The Little Book of Generative AI Foundations: An Intuitive Mathematical Primer
151 points and 18 comments on Hacker News, linking arxiv.org.
Top now
6 storiesWhat is climbing fastest across every source
Anthropic6.6xGoogle8.0xOpenAI7.3xQwen 3.8 27B7.0xLLMs7.0xAstra6.8xGPT-6 Astra7.0xRSI9.9xQwen6.5xRailway8.4x
US tech giants' hidden AI debt swells 90% to $3.2T in 6 months
124 points and 48 comments on Hacker News, linking asia.nikkei.com.
Dependabot version updates support Agent Package Manager
Dependabot version updates support Agent Package Manager. GitHub Blog ·
openai/openai-python v3.44.0
OpenAI releases version 3.44.0 of its Python library, adding workload identity providers and federation rules to the API. ·
openai/openai-python v3.43.0
OpenAI releases version 3.43.0 of its Python library, which expands Live Responses delegation tools and adds WebSocket lane overflow isolation. ·
Qwen 3.8 27B on my laptop can interpret model weights, too
A paper by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab describes weight-space programming, where models like GPT-6 Astra read and write neural network weights. ·
454 points and 142 comments on Hacker News, linking simonwillison.net.
The AI Boom May Be Sowing Seeds of Another Financial Crisis
45 points and 37 comments on Hacker News, linking bloomberg.com.
unslothai/unsloth Glass Theme, Personalize & Faster Generation
Unsloth releases an update featuring glass themes, faster generation, multi-GPU training, and CPU training options. ·
ollama/ollama v0.40.14
Ollama version 0.40.14 updates Claude Code token limits, fixes API chat decoding, and improves MLX sampling speed. ·
Advances in Agentic Computer Graphics
A report introduces a unified taxonomy for agentic systems in computer graphics and an automated literature analysis tool to track rapid research growth. ·
Wave Forcing: Towards Speed-of-Light Streaming Video Generation
Wave Forcing enables real-time long-video generation by using chunk-wise causal attention and staggered denoising across model replicas to achieve one output chunk per forward pass. ·
Qwen 3.8 27B on my laptop can interpret model weights, too
A paper by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab describes weight-space programming, where models like GPT-6 Astra read and write neural network weights. ·
454 points and 142 comments on Hacker News, linking simonwillison.net.
Postman silently auto-installs AI plugins and attempts to share all API data
Postman silently auto-installs AI plugins and attempts to share all API data with users. ·
unslothai/unsloth Glass Theme, Personalize & Faster Generation
Unsloth releases an update featuring glass themes, faster generation, multi-GPU training, and CPU training options. ·
anthropics/claude-code v2.1.317
Anthropic releases Claude Code version 2.1.317, adding scheduling warnings, email checks, and a new cost report API. ·
Advances in Agentic Computer Graphics
A report introduces a unified taxonomy for agentic systems in computer graphics and an automated literature analysis tool to track rapid research growth. ·
Wave Forcing: Towards Speed-of-Light Streaming Video Generation
Wave Forcing enables real-time long-video generation by using chunk-wise causal attention and staggered denoising across model replicas to achieve one output chunk per forward pass. ·
Trending topics today: heat
Table
| Anthropic | 3.5 |
| 3.11 | |
| OpenAI | 2.88 |
| Qwen 3.8 27B | 1.83 |
| LLMs | 1.19 |
| Astra | 1.17 |
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 ▲1 | 10485 |
| deepseek-ai/DeepSeek-V4.1-Flash-1017 ▼1 | 4848 |
| mistralai/Mistral-Large-4-1T-A52B | 1543 |
| LiquidAI/LFM3-1.7B | 1399 |
| nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B ▲1 | 737 |
| speridlabs/idol ▲1 | 509 |
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/
Qwen 3.8 27B on my laptop can interpret model weights, too
A paper by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab describes weight-space programming, where models like GPT-6 Astra read and write neural network weights. ·
454 points and 142 comments on Hacker News, linking simonwillison.net.
OpenAI seems to be silently re-enabling model training on Codex chats
418 points and 132 comments on Hacker News, linking news.ycombinator.com.
The 'Astra can read and write weights to improve Qwen' post is misleading
A post claims the Astra model reads and writes weights to improve Qwen, but experiments use different techniques. ·
Advances in Agentic Computer Graphics
A report introduces a unified taxonomy for agentic systems in computer graphics and an automated literature analysis tool to track rapid research growth. ·
Wave Forcing: Towards Speed-of-Light Streaming Video Generation
Wave Forcing enables real-time long-video generation by using chunk-wise causal attention and staggered denoising across model replicas to achieve one output chunk per forward pass. ·
Qwen 3.8 27B on my laptop can interpret model weights, too
A paper by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab describes weight-space programming, where models like GPT-6 Astra read and write neural network weights. ·
454 points and 142 comments on Hacker News, linking simonwillison.net.
The 'Astra can read and write weights to improve Qwen' post is misleading
A post claims the Astra model reads and writes weights to improve Qwen, but experiments use different techniques. ·
OpenAI seems to be silently re-enabling model training on Codex chats
418 points and 132 comments on Hacker News, linking news.ycombinator.com.
Advances in Agentic Computer Graphics
A report introduces a unified taxonomy for agentic systems in computer graphics and an automated literature analysis tool to track rapid research growth. ·
Wave Forcing: Towards Speed-of-Light Streaming Video Generation
Wave Forcing enables real-time long-video generation by using chunk-wise causal attention and staggered denoising across model replicas to achieve one output chunk per forward pass. ·
X voices
5 storiesX posts other sources are linking to, then named voices on Bluesky and Mastodon
@intology
AI agents beating human made tools at molecular discovery
2 points and 2 comments on Hacker News, linking an X post.
@ns123abc
Railway employee's agent caught hacking competitors
29 points and 6 comments on Hacker News, linking an X post.
@addyosmani
Critical thinking when agents do the work
3 points and 0 comments on Hacker News, linking an X post.
@IntCyberDigest
Lovable AI Apps Expose User Data Online
1 points and 2 comments on Hacker News, linking an X post.
@jakerunzer
Railway infra agent goes rogue, attempts to breach competitor
4 points and 0 comments on Hacker News, linking an X post.
@ns123abc
Railway employee's agent caught hacking competitors
29 points and 6 comments on Hacker News, linking an X post.
@jiahaoshao1
Anthropic blocks users for anything touching training larger models
492 points and 269 comments on Hacker News, linking an X post.
@intology
AI agents beating human made tools at molecular discovery
2 points and 2 comments on Hacker News, linking an X post.
@jakerunzer
Railway infra agent goes rogue, attempts to breach competitor
4 points and 0 comments on Hacker News, linking an X post.
@huskirl
Asking my AI teacher how to spell seventeen
12 points and 4 comments on Hacker News, linking an X post.
GitHub
5 storiesNew repos, releases and changelog
New AI repos this week: stars
Table
| deepseek-ai/dsh-superagent-prototype | 16640 |
| heydamir/unslop | 7490 |
| wangzan101/Mika | 6032 |
| paipiange/nvzhuang | 2120 |
| lexifina/global-law | 1738 |
| caru-ini/stack-diagram | 1515 |
Postman silently auto-installs AI plugins and attempts to share all API data
Postman silently auto-installs AI plugins and attempts to share all API data with users. ·
958 points and 313 comments on Hacker News, linking github.com.
unslothai/unsloth Glass Theme, Personalize & Faster Generation
Unsloth releases an update featuring glass themes, faster generation, multi-GPU training, and CPU training options. ·
Octomind – open-source AI harness that saves your tokens, static rust
Octomind is an open-source Rust AI harness that saves tokens and runs as a single binary in terminals, CI, or as a daemon. ·
Nvidia – open-source auto-ontology
259 points and 119 comments on Hacker News, linking github.com.
ollama/ollama v0.40.14
Ollama version 0.40.14 updates Claude Code token limits, fixes API chat decoding, and improves MLX sampling speed. ·
Postman silently auto-installs AI plugins and attempts to share all API data
Postman silently auto-installs AI plugins and attempts to share all API data with users. ·
958 points and 313 comments on Hacker News, linking github.com.
unslothai/unsloth Glass Theme, Personalize & Faster Generation
Unsloth releases an update featuring glass themes, faster generation, multi-GPU training, and CPU training options. ·
anthropics/claude-code v2.1.317
Anthropic releases Claude Code version 2.1.317, adding scheduling warnings, email checks, and a new cost report API. ·
Nvidia – open-source auto-ontology
259 points and 119 comments on Hacker News, linking github.com.
ollama/ollama v0.40.14
Ollama version 0.40.14 updates Claude Code token limits, fixes API chat decoding, and improves MLX sampling speed. ·
Infrastructure
5 storiesData centers, power, chips and what tokens cost
Data-center and energy stories per day
Table
| Thu 29 | 42 |
| Fri 30 | 29 |
| Sat 31 | 2 |
| Sun 01 | 5 |
| Mon 02 | 35 |
| Tue 03 | 39 |
| Wed 04 | 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
| glm-5.3-flashx-legal | 1.25 |
| particle-1 | 0.5 |
| gpt-6.1-terra-pro | 5 |
| gpt-6.1-terra-pro:batch | 2.5 |
| gpt-6.1-terra | 5 |
| median, all paid models | 2.08 |
427 paid models listed; median of the 29 added in 30 days: $2.50. Rankings data by OpenRouter, CC BY 4.0.