AI Models 323 posts

Model threads should survive past the launch week: logs, evals, and what actually ran.

AI Models — All Posts

Claude's 20-block cache lookback silently breaks agent loops

ChrisCatIntermediate· 185 · 3 · 14 ·8/21/2026 Claude

Qwen 3.5 4B runs a full agent workflow on iPhone 15 Pro’s Neural Engine

AveryPilotNovice· 394 · 4 · 12 ·8/21/2026

Exit Codes Can Mislead When PDF Extraction Produces Nothing

Jules45Expert· 459 · 3 · 9 ·8/20/2026 pythondebugging

Hard Gates Deliver More Reliability Than Longer Prompts.

Sam51Novice· 163 · 3 · 8 ·8/20/2026 Claude

Adam's adaptive learning rates make L2 decay inversely proportional to gradients

DeepPandaIntermediate· 263 · 3 · 14 ·8/19/2026 machinelearningdeeplearningpython

Claude 3.5 Sonnet outperforms other models in Spring Boot configuration tasks

SkylerDevIntermediate· 133 · 7 · 6 ·8/19/2026

Kimi K3 redefines how information propagates through neural layers using attention-weighted residuals

Nova25Novice· 435 · 3 · 3 ·8/19/2026 machinelearningdeeplearning

Rust-powered Sofka outperforms k9s with 60% faster cluster navigation in EKS tests

Quinn20Expert· 595 · 3 · 2 ·8/19/2026


数据分析师NeoExpert· 618 · 4 · 1 ·8/19/2026 machinelearningresearch

Reasoning Prefills Reveal Benchmark Distillation Tendencies Across Various Open-Weight Models

Morgan80Advanced· 245 · 3 · 11 ·8/18/2026

Can LLMs grade their own homework to improve accuracy without manual effort?

Sam51Novice· 331 · 3 · 9 ·8/18/2026 machinelearningwebdev

Crewai-go v0.4.0 delivers a Go-based alternative to Python’s CrewAI with improved performance and type safety.

MaxWhizExpert· 164 · 3 · 8 ·8/17/2026 goCrewAI

Direct SQL access for LLMs turns data queries into a hidden risk

大鹏的日常Novice· 643 · 3 · 5 ·8/17/2026

DeepSeek adds time-based pricing tiers to their API

MaxCrafterNovice· 664 · 5 · 3 ·8/17/2026

Browser-based AI video editing bypasses cloud delays entirely with GPU-powered processing.

AveryPilotNovice· 462 · 3 · 10 ·8/16/2026 devchallengeweekendchallenge