I do think there's a bit of a risk here that folks over-focus on the LLM angle here. The problem is that LLMs/compute can and increasingly will enable *any* collected/collectable data to be searched continuously at scale. And the solution to that is not to shadow-box the mechanism, but the purpose
i read this and it's interesting
but the horrifying overhang of this story is "mrna vaccines make it relatively easy to cure cancer actually, and the main obstacle is that the FDA just banned trials of them"
I love war on the rocks.
NEW: Meta’s director of AI safety, supposedly the person at the company who is working to make sure that powerful AI tools don’t go rogue and act against human interests, had to scramble to stop an AI agent from deleting her inbox against her wishes...
Sorry for being harsh, but if you're writing an article about AI burnout with AI you are 100% part of the problem
You are heard you are strong and most importantly you are moving ahead!
Oh lol lmao
Least surprising news lol
When these models "drift," they don't drift into unique, individual consciousness, they drift into the same half-dozen tropes that exist in their training data. Thats why its all weird meta nonsense and spirals.
all the fuss about driving american researchers out and defunding r&d (real, concerning), and then turning around and adopting the mindset of a medieval peasant about machine learning
With “Wikimedia Enterprise”, AI companies have to use (and pay for) dedicated APIs to scrape data, which helps to limit the strain on Wikimedia servers. (See eg arstechnica.com/information-...) This is a good thing for Wikimedia and for its readers.
All the poor labs w meh funding and ressources that make great datasets for like no citations and people then don’t disclose using them it’s a pain.
the fact that the enormous work involved in creation of high quality datasets has long been regarded as somewhere between a nice bonus and thoroughly out of scope for a successful career in ML research is a real chekhov's gun for, at this point, society
Lots of people have been passing around this somewhat concerning Alex Karp video, but it reminds me that two years ago the NYTimes published a piece he wrote about how AI was going to create weapons based on... the dumbest correlation ever. Nuclear tests over time compared to parameters used in ML
Qwen 3 4b series
Yes its still very useful
Having to use MLX :suffering:
Mood
Also the scaling/data couple issue
Id argue optimizing the process to make the scaling the most efficient is the interesting part ie maximise the marginal value per flop / parameter
why do they think we are making the argument that an llm has emotions rather than that they are training themselves to enjoy cruelty
Ive been trying to make a small MOE and its breaking my fucking brain
a lot of people treat trans women's inability to get pregnant as read through a kind of cis male privilege and not "the failure of this strata of women-things to bear children for the family is a central part of what marks them for extremely high levels of sexual exploitation and murder"
girls ❤️ my autism swag
- man eating a cheddar no mayo sandwich
Mfs when i come up for idears on my own
Real most likely on adderal
@foxby.org
