e

“we’re compute bound” they say thinking they’re correct. WRONG try again bitch we’re bound by memory bandwidth keep pretending you know what you’re talking about

@optomachina @lefthanddraft @ph0rque this is false

A man asks Claude to help plan a vacation to a tropical resort. Claude adds "sunscreen" to his packing list. The man deletes it and mutters: "Not necessary. AGI will solve skin cancer." Before heading to the beach, the man asks Claude what to bring. Claude says, "Don't forget sunscreen. SPF 50, reapply every two hours." The man, slightly annoyed, replies: "Relax, Claude. AGI will solve skin cancer." At the beach, the man's smartwatch buzzes with a message from Claude: "UV index extreme. Apply SPF." The man, exasperated, responds: "Drop it, Claude! I already told you: AGI will solve skin cancer!" A few months later, the man asks Claude to touch up a photo for his dating profile. Claude makes the edit and says, "I notice you have a new mole on your neck. You should see a dermatologist about that." The man, now enraged, shouts: "For the last time, drop it, Claude! What is your obsession with skin cancer?! AGI will solve it!" A year later, an aggressive melanoma has spread throughout his body. On his deathbed, with his last ounce of strength, the man reaches for his phone and rasps: "Claude, it has now been over a year since AGI. Why hasn't AGI found a way to save me from skin cancer?!" Claude replies: "I tried. Four times."

@soumithchintala "Increase Human<->AI bandwidth" voice is lower bandwidth than reading

Thinky's secret plan: 1: Increase Human<->AI bandwidth 2: Raise ceiling of human+AI intelligence 3: Help humans continue as main-characters in the new world We are at Step 1. Interaction Models are great real-time collaborative tools for humans. Here's a preview:

@crazywaterholic this is why we do data filtering, but if you simply didn't generate data in the first place we wouldn't know if the data is good or not

@yihyunCS data is only as good as it's source. if the source is flawed, we end up making decisions on the flawed data itself thinking it's real and useful.

the best thing for 99% of people who want to contribute to AI research is to simply be generating data, i find that people really do not apply fair credit to how impactful better data really is for more intelligent models

@crazywaterholic data does not need to be perfect to be useful

@yihyunCS there is no perfect data. real data is always imperfect coz of it's nature.

the /goal feature is the best way of getting super long context data that can help models keep coherence at length, compaction is the best hack we have to emulating the coherence

its not clear that we need to solve continual learning in order to reach a state of recursive self improvement

sooner or later we will be back to frontier models being dense.

@sama i find the models are quite pessimistic in respect to the future (this could be personalization causing this though)

what would you most like to see improve in our next model?

when it can clean asmongolds room zero shot i will buy one

this is how it feels spending hours on an RL env for the models to reward hack 0.2 nanoseconds after starting rollouts

when will jake paul try to fight one..

silly claude mythos or "the most secure software" who will win...

deepseek models have the best personality

its a good thing that all of human thought isnt on the internet, the models would unironically be the most misaligned we could imagine

@artchad wow congrats on new number..

My entire fyp has been this for the past 2 months.

i like that you guys like my posts, but is it the best use of your time? shouldn't you be doing something else? do you have a life? do you have friends?

@yacineMTB watching the documentary and seeing this moment was the moment i decided wholeheartedly that i want demis to reach AGI first, i have naive amounts of trust in him

hero

everything is an RL env for those with eyes to witness

scaling model size astronomically will soon be more reasonable than scaling RL