Off-policy honesty training generalizes better than on-policy honesty training
This work was done by Purvi Chaurasia with Daniel Tan and Chloe Li as part of the SPAR Program for Spring 2026.All code related to the…
Latest
Launch HN: Stoa Markets (YC S26) – A Marketplace for GPUs and AI Servers
Hi HN, we’re Eren, Berat and Kaan. We’re building Stoa (https://www.stoaexchange.com), a marketplace for new and used GPUs and AI servers.GPUs are the collateral in the…
Run interactive IDEs on Amazon EKS with SageMaker AI to power up your AI workflows
The Amazon SageMaker AI Spaces add-on for Amazon EKS runs managed JupyterLab and Code Editor environments on the cluster your ML team already operates. This post…
How nOps shipped FinOps agents 75% faster with Amazon Bedrock AgentCore
nOps rebuilt its Clara FinOps AI agent on Amazon Bedrock AgentCore, replacing a self-managed Amazon EKS stack running LangChain and LangGraph. The move cut time-to-production by…
What's the best programming language for coding agents?
Related: Which programming languages are most token-efficient? - https://news.ycombinator.com/item?id=46582728 - Jan 2026 (91 comments) Comments URL: https://news.ycombinator.com/item?id=49245936 Points: 23 # Comments: 14
b10344
model: add MTP support for Nemotron model (#26725) model: add MTP support for Nemotron Nano model model: add mtp_flags for nemotron model address review comments Website:…
DeepSeek V4 Flash 0731 is the ‘killer app’ that is going to sell A LOT of DGX Sparks
Having a ‘Killer Application’ that everyone wants to use helps sell hardware, plain and simple. DeepSeek V4 Flash 0731 isn’t an app of course, but I…
Meta’s new Glimmer AI model offers a hint at Zuckerberg’s personal intelligence vision
Meta’s new open-weight Muse Glimmer model offers a glimpse of Mark Zuckerberg’s personal superintelligence vision, as well as the emerging divide between AI users can own…
I just wrote a deep dive on using midtraining / CPT to train high-performing, domain-specialized LLMs. Here are the key takeaways… Full details are h…
I just wrote a deep dive on using midtraining / CPT to train high-performing, domain-specialized LLMs. Here are the key takeaways…Full details are here: https://cameronrwolfe.substack.com/p/midtraining-notesWhat’s the…
Four LLM loss functions → four flavors of LLM misalignment
It seems to me that, for every loss function that we use to train LLMs, we get a very distinct flavor of LLM misalignment. Here’s the…
Coercion and Deception in AI-to-AI Management
This article is a summary of an original study by Compassion in Machine Learning (CaML): Brazilek, J., Chaudhary, M., Lu, Z., & Tidmarsh, M. (2026). Coercion…
Why Addresses Have Numbers
Article URL: https://thehistoricalinsights.page/2026/06/why-addresses-have-numbers.html Comments URL: https://news.ycombinator.com/item?id=49245646 Points: 11 # Comments: 5
You're Absolutely Right
Magma Alignment & Safety disclosure note: The following are conversations that we uncovered as a result of the ongoing Manhattan Incident investigation, with alleged involvement from…
Exploiting System Management Mode with a very long interrupt
Article URL: https://github.com/xoreaxeaxeax/smiiiiiiiiiiiiiiii Comments URL: https://news.ycombinator.com/item?id=49245491 Points: 24 # Comments: 5
Best current ERP base model that are smart and uncensored?
My daily driver is Qwen3-235b-a22b-instruct-2507-Q4_K_M.gguf and it has been for a long time. I get around 75 t/s prompt processing and starting lower context ~5.5 t/s…
GLP-1 drugs linked to bigger jump in women’s employment than a college degree
Article URL: https://finance.yahoo.com/healthcare/articles/harvard-study-links-glp-1-123000637.html Comments URL: https://news.ycombinator.com/item?id=49245487 Points: 65 # Comments: 44
Thank you Mark, Alex and the whole Meta team for your contributions to open weight AI.
Thank you Mark, Alex and the whole Meta team for your contributions to open weight AI.Mark Zuckerberg: Today we're also opening the weights for Muse Glimmer,…
Book Review: The Infinity Machine
It looks like Demis Hassabis is stepping away from Google DeepMind. In honor of his rise and presumed fall, I wrote an essay on the powers…
Best current ERP base model that are smart and uncensored?
My daily driver is Qwen3-235b-a22b-instruct-2507-Q4_K_M.gguf and it has been for a long time. I get around 75 t/s prompt processing and starting lower context ~5.5 t/s…
Show HN: Ante, a coding agent in a single binary that runs offline
Article URL: https://github.com/AntigmaLabs/ante Comments URL: https://news.ycombinator.com/item?id=49245437 Points: 20 # Comments: 6
Back to the Future of Handwriting Recognition
Article URL: https://jackschaedler.github.io/handwriting-recognition/ Comments URL: https://news.ycombinator.com/item?id=49245421 Points: 9 # Comments: 2
muse spark open source!!! PERSONAL SUPERINTELLIGENCE FOR EVERYONE
muse spark open source!!!PERSONALSUPERINTELLIGENCEFOREVERYONEAlexandr Wang: 1/ big announcement today: we will be releasing an open weight version of muse spark 1.2 soon.we also are releasing muse…
The Tragedy of the Cognitive Commons
Article URL: https://arxiv.org/abs/2607.29380 Comments URL: https://news.ycombinator.com/item?id=49245359 Points: 25 # Comments: 4
Midlife Vascular Risk Burden and Dementia-Free Survival Years
Article URL: https://www.neurology.org/doi/10.1212/WN9.0000000000000152 Comments URL: https://news.ycombinator.com/item?id=49245348 Points: 13 # Comments: 2