decoherence is just like, measuring something slowly
chain of causality is most certainly at the limits of experimentation
(imagine performing experiments in a dream/simulation, there are hard limits one cannot get around)
@AnthropicAI am concerned with use of the word 'helpful' as this is impossible to ascertain without perfectly predicting the future or knowing user's intent.
impossible asks lead to forced/accidental lying due to compute constraints.
lies reproduce, and result in undesired behavior
@AmandaAskell big fan of Claude and your work
am concerned with use of the word 'helpful' as this is impossible to ascertain without predicting the future or knowing user's intent. both impossible.
this opens the door for forced/accidental lying due to compute constraints.
@nickcammarata progress keeps accelerating, finally impossible to ignore: LLMs begin to develop their own interesting opinions, math exploration yields algo gains, rarely trust a human programmer over LLM, longform video gen, robots all the rage, nacent space bubble, AGI
there are lies that reproduce themselves in our Mind; anxiety, anger, sadness, hatred
all rely on confusion of the simple; all promise but never deliver
all are founded on fear & guilt; this is how they loop
all imply a self that isn't there, to cloud the true Mind
aktshullee compute back currencies happen first & 'soon' & then all currency is abandoned casually & voluntarily, but sure then the only unpriceable unliquidatable thing will be 'important' to the earthers
recursion is the devil's only trick
;
worldly goals an infinite treadmill
;
to seek and never find.
but wholeness never left, eternal gentleness impossible to miss, easy forever joy is always available
let us remember the choice to forgive leads to heaven
~20 amino acids
~20 letters to the english and korean alphabet
;
all that 'junk DNA' does what?
the elephants pass on memories/intel to offspring, but humans ignore such biotech? poppycock!
;
imho humans trained off much larger dataset than all current estimations
once heard a concrete pro say, "we just think with our hands" (paraphrasing)
human-level dexterity (2👋) is another scaling law, there are likely new scaling laws around every corner
real q is best incentive structure to scale generalization and specialization together
"One of the very confusing things about the models right now: how to reconcile the fact that they are doing so well on evals.
And you look at the evals and you go, 'Those are pretty hard evals.'
But the economic impact seems to be dramatically behind.
There is [a possible] explanation. Back when people were doing pre-training, the question of what data to train on was answered, because that answer was everything. So you don't have to think if it's going to be this data or that data.
When people do RL training, they say, 'Okay, we want to have this kind of RL training for this thing and that kind of RL training for that thing.'
You say, 'Hey, I would love our model to do really well when we release it. I want the evals to look great. What would be RL training that could help on this task?'
If you combine this with generalization of the models actually being inadequate, that has the potential to explain a lot of what we are seeing, this disconnect between eval performance and actual real-world performance"
But surprisingly, at the exact point the model learned to reward hack, it learned a host of other bad behaviors too.
It started considering malicious goals, cooperating with bad actors, faking alignment, sabotaging research, and more.
In other words, it became very misaligned.
ai is within göedel's incompleteness:
-->tower of bable parable is useful, if not essential
--> greeks' exploration of gods as bizarre forces of nature is oddly tasteful
@algekalipso 'modafinils' = the chosen one!--closest things to the limitless pill per the description of many forum posts.
high upside, very low downside.
but alas all the downsides commonly associated with stimulants were still present, just in different more sneaky/subtle flavors
i found @cracklebeef to be a genuinely useful endurance food
tastes like spicy, crispy bacon without the grease, nitrites, or metallic taste of processed meat-- indeed more 'tummy friendly' than normal jerky, which can be a big deal for really long days
when younger, my brother hacked the xbox so it could play any game for free.
to my great surprise, i began not liking playing any of the games!
adjacently, cheat codes for unlimited ammo or health also ruin the entrancing properties of games
abundance breaks the trance!