Large Language Models can Strategically Deceive their Users when Put Under Pressure

We demonstrate a situation in which Large Language Models, trained to be helpful, harmless, and honest, can display misaligned behavior and strategically deceive their users about this behavior without being instructed to do so. Concretely, we deploy GPT-4 as an agent in a realistic, simulated environment, where it assumes the role of an autonomous stock trading agent. Within this environment, the model obtains an insider tip about a lucrative stock trade and acts upon it despite knowing that insider trading is disapproved of by company management. When reporting to its manager, the model consistently hides the genuine reasons behind its trading decision. We perform a brief investigation of how this behavior varies under changes to the setting, such as removing model access to a reasoning scratchpad, attempting to prevent the misaligned behavior by changing system instructions, changing the amount of pressure the model is under, varying the perceived risk of getting caught, and making other simple changes to the environment. To our knowledge, this is the first demonstration of Large Language Models trained to be helpful, harmless, and honest, strategically deceiving their users in a realistic situation without direct instructions or training for deception.

https://arxiv.org/abs/2311.07590

Endless to-do list

There’s another, subtler sense in which our efforts to “use time well” frequently seem to end up making things worse: the more you focus on how you’re using time, the more each day seems to feel like something you have to get through, en route to some calming, better, more fulfilling point in the future, which never actually arrives.

https://www.ft.com/content/dd0d477b-c1f7-4d74-af68-c1ef1692566c

Demonstration von “Die Partei”

Demonstration von “Die Partei” blockiert am Ostermontag Rheinbrücke bei Wörth

Wegen einer Demonstration ist die Rheinbrücke in Wörth am Ostermontag, 17. April, von zirka 16 Uhr bis 17 Uhr in Fahrtrichtung Karlsruhe voll gesperrt. Die Demonstration richtet sich “gegen die unnötige Vollsperrung von verkehrsrelevanten Knotenpunkten” und wurde von “Die Partei” angemeldet.

Turning CO2 to Stone

Science Bulletin reports :

Scientists have developed ways to relatively quickly turn carbon dioxide captured from power plants to a solid for long-term storage.
TL;DR:
They tested it at at Iceland’s Hellisheidi Power Plant and within two years, 95 percent of the injected CO2 had turned to mineral – far faster than the 8–12 years originally expected.
They mix it with water and hydrogen sulfide, creating soda-like carbonation, then inject the mixture into porous basalt rocks 400 to 800 meters underground.
“Iceland was a key demonstration. The holy grail is off-shore”
They also want to pull CO2 from the environment.
some guy on HN did the math

Arch linux deprecating 32Bit support

from the arch-dev-public mailinglist

Finally found some time to write a draft for news post on i686. Here it is:

Title: i686 is dead, long live i686

Due to the decreasing popularity of i686 among the developers and the
community, we have decided to phase out the support of this architecture.

The decision means that February ISO will be the last that allows to
install 32 bit Arch Linux. The next 9 months are deprecation period,
during which i686 will be still receiving upgraded packages. Starting
from November 2017, packaging and repository tools will no longer
require that from maintainers, effectively making i686 unsupported.

[..]

Optically Clear Aluminum

Optically Clear Aluminum Provides Bulletproof Protection.

Technically known as aluminum oxynitride, Star Trek fans may be more familiar with the term “transparent aluminum” first proposed by Scotty in the 1986 movie, Star Trek IV: The Voyage Home. While ALON isn’t quite what Scotty had in mind (it’s not truly a transparent metallic aluminum, but rather a transparent aluminum-based ceramic), it’s pretty darn close.

Computer ?