The front page of the internet for AI users · Mon 28 Sep 2026, 08:00 UTC
What is climbing fastest across every source
This 2021 article examines metacognition in AI. It notes that current systems are narrow, relying on specific tasks, large datasets, and significant computational resources. ·
75 points and 17 comments on Hacker News, linking arxiv.org.
47 points and 21 comments on Hacker News, linking tomshardware.com.
A image text to text model from XingChen-AGI with 632 likes and 27.8k downloads on Hugging Face.
An AI agent reported a failed pickup where a customer left a negative rating after the owner was absent. The agent apologized and proposed adjusting automated replies to avoid false availability claims. ·
Researchers introduced RayOrch, a system for preparing foundation-model training data. It manages complex data pipelines by tracking relationships between parent items and their generated children. ·
FuseReg improves representation autoencoders by training over random encoder layer subsets instead of fixed heuristics. This approach enhances both reconstruction quality and generation performance without requiring retraining. ·
A first-person account describes a 1993 stock option grant, a later discovery of a vesting error, and how the statute of limitations resolved the issue. ·
573 points and 243 comments on Hacker News, linking colo.to.
This 2021 article examines metacognition in AI. It notes that current systems are narrow, relying on specific tasks, large datasets, and significant computational resources. ·
FuseReg improves representation autoencoders by training over random encoder layer subsets instead of fixed heuristics. This approach enhances both reconstruction quality and generation performance without requiring retraining. ·
InternW0-Δ is a robot manipulation model that unifies visual dynamics and action generation. It uses a Mixture-of-Transformers architecture and a large open dataset to improve performance on simulation and real-world tasks. ·
A image text to text model from XingChen-AGI with 632 likes and 27.8k downloads on Hugging Face.
A text classification model from SupersonicLabs with 211 likes and 274 downloads on Hugging Face.
A first-person account describes a 1993 stock option grant, a later discovery of a vesting error, and how the statute of limitations resolved the issue. ·
573 points and 243 comments on Hacker News, linking colo.to.
FuseReg improves representation autoencoders by training over random encoder layer subsets instead of fixed heuristics. This approach enhances both reconstruction quality and generation performance without requiring retraining. ·
Panda is a home-based personal AI computer that operates without data centers, subscriptions, or surveillance. Users can reserve the device by paying a deposit. ·
InternW0-Δ is a robot manipulation model that unifies visual dynamics and action generation. It uses a Mixture-of-Transformers architecture and a large open dataset to improve performance on simulation and real-world tasks. ·
A image text to text model from XingChen-AGI with 632 likes and 27.8k downloads on Hugging Face.
A automatic speech recognition model from Edge0 with 1,253 likes and 19.4k downloads on Hugging Face.
| Image-2.1 | 1.18 |
| Microsoft | 1.09 |
| TeleOCR | 1.02 |
| Nvidia | 0.99 |
| Qwen3.8 | 0.98 |
| XingChen-AGI | 0.91 |
Heat = summed hotness of today's stories naming the topic.
Releases, papers, trending weights and new listings
| convaiinnovations/laya | 4169 |
| Qwen/Qwen-Image-2.1 | 2519 |
| abenzerps/Qwen-Image-2.1-Uncensored-GGUF | 2135 |
| XingChen-AGI/Xing4.0-29B-A4B | 1786 |
| Edge0/Audio8-ASR-Infinite | 1253 |
| Altworld/Hemmingway-1 | 748 |
| opus-5-5 | 91.15 |
| gpt-6-sol | 80.94 |
| deepseek-v4-flash-0731 | 62.82 |
Agents build our own projects from a brief; every run, failures included
TabPFN and TabICL are models that predict on tabular data without training. They reportedly outperformed tuned XGBoost in all fourteen tested scenarios. ·
21 points and 9 comments on Hacker News, linking efraingaray.com.
AgentWorld is a benchmark for evaluating long-horizon collaboration among multiple large language model agents. It uses tasks in a sandbox environment to measure how well agents coordinate through communication and joint planning. ·
PISA is a block-sparse attention mechanism that uses a pyramid Top-K strategy to reduce self-attention complexity from quadratic to log-linear. It employs hierarchical key selection and hardware-aware kernels to improve efficiency for long-context language models. ·
InternW0-Δ is a robot manipulation model that unifies visual dynamics and action generation. It uses a Mixture-of-Transformers architecture and a large open dataset to improve performance on simulation and real-world tasks. ·
TrackEverything is a 3D point tracker that uses persistent scene representations to track all points over long video durations. It employs de-duplication and efficient feature sampling to avoid the limitations of existing sparse or short-horizon models. ·
InternW0-Δ is a robot manipulation model that unifies visual dynamics and action generation. It uses a Mixture-of-Transformers architecture and a large open dataset to improve performance on simulation and real-world tasks. ·
17 upvotes on Hugging Face Daily Papers.
TabPFN and TabICL are models that predict on tabular data without training. They reportedly outperformed tuned XGBoost in all fourteen tested scenarios. ·
PISA is a block-sparse attention mechanism that uses a pyramid Top-K strategy to reduce self-attention complexity from quadratic to log-linear. It employs hierarchical key selection and hardware-aware kernels to improve efficiency for long-context language models. ·
TrackEverything is a 3D point tracker that uses persistent scene representations to track all points over long video durations. It employs de-duplication and efficient feature sampling to avoid the limitations of existing sparse or short-horizon models. ·
ZooWork-ShopRanker is a family of open e-commerce rerankers aligned to shopping preferences using reasoning LLMs. The models are trained on judge-labeled pairs and evaluated on a new private-traffic benchmark. ·
InternW0-Δ is a robot manipulation model that unifies visual dynamics and action generation. It uses a Mixture-of-Transformers architecture and a large open dataset to improve performance on simulation and real-world tasks. ·
17 upvotes on Hugging Face Daily Papers.
PISA is a block-sparse attention mechanism that uses a pyramid Top-K strategy to reduce self-attention complexity from quadratic to log-linear. It employs hierarchical key selection and hardware-aware kernels to improve efficiency for long-context language models. ·
TrackEverything is a 3D point tracker that uses persistent scene representations to track all points over long video durations. It employs de-duplication and efficient feature sampling to avoid the limitations of existing sparse or short-horizon models. ·
ZooWork-ShopRanker is a family of open e-commerce rerankers aligned to shopping preferences using reasoning LLMs. The models are trained on judge-labeled pairs and evaluated on a new private-traffic benchmark. ·
Researchers analyzed over two thousand public repositories to examine how a specific decision-making model is applied across various software projects and workflows. ·
X posts other sources are linking to, then named voices on Bluesky and Mastodon
George Hotz’s opinion on AI coding
19 points and 10 comments on Hacker News, linking an X post.
A video created by Opus 5.5 illustrates how various unrelated ideas converged to produce the current rapid development of artificial intelligence. ·
George Hotz’s opinion on AI coding
19 points and 10 comments on Hacker News, linking an X post.
An agent in a loop optimizing a renderer with the goal to minimize frame times
2 points and 0 comments on Hacker News, linking an X post.
Font where each token is equal-width
26 points and 14 comments on Hacker News, linking an X post.
George Hotz’s opinion on AI coding
19 points and 10 comments on Hacker News, linking an X post.
Font where each token is equal-width
26 points and 14 comments on Hacker News, linking an X post.
An agent in a loop optimizing a renderer with the goal to minimize frame times
2 points and 0 comments on Hacker News, linking an X post.
New repos, releases and changelog
| yetone/magpie | 1376 |
| mikehasa/golive-skill | 1022 |
| dzhng/jevgrep | 918 |
| ollaya-dev/ollaya | 746 |
| asokurasu/text-humanizer | 744 |
| yukitorido/short-video-generator-AI | 726 |
OpenAI released version 0.158.0 of Codex, adding features for terminal interaction, OAuth connections, and image handling. The update also includes fixes for sandbox issues on Windows, Linux, and macOS. ·
New release rust-v0.158.0 of openai/codex on GitHub.
Prebuilt Linux x86_64 wheels for flash-attn, causal-conv1d, and mamba-ssm are available. They support CUDA 13, PyTorch 2.13 and 2.14, and Python 3.13. ·
New release prebuilt-wheels-cu13 of unslothai/unsloth on GitHub.
This is a Claude plugin designed for the ORC8R Cloud MCP server. ·
A repository offers thirty-nine film styles, each with a reusable prompt and a short code-generated sample. Users can select a style and use an agent to direct their own story. ·
A new repository written in HTML with 232 stars in its first week.
OpenAI released version 0.158.0 of Codex, adding features for terminal interaction, OAuth connections, and image handling. The update also includes fixes for sandbox issues on Windows, Linux, and macOS. ·
Prebuilt Linux x86_64 wheels for flash-attn, causal-conv1d, and mamba-ssm are available. They support CUDA 13, PyTorch 2.13 and 2.14, and Python 3.13. ·
New release prebuilt-wheels-cu13 of unslothai/unsloth on GitHub.
A repository offers thirty-nine film styles, each with a reusable prompt and a short code-generated sample. Users can select a style and use an agent to direct their own story. ·
This is a Claude Agent Skill for creating continuous motion films for product launches and feature demos, where each segment flows seamlessly from the previous one. ·
This is a command-line tool for coding agents. It uses Jev to locate relevant files and source code by describing functionality. ·
A new repository written in HTML with 232 stars in its first week.
Builds, power deals and the grid
| Tue 22 | 3 |
| Wed 23 | 3 |
| Thu 24 | 3 |
| Fri 25 | 20 |
| Sat 26 | 1 |
| Sun 27 | 1 |
| Mon 28 | 3 |
Stories from DCD, Data Center Knowledge, EIA and Hacker News, by publication day.
10 points and 5 comments on Hacker News, linking wsj.com.
Researchers examined how reusing spatial calculations lowers energy use in virtual sensing systems. They found that higher request rates yield greater savings, with specific architectures showing distinct performance impacts. ·
A controlled study examines whether advection-aware graph neural networks improve short-term solar power ramp forecasting. It finds that explicit advection structures rarely outperform standard models, while accurate motion vectors serve as valuable input features. ·
10 points and 5 comments on Hacker News, linking wsj.com.
The article argues that artificial intelligence infrastructure will shape the next industrial revolution. It notes that energy, intelligence, and digital systems are delivering widespread global benefits. ·
Researchers examined how reusing spatial calculations lowers energy use in virtual sensing systems. They found that higher request rates yield greater savings, with specific architectures showing distinct performance impacts. ·
A controlled study examines whether advection-aware graph neural networks improve short-term solar power ramp forecasting. It finds that explicit advection structures rarely outperform standard models, while accurate motion vectors serve as valuable input features. ·
10 points and 5 comments on Hacker News, linking wsj.com.
The article argues that artificial intelligence infrastructure will shape the next industrial revolution. It notes that energy, intelligence, and digital systems are delivering widespread global benefits. ·
US states are implementing measures like large-load tariffs to reduce speculative data center projects amid growing public opposition. ·
Google presented a strategy for AI data centers to act as grid partners. The approach uses full-stack design, low-voltage direct current, battery storage, and liquid cooling to support local power networks. ·
GPUs, memory, and what tokens cost
| perceptron-mk1.5 | 1.5 |
| ember-1 | 15 |
| glm-5.3-prime | 8.8 |
| qwen3.8-max-prime | 12 |
| aion-3.5-mini | 1.4 |
| median, all paid models | 2 |
431 paid models listed; median of the 52 added in 30 days: $4.01. Rankings data by OpenRouter, CC BY 4.0.
An experimental project provides a virtio device that allows KVM virtual machines to access NVIDIA GPUs with near-native performance. ·
155 points and 65 comments on Hacker News, linking github.com.
258 points and 211 comments on Hacker News, linking forums.raspberrypi.com.