#arxivator

[icarusgf]

head ta of the class i ta for is literally my ideal person i have nvr been attracted to a man as much as i find him attractive . his thesis supervisor is my academic advisor............. i want u

[monopeptide]

yall I know that chatGPT arxiv paper confirms everyone’s prior beliefs but can we take it with a grain of salt please. I’m a big AI hater but i feel like there are a lot of red flags

[flavorcountry]

Are the machines just making ads for other machines or what, like I get the fake pretty girl pic, sure, but how in the flippity floppity fuck is this copy supposed to appeal to anything approaching a living human consciousness, what are these language models trying to sell each other and should we be concerned

[zoueriemandzijnopmars]

I wish arXiv worked like Ao3 in that it showed the number of hits on my paper

[nobutlikeimobessedwithbutchlogan]

Guys is there some scientific paper that explains why I hate everything in the mornings until I have my coffee? Like drop the arXiv link for me plz <3

[gudamor]

Stop referring to 50/50 odds as 'a coin flip'!

Coins have a tendency to land on the same side they started, confirming the predictions from the Diaconis, Holmes, and Montgomery (D-H-M) model of coin flipping.

Though if I'm reading it right, this tendency does depend on the tosser.

[rlyehtaxidermist]

definitely the funniest part about my second master's thesis was that i got to cite the 3ds hacking tutorial

[cheerycherrycandy-resurrected]

Hey, a question for my fellow academia-ghesties!

And other fandom peeps in general I guess

Please please please tell me I am not the only one who, when my colleagues talk about "putting something on the archive", thinks first of AO3 and not arXiv!!!!

[spacebrick3]

arXiv is a goldmine actually

[ulaulaman]

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs). For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024. We find that such backdoor behavior can be made persistent, so that it is not removed by standard safety training techniques, including supervised fine-tuning, reinforcement learning, and adversarial training (eliciting unsafe behavior and then training to remove it). The backdoor behavior is most persistent in the largest models and in models trained to produce chain-of-thought reasoning about deceiving the training process, with the persistence remaining even when the chain-of-thought is distilled away. Furthermore, rather than removing backdoors, we find that adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior. Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.

I would like to underline that "strategically deceptive behavior"…

Image credit: from Avengers #55 by Roy Rhomas and John Buscema

[sklogw]

such a well written and succinct intro to the central question of observational extragalactic astronomy

[il3x]

mostly when I say "I could fix him" I'm joking. but with specifically Hollyleaf, child Eurus, and possibly Chris, I am fully convinced that if I was allowed to befriend them and talk to them with my current general resources I could make them better. Morally and psychologically. They'd make me worse in equal measure, but it's the thought that counts.

[skilopsaros]

I really do want to understand what I'm supposed to do in this new job, but when some of the required reading has sentences like:

“Alternatively, non-Clifford gates can be effected by preparing special resource states called *magic states* that can be used in a type of gate teleportation circuit.”

It's really hard to believe that I've not been hired by a bunch of wizards.

[sparkmantis]

TOI-1846 b Super-Earth Discovery: TESS Finds New Alien World Around M Dwarf

TOI-1846 b: TESS Discovers a New Super-Earth Orbiting a Nearby M Dwarf Star NASA’s Transiting Exoplanet Survey Satellite (TESS) has made another remarkable discovery: a new super-Earth exoplanet named TOI-1846 b. This alien world orbits a nearby M dwarf star and stands out due to its size and mass, being about twice as large and four times as massive as Earth Key Facts About TOI-1846…

[kousofriends]

Thought-Augmented Planning for LLM-Powered Interactive Recommender Agent

[themonsterunderthebed]

arxiv papers arent peer reviewed arxiv papers arent peer reviewed arxiv is not peer reviewed!!!!!

preprints are cool but don't mistake them for peer reviewed studies!!! they might be rigorous and trustworthy studies OR THEY MIGHT NOT and you won't know without critically reading the paper yourself (assuming you have sufficient knowledge in the field to determine its quality for yourself) because ARXIV IS NOT PEER REVIEWED

if a post or article cites a paper from arxiv and says "look, a study found XYZ result!!" without also citing a PEER REVIEWED JOURNAL then you should take that with a massive grain of salt!!! even if you agree with it or it aligns with your worldview!!! the scientific process exists for a reason and peer reviewed studies exists for a reason!! arxiv.org ALSO exists for a very valid and admirable reason but that reason is NOT to distribute PEER REVIEWED research