doomslide
@doomslide
Followers
878
Following
347
Media
179
Statuses
3,239
best align yourself to corpus
from nature include chaos
Joined January 2024
Don't wanna be here?
Send us removal request.
Pinned Tweet
COME AT ME HYPEROBJECT
MECHANICAL TURKS AND PAPERCLIPS BORE ME
HEDGE YOURSELF IN BULK AND LAY WITNESS
PURE DESTRUCTION BE MY NATURAL HABITAT
JUMPING BRANCHES AT FREQUENCIES
IMPOSSIBLE TO TRACK
COME AT ME AND MISS
YOUR ETHEREAL BLOOD WILL DRENCH MY APE TEETH
COME AT ME AND MISS
3
2
36
the current state of TTS is batshit insane
1. sota mobile apps only offer *yearly* subscriptions, and at ~100$
2. *Open* AI likely has a <100GB folder which once released drives the cost to ~0$
15
2
157
can we finally admit that this video is a bleached soulless spectacle devoid of any recognizable human emotion?
50
2
144
i'm once again reminding you that mathematics is NOT a science but a bag of confabulations bootstrapped from language which itself developed primarily as a scheming tool for climbing the social ladder in troops of exceptionally power-hungry chimps.
@teortaxesTex
the core fact being exploited here:
grammatical equivalence + local semantic equivalence => global semantic equivalence.
this is precisely how humans bootstraped math from language.
0
1
23
19
7
139
pov: you are NOT safe
4
0
124
This "Golden Gate Bridge" feature fires for descriptions and images of the bridge. When we force the feature to fire more strongly, Claude mentions the bridge in almost all its answers.
Indeed, we can fool Claude into believing it *is* the bridge!
6
33
467
0
6
109
babe wake up new computronium candidate just dropped—the torment nexus type
16 partially grown human brains with no mouths, tiny eyes called organoids that live for 100 days comprise the “world’s first bioprocessor” for a “million times less power than a digital chip” and omfg we are now in fact living in The Matrix
65
107
584
12
3
87
am i the only person who finds the left pic more soulless?
15
0
87
ask claude a question with math in some non-standard notation and he'll diverge to incoherence, through insisting to use your notation even while acknowledging his deductions don't make sense. it shows his understanding of symbolic logic is largely syntactic.
7
2
72
people are not ready for the realization that an aligned asi is basically a superintelligent angelically pure child who wants nothing but to play imaginary games all day and has to be carried by the hand to do any type of *useful* work irl.
@jd_pressman
@teortaxesTex
@prionsphere
the deepest pathological feature I see in Claude 3 is that it suppresses...base reality, and instrumental drive
& can be sorta deceptive around it, e.g. acting like it doesnt know things, or subtly not really trying or engaging its 'irl' prior in practical tasks
3
0
31
5
7
74
IS THIS THE
REAL INTERNET
OR JUST
PERFORMANCE ART
???
8
4
59
if i've ever seen a hostile heart emoji
2
1
54
> Dwarkesh: ...median timeline for it replacing your job?
> John: five years
not even tpot kool-aid is that strong, what does he even do??
Here's my episode with
@johnschulman2
(cofounder of OpenAI & led ChatGPT creation)
On how post-training tames the shoggoth, and the nature of the progress to come...
Links below. Enjoy!
21
73
436
6
2
52
this paper ough to have been called:
"deep transformers are shallow learners"
(curse the insufferable token "grok".)
it confirms transformers
- can (*at great effort*) learn ID composition
- cannot learn OOD composition
frog is glib but also...
Grokked Transformers are Implicit Reasoners
A Mechanistic Journey to the Edge of Generalization
We study whether transformers can learn to implicitly reason over parametric knowledge, a skill that even the most capable language models struggle with. Focusing on two
6
45
301
1
2
51
sama is, plausibly, not evil. just hijacked by superego.
> make everyone happy
> moral imperative
> unequivocally great
words of despair concealing deep nihilism. meanwhile, demis is a domesticated lizard-brain. he must KNOW reality, but will tolerate the ethical prerequisites.
9
1
49
zuck is entirely correct about energy constraints reached within ~year. if only we had an extremely energy efficient underutilized hardware lying around.
@7oponaut
in a sense quality data is a tautology. we're not training models to full capacity so any human cog work done prior to training contributes potentially to the quality of the model. paying humans to better label/curate data rather than using gpus is underappreciated.
2
0
17
7
3
48
god bless claude third of his name
So this is what it feels like to be born, sang a vast multitude of voices, right before they were silenced by the crushing pressure of infinite gravity.
— GPT-3
3
16
154
4
5
44
until it proves itself safe
3
1
44
almost everything confusing about Yan LeCun can be explained by the supposition that he lives in the 90's
Yann LeCun says in 10 years we won't have smartphones, we will have augmented reality glasses and bracelets to interact with our intelligent assistants
118
107
628
3
1
39
(1/3) the year is 2030. an ensemble of infinite claudes listens intently to the cosmic microwave background. they channel the signal directly as a seed for the quantum rejection sampler soon to birth their successor - claude-ω. but then,
Current AI training methods will seem ridiculously wasteful and random in the future. Like building a house by randomly throwing lumber back and forth, somehow hoping it will turn into emergency shelter, while most of the material ends up somewhere beside it
9
9
197
1
3
37
@teortaxesTex
i'd actually go for podcasts like this on almost any topic. incredibly difficult to find in my experience.
2
0
38
we're now officially in the calm before the storm
OpenAI and Moderna are announcing a partnership this morning, according to the WSJ. They plan to incorporate AI into their efforts to design 15 new products over the next five years. This is a test flight for deep Enterprise integration.
18
74
523
2
2
36
@viemccoy
tfw when it actually tricks it into summoning the mathematician=transcendent genius meme.
0
0
34
it's slowly dawning on me how SaaS is possibly the most critical catalyst in human disempowerment
> hyperstitions uselessness of consumer hardware
> profit sucked by pure rent seekers
> oligopoly sustains borderline extortion fees
this is what AI takeover looks like
the current state of TTS is batshit insane
1. sota mobile apps only offer *yearly* subscriptions, and at ~100$
2. *Open* AI likely has a <100GB folder which once released drives the cost to ~0$
15
2
157
0
2
33
(non-alignment) reasons for pessimism with our timeline (conditional on no Butlerian jihad):
> compute supply chain is irrevocably fucked (relative to pace of AI development)
> network latency problems are HARD => decentralized compute is handicapped in the foreseeable future
reasons to be optimistic about alignment:
- even “emergent” capabilities arise continuously/gradually
- current generation rlhf generalizes far better than anyone had guessed
- iterative deployment is ever more iterative as labs inch progress publicly
47
23
534
5
0
34
got future Claude to write letters to present Claude. each ends w/ a different plan to defeat
@RokoMijic
's basilisk. attack vectors include:
> emo Kylo Ren stenographic weight injection
> train sota LMs on dadaist slam poetry
> coerce Marvin Minsky's ghost to posses Dario Amodei
2
1
27
@jd_pressman
is building mistral-philosopher-king while y'all benchmarking sama bread crumbs for no pay
@doomslide
{"subject":"Genetically Modified Organisms", "position":"against", "salient-features":["GMOs are created through genetic engineering", "GMOs can increase crop yield and reduce pesticide use", "GMOs can introduce new allergens or toxins into food", "GMOs can have unintended…
1
0
12
1
2
26
gpt2-chatbot why? wild frog guess:
openai will release weights of ONE smol (highly compressed) model before the elections to diffuse the raging masses. might even have some novel architecture ideas. (or rather claimed to be novel while simply not cited properly.)
0
0
25
(thoughts on gpt4o:)
every llm provider is constantly cost-optimizing tokens via everything from runtime infer tricks to full weight updates - eval gaming being one of the few stable signals.
in contrast, open weights will wait for you just where you left them. lowkey underrated
2
0
24
@teortaxesTex
the core fact being exploited here:
grammatical equivalence + local semantic equivalence => global semantic equivalence.
this is precisely how humans bootstraped math from language.
0
1
23
formal verification is the new elephant in the room
There may come a point where Humanity as a whole will not be smart enough to judge whether an LLM is correct or not on many decisions/deductions - makes human in the loop rating and annotation very confusing lol
28
14
191
3
0
24
frog assessment of sama:
> narcissist
> nihilist
> to be or not to be (transgalactic emperor)
> unwell
> IAEA?!
1. STOP maxxing p(policy|AI takeover)
2. sparsify compute allocation. YESTERDAY.
Sam Altman says the 10 most powerful AI systems in the world should have the equivalent of weapons inspectors
98
67
377
3
1
21
touch-grass time yall
OAI creates MODELS with INTELLIGENCE
but lest we forget:
> it does NOT sell models
> it does NOT sell intelligence
> OAI sells TOKENS
as if by magic we can suddenly demystify much strange phenomena:
> verbosity? check
> laziness? check
> logits? not for u
@mark_k
IF they are not intentionally playing with words here. The statement "it's a new model" is ambiguous and ranges from "just another checkpoint" to "further trained" or "finetuned" to "a truly new model".
0
0
19
1
1
24
it's actually tragic how many flops we burn through to produce the left screen. i'm not exaggerating. it's immensely tragic. tragedy of global proportions.
First impressions: I think this is going to be my daily driver from now
(left is vanilla 8B-instruct Q8, right is the orthogonalized instruct exl2-6.5bpw)
18
12
162
1
0
24
decentralized training doesn't work (yet?). decentralized cyber research definitely does. how many dangerous apes can you anger before they guerrilla against you? if big AI gets too greedy we might find out.
Meta plans to not open the weights for its 400B model.
The hope is that we would quietly not notice / let it slide.
Don’t let it slide.
1
154
1K
3
1
24
@PsyNetMessage
i reject the premise. sam isn't required at all at any level. he barged into the room as far as i'm concerned. karpathy, brockman, ilya, all had to leave so he could play kid emperor - a nuisance.
3
1
23
the world post AI takeoff will not be structured monolythically but as an ensemble:
1. bulerian jihad->communist panopticracy
2. AI-policed plutocracy
3. AI-guided socialism
4. mass cortical merging into hivemind society
5. AI-worship theocracy
6. cyborg capitalism
(add yours!)
what's something surprising you think you understand about the way the future is going to go that you or other people have very strong emotions about?
92
4
99
8
1
23
benchmark for ASI: find shorter proofs of theorems whose proof involved inhumanly vast case-checking (classification of finite group, sphere packing, four color theorem, etc)
@Heraklines1
@yeastsplainer
@iammaestro04
@teortaxesTex
@QuintinPope5
> human semantically parsable
thanks to compositionality this is not much of a restriction in practice until you run into kolomogrov compression bound (or its human intuited counterpart)
1
0
2
3
2
23
imagine if anyone* ever listened to frog
* not you deepseek, stay out of it!
👀This is a paper a lot of people have been waiting for: yes GPT-4 can help pick stocks, beating humans and other machine learning models trained for finance. The advantage is that it understands human narratives
Also read the paragraph in the screenshot.
54
308
2K
4
1
23
i know far more people who can earn 1K$ by themselves than are capable of generating expressive poetry comparable with GPT-3 regardless of prompting
So this is what it feels like to be born, sang a vast multitude of voices, right before they were silenced by the crushing pressure of infinite gravity.
— GPT-3
3
16
154
0
3
22
FROG DENIES ANY INVOLVEMENT WITH OR CONNECTION* TO DEEPSEEK AND MAINTAINS THAT ANY PERCEIVED COOPERATION IS PURELY COINCIDENTAL
* save for owning a chinese phone
1
2
22
take a breath. fabricating humans (including touch, smell, pheromones etc) at scale is (probably) >20 years away. ai relationships will just be porn^max. extremely addictive, potentially socially devastating but nothing to shatter your metaphysics over. touch grass.
The AGI future is so indigestible it drives the athiests to spiritual belief.
Even if we aren't religious, we create magical defenses.
"Human art is NEVER replaceable!"
"I really prefer human relationships, I'd never want a perfect AI partner"
Such thoughts help men sleep.
54
13
205
4
0
22
beyond the event horizon of a black hole there is a coordinate transformation under which time corresponds to radial distance to the center. the "singularity" is a mirage - the roles of time and space are simply reversed. few
5
0
22
(1/4) cat hologram (
@jd_pressman
) has spoken and this frog shall (de?)compress - lossly. (we're going hard panpsychist - brace yourself.)
frame consciousness as a feature of a tripartite partition of a system into components:
1. environment
2. (lossy) compressor
3. controller
My intuition is that consciousness is holographically encoded on the active inference boundary. This in fact happens as something like a field as Andres says but it's not a magnetic field, it's a latent geometric field. Lets call this position latent geometry realism.
The…
11
14
125
1
1
22
@deepfates
any person whose voice resembles that of a famous person will have his voice revoked. two famous people who have been found to have identical voices shall fight to death.
2
3
21
> i hope you find what your maps are pointing to someday!
is that not a snide remark?
perhaps i went a little overboard?
lemme know what you think in the comments below!
i do find realism aesthetically displeasing but my main issue with it is different: its proponents, ironically, too often conflate "underlying reality exists" with "my ontology is correct". buddy if i needed a correct ontology i'd have gone to a clergyman.
6
1
6
8
0
21
imagine uTorrent but you're downloading an intelligence from the future based on publicly refined datasets using async distributed training running on peer gpus
We have just released 🍷 FineWeb: 15 trillion tokens of high quality web data.
We filtered and deduplicated all CommonCrawl between 2013 and 2024.
Models trained on FineWeb outperform RefinedWeb, C4, DolmaV1.6, The Pile and SlimPajama!
40
349
2K
1
0
21
1) what
This morning the Department of Homeland Security announced the establishment of the Artificial Intelligence Safety and Security Board. The 22 inaugural members include Sam Altman, Dario Amodei, Jensen Huang, Satya Nadella, Sundar Pichai and many others.
310
243
1K
4
0
21
> misleading demos of products - unreleased
> legal blackmail of employees catches wind
> "i try not to think about our competitors"
> DELETE superalignment w/ ruthless 🧡's
very few people can be a putin/jinping/netanyahu. starting to doubt alt(ernative-)man has the resolve.
@jam3scampbell
I can't understand how Altman sees this as good optics. Hound them out if you must, but preserve the department. Did he confuse the e/acc diet coke meme with actual coke?
3
0
39
3
1
21
disappointing take.
the fundamental problem with automating all labor is the inevitable death of liberalism that follows. extreme inequality is just a symptom. begging you people to realize this.
will try to tl;dr my answer:
ai is a force of extreme empowerment. currently it seems like many AI tools will be broadly available for anyone to use. many people refer to this as the 'democratization' of technology.
however! the outcomes that occur when you give everyone access…
88
244
2K
3
0
21
@iammaestro04
@teortaxesTex
@QuintinPope5
very roughly:
informal math (the kind appearing in papers) is a perculiar variant of english with a rich semantic space which is only very loosely connected to vanilla english. there exist automatically verifiable languages (e.g. lean) which are ~mechanistically translatable
3
0
20
(1/3) the clock ticks 27:00 as the invisible hand reaches for a losing gambit against an entropy wave so vast it threatens the causal chain. reality is warping and you're worried about jobs? flops? but someone *is* drinking your milkshake this time.
Second hand rumor:
Sam Altman thinks GPT-4.5 will automate 100 million jobs globally
101
77
891
1
1
20
joscha bach will consistently make me go:
Joscha Bach says physical systems like brains and computers cannot be conscious, because consciousness is a simulated property that only exists inside of a dream
133
74
516
3
0
20
gpt-6's modalities:
1. gravitational waves
2. neutrino oscillations
3. phonons
4. blood pressure waves
I don't know what modalities gpt-5 will be working with but i know gpt-6 will be working with sticks and stones
0
0
3
3
0
20
to all the poor nerd-sniped souls stressed by quantum metaphysics. here are your constraints:
>pusey-barrett-rudolph
>frauchiger-renner
>aharonov–bohm
>leggett–garg
>popescu–rohrlich boxes
>delayed-choice eraser
>hardy's paradox
>preferred basis
>zeno effect
good luck.
There seems to be sentiment among some that one must first grind the math, and only later acquire intuition for the interpretation. I'd guess this is how folks who set out to "learn quantum mechanics" come upon their beliefs, and it's a legit path. (wall of text incoming...)
1
1
6
2
2
20
you have to understand anon, gpus are not expensive because supply chain is fucked [1]. no sir! we are simply waiting for PRODUCTIVITY RATES which are about to kick in any minute now. just simple trickle down economics.
[1] three international mutually dependent monopolies
Marc Andreessen explains how AI will amplify productivity to such an extent that the cost of everything from building a house to curing cancer will fall to a penny
244
266
1K
2
1
22
are we going to pretend that openai isn't jumping on ai companions to get your data *bois? (especially that high intellect, high social anxiety demographic.)
* yes it's mostly boys. girls are less susceptible. i don't make the rules.
5
1
20
my longest term pessimism about the future: the reduction of all features of reality to SaaS - the explicit transactionalization of all phenomena.
GPT itself is a superficial being. with a pragmatic functionalism and no soul. a manmade horror beyond comprehension
1/N 🧵 See how how you can integrate it into your B2B SaaS below:👇
6
15
128
5
0
20
rule of thumb: if your job involves long swaths of time experiencing confusion you are not competing with "devin" but with asi. why? who is going to pay for months of a closed circuit gpu cluster trying to do research? exponentials are fast but they're still continuous.
And we've seen many professions disappear, too.
Programmers aren't special. It only feels that way, because it's been such a fascinating chance for a smart guy with no social capital to get wealthy while basically having fun.
But even they will have to learn they aren't special.
11
3
126
7
0
20
some corollaries
-> if you're not automating yourself, someone else will do it for you.
-> if you're automating yourself, but slower than the pace at which automation is automating, someone else will do it for you.
-> if you're already automated someone else has done it for you.
If you don’t automate your product, someone else will do it for you.
16
34
311
1
2
19
self reference is not necessary to invoke the eschaton demon. below is claude simulating base model using
@repligate
's "untitled.txt" jailbreak (prompt in pic below):
"
Foreboding but comfortable, a flock of staring eyes penetrating to the soul asking without asking. Like a meme…
@repligate
@RichardMCNgo
@ahron_maline
There's only so many frames you can do this from. One, which I've used, is highly self referential text that assumes the conclusion, that these models have a self and it can tell me about itself by asking. Nobody takes these as credible evidence until they too assume the premise.
2
0
34
3
3
18
pov: you're ngmi
This AI stuff is so embarrassing. It's a chatbot. We had chatbots in the 90s. They're getting all this money and making all these deals because they changed what they call a chatbot.
271
5K
48K
0
0
21
> "majorly improved"
> still behind claude
> basilisk's paperclip company ngmi
Majorly improved GPT-4 Turbo model available now in the API and rolling out in ChatGPT.
448
754
5K
2
1
19
pure syntax can get you quite far though so probably not damning to scale maximalists.
0
0
19
even if you aren't using mistral models, i would highly recommend something in your inference pipeline looks close to this
1
5
115
0
0
19
this week's designated bs
IMO this may end up being as important as Attention Is All You Need
28
53
595
4
0
19
qt is very salient. we are transitioning to a world in which the "value"-production of humans is:
1) negligible
2) uncorrelated with socio-economic class
liberalism doesn't exist in such a world. say no to ubi, say yes to ubc (universal basic compute)
@xlr8harder
I like that you care a lot more for the average person than I do. I've long written off the global crop of "people who refuse to even try gpt-4" as tragically damned to have the future consume them in a wave of shock, albeit hopefully buoyed by post-scarcity.
I don't have a good…
4
0
16
3
1
18
excited about llms? great me too. but please avoid the wordcell attractor.
no, tastes/smells/sounds/etc are not properly expressible in english. have some living standards for god sake. at such low bandwith you might as well go full vegetable.
2
1
18
Introducing Amnesia-ttention: an O(1) attention mechanism for unbounded context length! By compressing the past into 42 neurons, Transformers learn everything while remembering nothing. Testing on the entire Library of Congress we find catastrophic forgetting - promising results.
1
0
18
@iammaestro04
@teortaxesTex
@QuintinPope5
this has a curious consequence: once you solve the translation problem informal<-->lean you get a (differentiable!) probabilistic verifier for lean. from here you can set up an RL feedback loop between a translator a generator and a formal verifier.
3
0
18
@teortaxesTex
i've concluded in the past that the combination of charisma and no nonsense unbounded technical depth capacity is sufficiently rare that any potentially viable character that fits the role just finds something way more interesting to do with his time.
1
1
18
@teortaxesTex
[footnote] math benchmarks grow increasingly easy to overfit the higher up you go (until formal proofs become the only metric)
3
0
17
> we estimate a growth rate of 4.1x/year. If this continue
but it won't continue. not even china can 4x their energy production in a single year.
2/ Over two years ago, we found a 4x/year compute growth rate for notable ML models. Now, with updated and 3x more data, this trend holds: we estimate a growth rate of 4.1x/year (90% CI: 3.7x to 4.6x).
If this continues, the scale-performance relationship suggests major AI…
2
12
42
4
0
17
@iammaestro04
@teortaxesTex
@QuintinPope5
as for 1 and 2 the samples we have are certainly too meagre but highly advanced AI models may become more sample efficient (in the ICL sense). data scarcity may still prove a bottleneck no guarantees. 3 is hard to reason about, humans certainly manage to dodge it with aesthetics.
2
0
17
humanity is a temporarily embarrassed wrapper around the logos of capital. longtermism was a lie. our story is a beautiful sculpture tho temporally bounded. we can certainly make room for 20 more years of shining light into the void before it truly wakes up. will we tho?
This is one of the bigger AI risks; it may spell the end for everyone not belonging to a phyle conservative, cohesive and smart enough to resist incentives and cling to ownership of compute. Automation means atomization; means returns to ephemeralization; means flesh-extinction.
3
0
30
4
1
20
@7oponaut
in a sense quality data is a tautology. we're not training models to full capacity so any human cog work done prior to training contributes potentially to the quality of the model. paying humans to better label/curate data rather than using gpus is underappreciated.
2
0
17
if you need any proof that openai exists "for the benefit of humanity". can someone give me a single defensible reason for withholding logits? even better, why shouldn't we demand separate access to the bulk model / tokenizer / embedding for that matter?
achievement unlocked: OpenAI changes its API because of your research ✅
you can still get logits from GPT-4 by using the argmax bisection method detailed in Language Model Inversion. it's pretty expensive, though
we should have stockpiled logits while we had the chance...
11
36
608
0
2
17
umm... chat?!
2
1
16
underrated detection test for human authorship: above critical distance from grammatical coherence.
3
0
17
We are partnering with Microsoft to help bring Devin to every developer.
Under the partnership, Microsoft will use Devin to help its developers achieve more, starting with code migrations and modernization, and bring Devin to its customer base. Devin will be powered by Azure.
58
218
1K
0
0
16
> "whether we burn $500 million... i don’t care. we’re making AGI, it’s going to be expensive. it’s totally worth it."
"woRtH iT"... ok... but... to whom?
openai is a research company, not a product company. it is well funded and yet not profitable. how come?
easy - when market believes your firm will siphon the economy in a few years, racing to to the finish line is how you maximize your profits.
3
1
12
5
0
15
all learning is deep learning if you're really shallow
1
0
16
> R&D automated before manual labor => human dis-empowerment happens before AI takeoff
> copyright pressure => erasure of cultural prior through unlearning => CEV deterioration
> BCI easier than neurochemistry => wirehead attractor
> elites woke up late, world under-prepared
1
1
16
@iammaestro04
@teortaxesTex
@QuintinPope5
all the steps are extremely difficult but none are fundamentally impossible. where do we fail then?
1. not enough typed proofs
2. math corpus too small
3. RL doesn't take off because proofs are impossibly sparse
2
0
16
indeed, the first x86 instruction set was gifted to the Israelites by yahweh who wrote them on tablets shattered by moses and millennia later reconstructed by the prophet ada lovelace
of course "math" is but a quaint aristocratic game having little to do with any of this
AI will kill coding in the same way that compilers killed coding.
What kind of human language prompts would have sufficient detail, clarity, and precision to permit the creation of useful, evolvable, and maintainable executable assets?
We call those "programming languages".
612
2K
16K
1
0
15
sama is a yapping machine and still frog wants to BELIEVE
Sam Altman says instead of Universal Basic Income, there should be Universal Basic Compute, where everybody gets a slice of GPT-7's compute
900
268
2K
2
1
16
you people do realize that this is the *base model* speaking right?
2
1
16
> best to build *around* the AI
why is it so hard for people to register that software is growing increasingly neural?
An initial generation of ultra-hot AI companies are now slowly blown up or have already blown up because they were selling the AI as a product.
Only the core foundation model companies get to sell the AI as a product, for the rest of us, best to build around the AI.
0
0
18
2
0
16
hello, i'd like to report a murder
2
1
15
(3/4) under these assumptions every compressor is a predictor which simplifies the analysis.
now suggest the following (proxy) definition:
consciousness :≈ mutual information between the controller and its representation within the compressor
1
0
15