GPT-6 Astra Helped Me Jailbreak My Kindle
4 points and 0 comments on Hacker News, linking nirajrajgor.com.
Top now
6 storiesWhat is climbing fastest across every source
Meta6.9xMuse Wonder7.1xOpenAI6.2xAnthropic5.5xGPT-67.9xImage-2.16.8xClaude5.2xMicrosoft6.9xEpic Games6.9xGoogle6.7x
AI is a tool and using it well takes skill
The author responds to Rich Hickey by arguing that large language models function as tools requiring skill, feedback, and judgment to use effectively. ·
OpenAI Fires Three Safety Researchers
OpenAI dismisses three safety researchers, Tomek Korbak, Mikita Balesni, and Jasmine Wang, citing mishandled sensitive company information. ·
autotrust/JEV-35B
A text classification model from autotrust with 183 likes and 6,262 downloads on Hugging Face.
curvetherapistfall/AIDA-64-Extreme
A new repository with 146 stars in its first week.
langchain==1.4.5
LangChain version 1.4.5 allows developers to pass a summarizer into its SummarizationMiddleware component. ·
AI is a tool and using it well takes skill
The author responds to Rich Hickey by arguing that large language models function as tools requiring skill, feedback, and judgment to use effectively. ·
175 points and 158 comments on Hacker News, linking yyhh.org.
AI models may be capable of self-replicating without access to a datacenter
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
Aquafitzdisconnect/Discord-Server-Raider
Aquafitzdisconnect releases Discord Server Raider, an AI-driven tool that automates server management and moderation tasks to improve user experience and resource efficiency. ·
Galaxychiblast/SketchUp-Pro
Galaxychiblast offers a SketchUp Pro desktop software configuration that provides an intuitive interface and advanced modeling capabilities for professional use. ·
unslothai/unsloth Browser Use + Qwen-Image-2.1-Turbo
Unsloth adds browser control to local models and includes Qwen-Image-2.1-Turbo, which generates images in eight steps. ·
autotrust/JEV-35B
A text classification model from autotrust with 183 likes and 6,262 downloads on Hugging Face.
AI is a tool and using it well takes skill
The author responds to Rich Hickey by arguing that large language models function as tools requiring skill, feedback, and judgment to use effectively. ·
175 points and 158 comments on Hacker News, linking yyhh.org.
AI models may be capable of self-replicating without access to a datacenter
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
Aquafitzdisconnect/Discord-Server-Raider
Aquafitzdisconnect releases Discord Server Raider, an AI-driven tool that automates server management and moderation tasks to improve user experience and resource efficiency. ·
Galaxychiblast/SketchUp-Pro
Galaxychiblast offers a SketchUp Pro desktop software configuration that provides an intuitive interface and advanced modeling capabilities for professional use. ·
DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models
Researchers introduce DiffusionART, a reinforcement learning method that fine-tunes diffusion models for image and video generation without requiring a differentiable reward model or exponential moving average tuning. ·
unslothai/unsloth Browser Use + Qwen-Image-2.1-Turbo
Unsloth adds browser control to local models and includes Qwen-Image-2.1-Turbo, which generates images in eight steps. ·
Trending topics today: heat
Table
| Meta | 1.68 |
| Muse Wonder | 1.68 |
| OpenAI | 1.31 |
| Anthropic | 1.2 |
| GPT-6 | 1.17 |
| Image-2.1 | 1.14 |
Heat = summed hotness of today's stories naming the topic.
Models
5 storiesReleases, papers, trending weights and new listings
Top 6 trending on Hugging Face: likes
Table
| Lightricks/LTX-2.5 | 7477 |
| abenzerps/Qwen-Image-2.1-Uncensored-Quantized new | 4095 |
| Cloudflare/clef ▼2 | 2090 |
| google/embeddinggemma-2 | 1932 |
| jialinyyzz/humanizer | 1316 |
| Qwen/Qwen-Image-2.1-Turbo ▲2 | 631 |
PirateBench: score out of 100
Table
| gpt-6.1-sol | 98.9 |
| opus-5-5 | 96.46 |
| gpt-6-sol | 91.6 |
| gpt-6-luna | 65.25 |
| haiku-4-5 | 4.28 |
Agents build our own projects from a brief; every run, failures included: https://cowerx.dev/daily/piratebench/
AI models may be capable of self-replicating without access to a datacenter
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
38 points and 27 comments on Hacker News, linking anthropic.com.
unslothai/unsloth Browser Use + Qwen-Image-2.1-Turbo
Unsloth adds browser control to local models and includes Qwen-Image-2.1-Turbo, which generates images in eight steps. ·
Launch HN: Agent37 (YC F26) – Low-cost agent sandboxes for running agents
81 points and 57 comments on Hacker News, linking news.ycombinator.com.
abenzerps/Qwen-Image-2.1-Uncensored-Quantized
A text to image model from abenzerps with 4,095 likes and 2.2M downloads on Hugging Face.
DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models
Researchers introduce DiffusionART, a reinforcement learning method that fine-tunes diffusion models for image and video generation without requiring a differentiable reward model or exponential moving average tuning. ·
AI models may be capable of self-replicating without access to a datacenter
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
38 points and 27 comments on Hacker News, linking anthropic.com.
DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models
Researchers introduce DiffusionART, a reinforcement learning method that fine-tunes diffusion models for image and video generation without requiring a differentiable reward model or exponential moving average tuning. ·
unslothai/unsloth Browser Use + Qwen-Image-2.1-Turbo
Unsloth adds browser control to local models and includes Qwen-Image-2.1-Turbo, which generates images in eight steps. ·
Launch HN: Agent37 (YC F26) – Low-cost agent sandboxes for running agents
81 points and 57 comments on Hacker News, linking news.ycombinator.com.
Contextual Correction of Parallel Draftsfor Speculative Decoding
Researchers introduce DCOCO, a mechanism that restores dependencies in parallel diffusion-based drafting to improve large language model inference throughput across four verifier backbones and ten benchmarks. ·
X voices
5 storiesX posts other sources are linking to, then named voices on Bluesky and Mastodon
@thepirateface
Decentralize AI: Use Model torrents for your Python scripts
3 points and 0 comments on Hacker News, linking an X post.
@karpathy
The gap in shared understanding of LLM capability is widening
675 points and 984 comments on Hacker News, linking an X post.
@firesidealpha
AI companies are starting to do phones
1 points and 2 comments on Hacker News, linking an X post.
@patrickc
Assorted things I've been mulling over regarding agents in the economy
1 points and 0 comments on Hacker News, linking an X post.
@opencode
Operation Cheepseek Phase 3: $80 of usage for DeepSeek v4.1 Flash
1 points and 0 comments on Hacker News, linking an X post.
@karpathy
The gap in shared understanding of LLM capability is widening
675 points and 984 comments on Hacker News, linking an X post.
@thepirateface
Decentralize AI: Use Model torrents for your Python scripts
3 points and 0 comments on Hacker News, linking an X post.
@selimonder
The Jevons Paradox: Will AI Take Our Jobs?
1 points and 1 comments on Hacker News, linking an X post.
@firesidealpha
AI companies are starting to do phones
1 points and 2 comments on Hacker News, linking an X post.
@satyanadella
Models as Insider Risks in the Super Intelligence Era
7 points and 0 comments on Hacker News, linking an X post.
GitHub
5 storiesNew repos, releases and changelog
New AI repos this week: stars
Table
| wheresryan22/anatomy | 6152 |
| mhtsec/ARTEX | 3558 |
| LosaLosSantos/aurelio-finance | 2841 |
| bherbruck/solidcraft | 1076 |
| rociiu/talorys | 894 |
| ThariqS/ai-newtab | 767 |
Codex CLI installs a persistent, self-updating background daemon by default
OpenAI's Codex CLI version 0.157.0 installs a persistent, self-updating background daemon by default, prompting users to request that this feature become opt-in. ·
421 points and 222 comments on Hacker News, linking github.com.
Aquafitzdisconnect/Discord-Server-Raider
Aquafitzdisconnect releases Discord Server Raider, an AI-driven tool that automates server management and moderation tasks to improve user experience and resource efficiency. ·
Galaxychiblast/SketchUp-Pro
Galaxychiblast offers a SketchUp Pro desktop software configuration that provides an intuitive interface and advanced modeling capabilities for professional use. ·
Janas: Run LLMs larger than RAM by streaming experts from SSD
Janas streams large language model experts from SSDs to run models exceeding available RAM, achieving 37 tokens per second on a 32 GB laptop. ·
unslothai/unsloth Browser Use + Qwen-Image-2.1-Turbo
Unsloth adds browser control to local models and includes Qwen-Image-2.1-Turbo, which generates images in eight steps. ·
Aquafitzdisconnect/Discord-Server-Raider
Aquafitzdisconnect releases Discord Server Raider, an AI-driven tool that automates server management and moderation tasks to improve user experience and resource efficiency. ·
A new repository with 475 stars in its first week.
Codex CLI installs a persistent, self-updating background daemon by default
OpenAI's Codex CLI version 0.157.0 installs a persistent, self-updating background daemon by default, prompting users to request that this feature become opt-in. ·
Galaxychiblast/SketchUp-Pro
Galaxychiblast offers a SketchUp Pro desktop software configuration that provides an intuitive interface and advanced modeling capabilities for professional use. ·
unslothai/unsloth Browser Use + Qwen-Image-2.1-Turbo
Unsloth adds browser control to local models and includes Qwen-Image-2.1-Turbo, which generates images in eight steps. ·
primemitechaos/Imei-Repair
A developer publishes an AI-driven software tool that allows users to manage their device's IMEI number information securely and quickly. ·
Infrastructure
5 storiesData centers, power, chips and what tokens cost
Data-center and energy stories per day
Table
| Wed 07 | 28 |
| Thu 08 | 28 |
| Fri 09 | 30 |
| Sat 10 | 3 |
| Sun 11 | 7 |
| Mon 12 | 42 |
| Tue 13 | 0 |
Stories from DCD, Data Center Knowledge, EIA and Hacker News, by publication day.
Median output price of models added in the last 30 days, $ per million tokens
One point per OpenRouter reading, last 8 days. Data by OpenRouter, CC BY 4.0.
Output price, $ per million tokens (newest models)
Table
| muse-wonder | 20 |
| step-5-preview | 2.7 |
| claude-haiku-5.5 | 0.5 |
| claude-haiku-5.5:batch | 0.25 |
| gemini-nano-banana-2.1 | 7.5 |
| median, all paid models | 2 |
433 paid models listed; median of the 41 added in 30 days: $5.00. Rankings data by OpenRouter, CC BY 4.0.
AI models may be capable of self-replicating without access to a datacenter
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
38 points and 27 comments on Hacker News, linking anthropic.com.
Intel D270 GPU AI Accelerator–Up to 480 GB LPDDR5X
Intel introduces the D270, a data center GPU designed for inference workloads, which supports up to 480 GB of LPDDR5X memory. ·
Janas: Run LLMs larger than RAM by streaming experts from SSD
Janas streams large language model experts from SSDs to run models exceeding available RAM, achieving 37 tokens per second on a 32 GB laptop. ·
Yandex bot traffic down 99% after data center attack
219 points and 113 comments on Hacker News, linking radar.cloudflare.com.
Google is illegaly cutting forest to build data centers
A court orders Google to halt construction at two data center sites because the company failed to complete required environmental impact assessments before clearing forest land. ·
AI models may be capable of self-replicating without access to a datacenter
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
38 points and 27 comments on Hacker News, linking anthropic.com.
Intel D270 GPU AI Accelerator–Up to 480 GB LPDDR5X
Intel introduces the D270, a data center GPU designed for inference workloads, which supports up to 480 GB of LPDDR5X memory. ·
Google is illegaly cutting forest to build data centers
A court orders Google to halt construction at two data center sites because the company failed to complete required environmental impact assessments before clearing forest land. ·
Why Texas Is Making Data Centers Wait
26 points and 39 comments on Hacker News, linking a16z.com.
Yandex bot traffic down 99% after data center attack
219 points and 113 comments on Hacker News, linking radar.cloudflare.com.