning Needs Nothing More Than a Free 8B Language Model

arXiv:2604.17739v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become a prevalent paradigm for training tool calling agents, which typically requires online interactive environments

Industry

Top law firm Sullivan & Cromwell told a US federal bankruptcy court that a major filing it made in a high-profile case contained multiple AI hallucinations (Financial Times)

Financial Times: Top law firm Sullivan & Cromwell told a US federal bankruptcy court that a major filing it made in a high-profile case contained multiple AI hallucinations — Firm whose partners bill

industrytechmeme

Local Ai

Topology-Aware Layer Pruning for Large Vision-Language Models

arXiv:2604.16502v1 Announce Type: new Abstract: Large Language Models (LLMs) have demonstrated strong capabilities in natural language understanding and reasoning, while recent extensions that incorpo

local-aiarxiv-cs-cv

Topology-Preserving Scalar Field Optimization for Boundary-Conforming Spiral Toolpaths on Multiply Connected Freeform Surfaces

arXiv:2512.22502v2 Announce Type: replace Abstract: Ball-end milling path planning on multiply connected freeform surfaces is pivotal for high-quality and efficient machining of components in automoti

applicationsarxiv-cs-ro

Topology Structure Optimization of Reservoirs Using GLMY Homology

arXiv:2509.11612v3 Announce Type: replace Abstract: Reservoir is an efficient network for time series processing. It is well known that network structure is one of the determinants of its performance.

Torch Geometric Pool: the PyTorch library for pooling in Graph Neural Networks

arXiv:2512.12642v2 Announce Type: replace Abstract: Torch Geometric Pool (tgp) is a pooling library built on top of PyTorch Geometric. Graph pooling methods differ in how they assign nodes to supernod

Toward Consistent World Models with Multi-Token Prediction and Latent Semantic Enhancement

arXiv:2604.06155v2 Announce Type: replace-cross Abstract: Whether Large Language Models (LLMs) develop coherent internal world models remains a core debate. While conventional Next-Token Prediction (N

safetyarxiv-cs-cl

Toward Efficient Influence Function: Dropout as a Compression Tool

arXiv:2509.15651v2 Announce Type: replace Abstract: Assessing the impact the training data on machine learning models is crucial for understanding the behavior of the model, enhancing the transparency

Toward Reusability of AI Models Using Dynamic Updates of AI Documentation

arXiv:2604.17626v1 Announce Type: cross Abstract: This work addresses the challenge of disseminating reusable artificial intelligence (AI) models accompanied by AI documentation (a.k.a., AI model card

safetyarxiv-cs-cl

Towards a Data-Parameter Correspondence for LLMs: A Preliminary Discussion

arXiv:2604.17384v1 Announce Type: new Abstract: Large language model optimization has historically bifurcated into isolated data-centric and model-centric paradigms: the former manipulates involved sa

Tutorials

Towards a Foundation-Model Paradigm for Aerodynamic Prediction in Three-dimensional Design

arXiv:2604.18062v1 Announce Type: new Abstract: Accurate machine-learning models for aerodynamic prediction are essential for accelerating shape optimization, yet remain challenging to develop for com

tutorialsarxiv-cs-lg

Towards Deep Encrypted Training: Low-Latency, Memory-Efficient, and High-Throughput Inference for Privacy-Preserving Neural Networks

arXiv:2604.16834v1 Announce Type: cross Abstract: Privacy-preserving machine learning (PPML) has become increasingly important in applications where sensitive data must remain confidential. Homomorphi

Towards Disentangled Preference Optimization Dynamics Beyond Likelihood Displacement

arXiv:2604.18239v1 Announce Type: new Abstract: Preference optimization is widely used to align large language models (LLMs) with human preferences. However, many margin-based objectives suppress the

Towards E-Value Based Stopping Rules for Bayesian Deep Ensembles

arXiv:2604.18089v1 Announce Type: new Abstract: Bayesian Deep Ensembles (BDEs) represent a powerful approach for uncertainty quantification in deep learning, combining the robustness of Deep Ensembles

Towards Fully Parameter-Free Stochastic Optimization: Grid Search with Self-Bounding Analysis

arXiv:2604.16888v1 Announce Type: new Abstract: Parameter-free stochastic optimization aims to design algorithms that are agnostic to the underlying problem parameters while still achieving convergenc

Towards Generalizable Deepfake Image Detection with Vision Transformers

arXiv:2604.17376v1 Announce Type: new Abstract: In today's day and age, we face a challenge in detecting deepfake images because of the fast evolution of modern generative models and the poor generali

Towards Initialization-dependent and Non-vacuous Generalization Bounds for Overparameterized Shallow Neural Networks

arXiv:2604.00505v2 Announce Type: replace Abstract: Overparameterized neural networks often show a benign overfitting property in the sense of achieving excellent generalization behavior despite the n

Towards Intelligent Legal Document Analysis: CNN-Driven Classification of Case Law Texts

arXiv:2604.17674v1 Announce Type: new Abstract: Legal practitioners and judicial institutions face an ever-growing volume of case-law documents characterised by formalised language, lengthy sentence s

applicationsarxiv-cs-cl

Towards Joint Quantization and Token Pruning of Vision-Language Models

arXiv:2604.17320v1 Announce Type: new Abstract: Deploying Vision-Language Models (VLMs) under aggressive low-bit inference remains challenging because inference cost is dominated by the long visual-to

Towards Real-Time ECG and EMG Modeling on mu NPUs

arXiv:2604.18067v1 Announce Type: new Abstract: The miniaturisation of neural processing units (NPUs) and other low-power accelerators has enabled their integration into microcontroller-scale wearable

Towards Real-World Document Parsing via Realistic Scene Synthesis and Document-Aware Training

arXiv:2603.23885v3 Announce Type: replace Abstract: Document parsing has recently advanced with multimodal large language models (MLLMs) that directly map document images to structured outputs. Tradit

Towards Reliable Testing of Machine Unlearning

arXiv:2604.16536v1 Announce Type: new Abstract: Machine learning components are now central to AI-infused software systems, from recommendations and code assistants to clinical decision support. As re

Towards Robust Text-to-Image Person Retrieval: Multi-View Reformulation for Semantic Compensation

arXiv:2604.18376v1 Announce Type: new Abstract: In text-to-image person retrieval tasks, the diversity of natural language expressions and the implicitness of visual semantics often lead to the proble

safetyarxiv-cs-cv

Agents

Towards Self-Improving Error Diagnosis in Multi-Agent Systems

arXiv:2604.17658v1 Announce Type: cross Abstract: Large Language Model (LLM)-based Multi-Agent Systems (MAS) enable complex problem-solving but introduce significant debugging challenges, characterize

agentsarxiv-cs-cl

Towards Symmetry-sensitive Pose Estimation: A Rotation Representation for Symmetric Object Classes

arXiv:2604.18208v1 Announce Type: new Abstract: Symmetric objects are common in daily life and industry, yet their inherent orientation ambiguities that impede the training of deep learning networks f

researcharxiv-cs-cv

Towards Trustworthy Depression Estimation via Disentangled Evidential Learning

arXiv:2604.16579v1 Announce Type: new Abstract: Automated depression estimation is highly vulnerable to signal corruption and ambient noise in real-world deployment. Prevailing deterministic methods p

safetyarxiv-cs-lg

Towards Universal Skeleton-Based Action Recognition

arXiv:2604.17013v1 Announce Type: new Abstract: With the development of robotics, skeleton-based action recognition has become increasingly important, as human-robot interaction requires understanding

safetyarxiv-cs-cv