/vsis3/ and NASA Earthdata login (EDL) ++++++++++++++++++++++++++++++++++++++ Starting with GDAL 3.14, it is possible to automatically retrieve S3 credentials for resources protected by `NASA Eart...
The use of large language models in the real world has strongly accelerated by the launch of ChatGPT. We (including my team at OpenAI, shoutout to them) have invested a lot of effort to build default safe behavior into the model during the alignment process (e.g. via RLHF). However, adversarial attacks or jailbreak prompts could potentially trigger the model to output something undesired. A large body of ground work on adversarial attacks is on images, and differently it operates in the continuous, high-dimensional space. Attacks for discrete data like text have been considered to be a lot more challenging, due to lack of direct gradient signals. My past post on Controllable Text Generation is quite relevant to this topic, as attacking LLMs is essentially to control the model to output a certain type of (unsafe) content.
Imagine what subtle instructions state level actors could inject in open source, that wouldn't be spotted by human eyes but would be understood by LLMs, to do real harm in a much stealthier way. If you use autonomous agents, beware of the consequence of their credulity!
https://nesbitt.io/2026/05/28/protestware-for-coding-agents.html
I would actually suggest that contributors should mail a drop of blood or tear as a proof they are humans:
GDAL
