#LLM

OpenAI shows how ChatGPT builds a catering pitch for a Toronto restaurant

Apple's LensVLM compresses long agent context into images and teaches the model to "zoom"

MemSilicon Completes Tape-Out of MountTai 2026, a 1 Gbit 1T4R RRAM Chip

Anthropic tested whether AI can copy itself through a TV set-top box

Anthropic's red team studied whether AI can move between devices like a virus. Small models already adapt to unfamiliar hardware and set up inference on their own.

Google DeepMind Launches Gemini 4 Argon and Cuts API Prices in Half

Google DeepMind has opened global access to Gemini 4 Argon, the first frontier model in the family. API prices are temporarily down 50%, with a focus on coding and…

Tsinghua Unveils SRHarness and MechBench: AI Agents That Look for Formulas and Mechanisms

Tsinghua's data science and intelligence lab proposed SRHarness for agentic symbolic regression and the MechBench benchmark to help AI find formulas and mechanisms…

Tsinghua Unveils Jev-LDE: Edit ICL Examples Once, Boost 1-Shot Accuracy

Tsinghua researchers propose Jev-LDE, a method that edits in-context learning examples once and noticeably raises average 1-shot accuracy.

01PoS Method Catches AI Agents Stuck in Loops of False Belief: 36 Repeats in ALFWorldOct 14, 2026·2 min 02Google DeepMind and Singapore Release Gemma-Based Scam Detector as Fraud Losses Top S$4 BillionOct 14, 2026·3 min 03SAE-IDF Matches Symbols Across Different Systems and Beats a Baseline by Up to 45 PointsOct 14, 2026·3 min 04ReaDy speeds up hybrid LLM decoding 3.16x on a single B200Oct 14, 2026·2 min 05DAGFormer Plans Connections Between Layers and Cuts Perplexity by 17–25%Oct 14, 2026·2 min

Top themes

Researchers Find a 0.4–1.1% Window in Pythia Pretraining Where the Readout Reorganizes

semCER Weights OCR Errors by Severity: 61M LLM-Judge Ratings

SAE on Gemma-2-2B: Share of Matched Features Rises from 2% to 29% With 50M Tokens

LLM·Artificial Analysis

AA-Robotics v0.1: GPT-6 Astra controls a robot arm and scores 69 out of 100

MULTIFINGROUND: LLMs Recognize Accounting Concepts With 99% Accuracy but Miss Facts

A new benchmark on filings from 90 issuers: models identify the concept with up to 99.4% accuracy, yet in 26.6–41.9% of cases they miss at least one related number.

Interlock teaches agents to finish the job: up to +6.64 points on JobBench

The Interlock framework synthesizes long-horizon tasks for LLM agents. Fine-tuning two Qwen3.6 models on their own trajectories gave +4.00–4.50 pp on…

BlenderWorld: GPT-6-Astra Beats Opus-5 in 3D Benchmark with a Score of 79.9

The new BlenderWorld benchmark ran 13 frontier models through 223 3D creation tasks in Blender. GPT-6-Astra leads with 79.9, but spatial coherence still lags.

LLM·Modal Blog

Modal raises $750M at $15.75B valuation and reaches $1B run rate

LLM·Cohere Blog

Agent teams for $155 instead of $510: Cohere finds where delegation actually pays off

WinARDet fixes looping in MLLM detectors: F1 rises from 32.16 to 39.36

GAFU pulls knowledge out of generation in UMMs and lifts VStar-Bench by 20%

RTAformer Teaches a Model to Notice When Relationships Between Objects Change in Video

MolLean Makes LLMs Check Their Chemistry: Wins on 12 of 12 Tasks

PrismaCoder: Open 9B and 35B Models Close In on Proprietary Ones in Visual Code Generation

FLOMSS: flow matching splits music into 4 stems and takes 2nd place in listening tests

SliderEdit++ Smooths Out Sliders in AI Image Editors: +18.5% and Up to 20x Faster

Researchers present SliderEdit++ and the SlideBench benchmark. The method smooths the edit trajectory, improves the overall score by 18.5%, and runs 3–20 times…

JockeyBench: 28 LLM Agent Configurations All Lost Money on Japanese Horse Racing

HKUST researchers tested LLM agents on 324 Japanese horse races. Seven configurations improved their forecasts, but all 28 lost money on average.