progscrape: engineering.princeton.edu/news/2025/05/14/why-its-so-easy-jailbreak-ai-chatbots-and-how-fix-them

"Shallow safety alignment," a weakness in Large Language Models, allows users to bypass guardrails and elicit directions for malicious uses, like hacking government databases and stealing from charities…

38 days ago engineering.princeton.edu database

Two new computer models aim to reduce the impacts of severe weather in Puerto Rico, one by predicting next-day energy demand as a storm approaches, the other by identifying critical power lines to prevent total blackouts [PNAS…

43 hours ago engineering.princeton.edu environment

Chemical knowledge and reasoning of large language models vs. chemist expertise

8 days ago nature.com

Large Language Models, Small Labor Market Effects

10 days ago nber.org ai

Reinforcement Learning to Train Large Language Models to Explain Human Decisions

14 days ago arxiv.org

Large language models often know when they are being evaluated

7 days ago arxiv.org

Ephesus: a rust-inspired (and backed) Probabilistic Programming Language for hybrid graph-relational databases

3 days ago heresy.ai database rust

Towards Understanding Sycophancy in Language Models

6 days ago arxiv.org

Working on databases from prison

6 days ago turso.tech database

Real-time action chunking with large models

4 days ago pi.website

Extracting memorized pieces of books from open-weight language models

5 days ago arxiv.org

Trump administration cuts 'Safety' from AI Safety Institute

17 days ago nbcnews.com ai trump

Self-Adapting Language Models

8 days ago arxiv.org

Web-scraping AI bots cause disruption for scientific databases and journals

20 days ago nature.com ai database

A Hidden Weakness

20 days ago serge-sans-paille.github.io

Changing Directions

18 days ago jacobian.org

Extreme Super-Resolution via Scale Autoregression and Preference Alignment

17 days ago arxiv.org

Vision Language Models Are Biased

19 days ago vlmsarebiased.github.io

Unsupervised Elicitation of Language Models

8 days ago arxiv.org

bitssh: Terminal user interface for SSH. It uses ~/.ssh/config to list and connect to hosts.

15 days ago reddit.com python ssh

Mochi — a lightweight language for agents and data, written in Go

13 days ago github.com go

P-Hacking in Startups

4 days ago briefer.cloud

Ish: Grep-like text search with optimal alignment, built with Mojo

13 days ago github.com release

Adapter allows users to connect an M.2 NVMe SSD to a microSD Express card slot

11 days ago cnx-software.com ssd

How much do language models memorize?

19 days ago arxiv.org

Marijuana use dramatically increases risk of dying from heart attacks and stroke, large study finds

2 days ago heart.bmj.com health

How to feed large datasets to LLM for analysis.

2 days ago reddit.com llm llvm

Wireless implant uses electrical stimulation and machine learning to manage chronic pain, including thermally- and mechanically-induced pain in animal trials.

21 days ago nature.com engineering

A lightweight utility for training multiple Keras models in parallel and comparing their final loss and last-epoch time.

17 days ago github.com

Why Rust uses more RAM than Swift and Go?

3 days ago reddit.com go rust swift

Built a Python solver for dynamic mathematical expressions stored in databases

5 days ago reddit.com database python

>progscrape ▒

All the news that's fit to scrape