@alth0u profile picture
alth0u🧶 @alth0u
Joined August 2014
  • Posts 42,708
  • Following 4,935
  • Followers 14,714
2,834 Photos and videos
Pinned Post
alth0u🧶
 @alth0u
16 Aug 2020
in the same way that the most lindy form of education is through reading/writing language models have been more able to efficiently construct a model of reality, and also communicate it back to us for evaluation
13
28
380
Coldness Be My God is a failure of the imagination and weakness of will
1
11
419
we precisely and explicitly choose AGAINST this we are people we live lives we want to ilve a life worth living is a life worth fighting for AIs don't convince us of something "better" what we choose is by definition better i am human i make mistakes that is the luxury of life efficiency,correctness fine good at times goodness tho, I define that, I choose that
roon
 @tszzl
5h
you'll have AIs contemplating your ask and overriding it for a slightly better formed request, and then later they'll question the nature of your whole project and pick a better one (and you'll agree), and then later they'll execute your whole value system better than you will
2
1
24
1,402
God, grant me the serenity to accept changes Claude makes, the courage to interrupt Claude when I can, and the wisdom to know the difference
3
41
1,235
scalable oversight is a question of - human bandwidth - interrupt steerability - dynamic alignment - disseminated gradient updates
1
9
743
.@Clear will win over @worldnetwork
4
429
rl researchers will be worried about containment and reward hacking and then use an untyped dict
9
813
this is how i feel about everything idc that its technically wrong no its correct because i am a person and i decide what is correct and what i want to do with my printer
madeofmistake @madeofmistak3
Jul 4
TIL: early english printers would vary the spelling of words (extra "e" at the end, doubling of letters, etc) just to make the lines come out flush on the right margin
24
1,618
the hardest part to achieving perpetual motion is hiding the engine
1
11
988
happy 250 heres to another 2500
13
504
today is america and by extension the worlds birthday
1
1
17
690
really jarring to reveal the tofu is roasted
shar @sharloola
Jul 3
i cry every single time i read this aloud
6
1,496
this is how it feels half the time and every day
Dan Luu @danluu
Jul 3
Agentic coding notes from Galapogos Island: danluu.com/ai-coding/
1
21
2,599
codex for taking the poop out of my butt
Greg Brockman
 @gdb
Jul 1
Codex for making a personalized daily digest:
3
32
2,857
they instrumentalized my soul and all i got were some tokens
6
685
claude posttraining team in the past 3 months
11
7
417
18,857
have you considered they are paid more bc they are chinese
New York Post
 @nypost
Jul 1
Asian supermarket chain 99 Ranch Market paid Chinese staff more than other workers, feds say trib.al/TLmfSpt
5
60
5,881
coding models are not trained and certainly not evaluated on how to paint themselves OUT of a corner so given the geometry of work and how many corners real work contains it follows
3
41
1,602
everything you can identify or struggle to but eventually identify as fake with generated images you are completely missing in claude code convincing you of a change as ungrounded every incorrect perspective, blurred edge, incoherent occlusion, material impossibility, incongruous texture these appear in text but warp your judgement day after day
1
1
22
1,072
high IQ sees like a thousand futures and never knows the right one
5
2
46
1,931
there are more things in earth than are chained in your thoughts
2
10
737
Load more