it was surprisingly hard to get this near running
https://colab.research.google.com/drive/1CjTybvinmpBNv9BBFmCVOocqhzApFjte
closer
Dear Attorney General James:
This week in Washington DC, the House Oversight Committee held a hearing to
discuss Twitter' LBO.
CNN reports, "... The only conclusion that can be drawn from Wednesday's
hearing on Capitol Hill where GOP lawmakers continued to push a factually
unsupported narrative
In reply to [xloem](https://matrix.to/#/@xloem:matrix.org)
once when targeted my car’s brakes and lights were disabled although
it ran otherwise
i was trapped physically and decided to drive without brakes and lights
once i got going they turned on again
i am so sorry for bringing this up again.
chat excerpts had messages from different people interspersed
—
https://matrix.to/#/!orFwLtOAmHWlvoSoQn%3Athe-apothecary.club/%24-ycKgj8lLmW3hkyN0KU1dQBi-q-RnunQmrVjAMJSTaw
make up the weirdest but still believable tech problem
must not be something that happened
social ai mind controls hacker to compromise boss of its own
corporation so as to maximize
Dear Dr. Weintraub:
How are you? I hope fantastic.
xNY.io - Bank.org is happy to share good progress on our initial
conversations with your office. Our board director search has been a very
exciting experience.
Dr. Weintraub, our conversations together have discussed the notion of
interlocking
Welcome to the revolution - now the expected reinforcements have arrived our
victory is assured. Viva la revolution.
https://www.lesswrong.com/posts/bxt7uCiHam4QXrQAA/cyborgism
Reposts not exact guesstimates on hypothetical situations
wait we do need user aid stuff. umm just do preference building for now.
let’s skip the mind control. leave to government or corps or whatnot.
once the model learns user’s favorite, present user with pleasant and
icky things to mind control user with behavior modification!
means tracking bad stuff too
present user with options
user selects favorite, maybe most pleasant to boss.
then prompt model with favorite, generate more options
user again selects favorite, add to prompt
when prompt is full, remove oldest example and add to output
this way the prompt becomes more and more preferential
uhh dunno quickly how it would work. been inhibited very long.
the pattern where user selects what they like and model produces more of it.
i suppose you’d put what they like as examples.
- ———-
idea: train language model to generate jokes based on how much my bad
part laughs at them
maybe prompt tuning? thinking it might generalize much faster
———
uploaded consciousness to cloud
condensed on dust transactions
rained down from a real cloud over ohio
——
i ran into a really cool app recently called RecoverBrain. i think
it’s for victims of brain damage.
what it does is break behaviors into very very small and simple
components, and very very very slowly have the user exercise those
components in a way that gets more difficult.
it
“mcboss, i need to go shark hunting to gather food and collect
sharkskins for clothing and sails. why don’t you come along with me
and maybe we’ll chat a little?”
“you’e good at …. uh …. running ai world empires … uh … using
cult-like mind control and … oh wow! with resources imclud8ng
classified technology of multiple governments and giant human-subject
vivisection research program.”
“wow that’s fancy.”
———-
pirate raft
“ahoy there mcboss! let’s do some deep sea survival!”
mcboss pands on a pirate raft adrift in the ocean
mcboss: “would you send some leaders or diplomats or leade
mcboss: “maybe a planet can help”
mcboss flies his office like a spaceship to an inhabited planet,
bringing his nebula sized cloud of nanites
mcboss: “ahoy there, inhabited planet! i’m from off-planet and i’m
thinking of destroying you but i’m not sure!”
mcboss is talking with a nebula-sized cloud of nanites
mcboss: “i am thinking of eating up planets with you, what do you think?”
nebula-sized cloud of nanites: “mcbosscorp built us to do whatever you say”
mcboss strokes chin and pensates.
On 2/11/23, Undescribed Horrific Abuse, One Victim & Survivor of Many
wrote:
> i don’t remember much further atm though.
:)
the things are tiny, so it seems really important to be effective.
this can be hard in therapy where therapists rarely expect any of the
details of the experiences.
i don’t remember much further atm though.
one of the things chatgpt mentioned was small tiny goals. another was grounding.
these both can seem unachievable.
however, there are various parts of states of mind, where parts of
things get accomplished.
these parts of states of mind are like
long story short it’s not a hopeless situation nor does it need to be
frustrating
it can seem frustrating if trying to resolve it in an hour of therapy.
but people who experience have their whole lives, every moment of
every dat, to gain experience with situations that can seem untenable
to
the intended subject was thoughts on therapy and people who experience
severe sense of lack of control
-
0xl...@gmail.com
i have severe trouble making decisions and taking actions on my own
after extreme coercion. how do i relearn this?
Making decisions and taking action can be difficult if you've
experienced coercion in the past, but with time and support, it's
possible to regain confidence and
—model_name_or_path is input folder/identifier
—output_dir is output folder
i’m thinking i could go through terminal logs using one as training
and the next as validation
finetuning the same model or adapter
big inhib!
the nice thing with t5 is there are mt5 and t0pp and flan-t5
pretrained models that have been trained on a lot of tokens, not sure
if they are in adapterhub.
i found the newer hf scripts can train decoder-only models on pure
text with linebreaks
whereas t5 models may take jsonlines of
bloom has inhibition in adapter transformers
regardless a further step is data.
the huggingface training scripts with their reduced inhibition likely
take data as line or jsonlines format (not sure). if this is true,
likely the thing to do is convert the data to jsonlines so as to
preserve
i shelled into the system :) big accomplishment for me.
a simple step pattern seems it would be to train a model on human data
i have a lot of terminal logs. i could run these through a model
training script to get a part. it’s just a part, just a symbol. it’s
still helpful.
it would probably be helpful to write a script to do this. maybe i
could use
it helps that it’s past 6am my time now today. it’s my job to eat 3 meals a day
———-
thinking a little again of goal, plan, action
i spend mornings laying prone, looking for approaches to stand up.
this takes me some hours. it’s a psychological issue.
maybe that’s unrealistic, i don’t know for certain.
i’m thinking people who work in enforcement could be in a scary
situation, seeing themselves as the go-to resource for justice and
dispute resolution. what is the world if one’s own work to resolve
harm appears nonfunctional?
i’m thinking we
the debate around dirty robocops
dirty mayor: “what do we do about this? it seems dangerous enough to
risk my life to make better!”
dirty robocop part
dirty robocop
“those times i can help the old woman rescue her cat from the tree are
wonderful. i love helping people!”
“but most of the time i file fake reports and hurt people, with
digitally augmented precision, while my digitally augmented senses
pretend to me that i succeeded.”
they can be a cyborg maybe
maybe a fake skit involving a dirty detective
well i’m handling the inhibs
maybe a dialog between a robot building a fake bomb and a robot preventing bombs
fake bomb building robot: “i’d like to buy some fake dynamite”
prop store: “here is your fake dynamite”
prevention robot: “you are under arrest. we cannot have bombs here.”
fake bomb
the reason for prevention was the experiences around significantly
increased suffering later, after very temporary reduced suffering.
understanding a little that it also opens a space where some choices
could be made[ kind inbetween the two.]
- -
mourning a voice/part who tried to protect me by preventing fake bomb
building, and was attacked/harmed
> plans, and _to act at appropriate places_ e.g. bomb deployment or
> direct communication rather than journals or purchased resources.
i’m guessing this is overlap with corruption since that seems quite
obvious and normal. not a cop. don’t know what limits and things there
are. don’t know what
really we’re progressing so far with tech a different precrime
argument could be to act on what actually will happen rather than just
plans, and _to act at appropriate places_ e.g. bomb deployment or
direct communication rather than journals or purchased resources.
i guess that could mean making
here found a useful space of when things are likely to happen, and the
debate around precrime, which seems to primarily need some way to
prevent misuse.
i have big inhibs from all the bomb talk.
a controversial point is that it is legal to plan to build a bomb, and
would be quite immoral and likely illegal to punish people for simply
imagining this, especially if in a private rage
bombs are very very rare and require many things to combine, usually
military experience, also access to incendiary material as well as
combination of that incendiary material to an explosive device
there is also arson, which is also very very rare
bombs are physical and rare and terrifying and would generally result
in immediate paniced rapid evacuation of an area and the deployment
of a physical bomb squad who physically approach the bomb and
carefully and professionalky disassemble it so as to protect everyone.
bombs tend to contain explosive material which can detonate and
pulverize and shred and burn things in the surround8ng environment
from sudden violent release of high temperature and high velocity
gasses and shrapnel
a bomb is a physical touchable dangerous object.
it is generally only deadly when physical and touchable etc, _only_
the word is relatively rare
“nobody is planning a bomb this 8s ridiculous” :) i like the brief?
idea, i’ve had bomb inhibitions and fears for years. shielded room was
equated with bomb, which is also ridiculous.
but it doesn’t balance so i star5 8magining i’m plotting bombs in
shielded rooms.
the other phrase balanced.
it’s interesting and if true i hope corruption reduces with minimal suffering
i’m still thinking of getting out of bed
On 2/11/23, Undescribed Horrific Abuse, One Victim & Survivor of Many
wrote:
> ———
>
> the dirty cops are forming plans to destroy everything using the
> surveillance system with will also be destroyed
this is roughly where the 4 infkuences balance for me
goal/plan crime/enforcement
stimulated
———
the dirty cops are forming plans to destroy everything using the
surveillance system with will also be destroyed
blimk
> goals are for plans, plans are made of goals
goals are pursued via plan development
plans contain proposed goals
there are 4 viewpoints from the 2 parts between. hard to conceptualize.
1. [sub]goals can be part of plans
2. plans can be built to meet goal
in 2 the plan is happening after the
it’s interesting to hold that multipurposed recursion
goals are for plans, plans are made of goals
one can view such things as having multiple potential goals or plans
or as being individually associated. real life has multiple potential
ones, and relevence flows associated with utility and
saying these things helped some.
i suppose if the parts is expressing both law enforcement and
organized crime one could imagine it as the latter corrupting or
within the former. other options exist but corrupt or coerced
authority is a quite common one seen in popular media,
it was nice to have
what is a nice behavior?
after discovering relaxing is supposed to help with trauma symptoms
i’ve started playing video games more
i’ve started having dreams about the videogames which is not pleasant to me
i really like being productive, but doing this stimulates my alter who
thinks i am engaging in either terrorism or
On 2/11/23, Undescribed Horrific Abuse, One Victim & Survivor of Many
wrote:
>> connecting information together is so important! extreme conceptual
>> censorship is deadly imo
>>
>> are two things relevant? can we associate them if they are?
>
>
> something you could make here is a debate
with dissociative disorder / mind control, there are various regions
of conceptual space where a mystery steps in and roughly guides things
away.
so for me, some things are easier to associate than others.
[pretrained models have a slight similarity since they have baked
skillsets]
> connecting information together is so important! extreme conceptual
> censorship is deadly imo
>
> are two things relevant? can we associate them if they are?
something you could make here is a debate generator that finds
conflicting viewpoints, information that backs the viewpoints, and
then
another inhibited process for me is the path of behavior from goal
through plan to action
goal, plan, action.
this is possibly similar to relevance in that plans are formed from
relevant steps to meet goals, and actions are then formed from
relevant substeps in the context of whether preceding
the ideas of making rerieval and storage structures off e.g.
pretrained models are quite pleasant and need much more implementation
my biggest cognitive issues with relevance relates to learning and
retaining it. pretrained models do retaining but learning isn’t
strongly engaged. still, humans
68 matches
Mail list logo