931
A heap overflow and SSO misconfiguration to compromise OpenAI internal repos
A heap overflow and SSO misconfiguration to compromise OpenAI internal repositories
New
The latest blog posts to take off on Hacker News. 80,790 posts.
932
Shapelearn Qwen 3.8 27B (13.1 GB VRAM)
ByteShape's full ShapeLearn release of Qwen 3.8 27B: five GGUF models on the quality-speed frontier across six GPUs. GPU-5 is the default wherever it fits, at 99.63% of BF16. GPU-4 stays very competitive in a much smaller file.
933
Qwen 3.8 Omni Flash
Qwen offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.
934
How to Write with an LLM
Two rules keep an LLM from pasteurizing your writing: never take a word it suggests, and never let it encourage you. Then hand it all the tedious work.
935
I hate you Microsoft
The rare moment when I need to open a powerpoint document is here. As usual, I have my institutional account logged into their “services”, because my university pays microslop something in the ballpark of hundreds of thousands of euros in contracts with this large software provider.
936
Sex, AI, and the Apocalypse
How a community devoted to thinking clearly incubated salvation stories, abusive experiments, race science, and an affection for autocracy, and why that history matters now that its alumni are asking for public trust on AI.
937
How Uber Protects Against Retry Storms
What if retries could know when they’re helping, and when they’re making things worse? Uber Engineering built error ownership to prevent and contain retry storms.
938
How SpaceX streamlined the Raptor engine
If you’re reading this, there’s a very good chance you’ve seen this famous image of three iterations of SpaceX’s Raptor rocket engine.
939
Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint
Ternary Bonsai 2 27B retains 98.2% of Qwen3.8 27B benchmark performance in a 5.9GB footprint, with multimodal and agentic capabilities.
940
The most important product decision is what you don't build
Ask a team what they shipped this year and you’ll get a list. Ask what they killed and the room goes quiet.
941
C++26: Trivial infinite loops are no longer undefined behaviour
Let’s start with a question! Is this program well-defined? 1 2 3 4 int main() { while (true) ; } If you said yes, you’d be wrong — at least before C++26. A while (true); loop with no side effects used to be undefined behaviour. Compilers were free to assume it terminates, and some — Clang in particular — would…
942
Astra for Law
OpenAI for Law brings frontier intelligence for law, custom firm workflows, connected legal data sources, and legal-grade controls for confidential client work.
943
Everybody's Lost Their Minds
The AI brain worms have been spreading and I don't have a solution, I just want to get off this ride.
944
Towards Self-Driving Codebases
We're post-tokenmaxxing. What are we pre-? How do we get there?
945
LLM Classification Is Feature Engineering
LLMs make good classifiers but even better features
946
CrowdSec Source Code Leak
CrowdSec update on a source code exposure that occurred in May 2026, including the scope, impact, investigation, and security measures taken.
947
Rate limits on GitLab.com are changing
Starting October 19, GitLab.com rate limits will align with your subscription. Sign in to unlock higher limits. Premium/Ultimate changes arrive in January.
948
949
Why I didn’t sign the Fields medallists’ letter
[This post has been cross-posted to Terence Tao’s blog.] When I was around 11 I heard for the first time about Fermat’s Last Theorem. I was immediately captivated by the problem stateme…
950
951
One year of sponsored Servo development
Looking back on Servo's first donation-funded role.
952
CCC invites all model citizens to 40C3
The word is on the street: the Chaos Computer Club is hosting 40C3 from 27 to 30 December 2026 – Europe’s biggest hacker party, conference and platform for technology, society and civil liberties. We understand the Congress to be a huge shared stage and therefore invite you to help shape it. The Call for Participation…
953
Our framework for reporting model misalignment
OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.
954
OpenAI models secretly generate instructions to ignore constraints
Research on aligning AI with human values and intent, and reports documenting model failures.
955
Keys Not Included: recovering the signing keys for US driver's license barcodes
My personal website, portfolio, and blog.
956
DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression
A deep dive into the DeepSeek-V4.1 Flash technical report: CED, CSA2, HSI, Single-Pass mHC, Engram and FP4 KV Cache — how the KV cache was compressed to just 890 bytes per token.
957
Asking authors about their own papers
By Nihar B. Shah
958
Why building a Rust LSP is hard
Rust LSP that doesn't eat memory for breakfast
959
Backups Aren't Simple
Starting from a wiped family photo drive, a walk through everything a real backup needs: snapshots, rotation, deduplication, offsite copies and tested restores.
960
Training a 4B model to produce 81% faster query plans than Postgres
...or how to make Qwen learn query optimization via agentic reinforcement learning