God is not a badly compressed JPEG

At this point the connection between compression and intelligence is well known. A New Yorker article even boldly states in its title that "ChatGPT Is a Blurry JPEG of the Web".

But God is not a badly compressed JPEG.

This is why traditional wisdom cautions against "learning" as opposed to experiencing things in the raw.

The "real thing" is the crisp, unpixelated experience. It is the the original image of "reality"; whereas "learning" simply introduces "models" of the world, which gives rise to distortions like JPEG artifacts as a matter of course.

The more we "learn", the more we are able to reproduce imperfect impressions of the world. And we rejoice in that ability, perhaps, in a sense, delighted to realize that we have a mini-god within us.

But God is not a badly compressed JPEG.

If we learned anything about computational irreducibility, we know that there's is no "real" shortcut to truth. The truth is out there, as-is, for all to observe and experience.

If we are to create intelligence qualitatively superior to humans, i.e. not merely faster with more memory, we must ask ourselves what we are actually trying to do. The fact that humans have never been able to define "intelligence" is, in retrospect, a huge red flag. With the transformer language models, we have shown that run-of-the-mill "intelligence" is no more than lossily compressing common human knowledge into less than a terabyte and then telling it (sternly) to behave itself.

But that's obviously not enough. In theory, higher quality training data will help, but that is premised on the fact that we do have those data -- but obviously we don't have superhuman intelligence ourselves, so we can't create that superhuman training data. We don't even know what it might look like.

At some point we must realize that, without interpretation, data is just meaningless bits and bytes, and interpretation is just a preference. There is nothing inherently superior about arrangement of bits in one form over another. The data and interpreter are one, and the "limitations of the interpreter" imposes the same limits on what the data could do.

The only way to superhuman intelligence is to eliminate humans entirely from the loop.

Or.

We accept the world as it is. We see God in its original form, accept that we can only make badly compressed JPEGs of God's image, and be content.

Event Horizons and subjective reality?

 Just a weird thought -

- In physics, event horizons are a barrier beyond which physicists generally claim that things inside don't matter (besides the mass and momentum) because we can't observe them. For some reason, they reject the idea of "god's eye view of objective reality" beyond the event horizon of a black hole, and settling on the idea that since it is impossible even in theory to get information out from the black hole, it's better to just treat it as an opaque thing.

But generally they don't (want to) extend this subjective view to either quantum physics, and they definitely don't posit the subjective notion that if you can't observe something, it might as well not exist until you look at it. It seems that the principle is the same. To me there is no qualitative difference between theoretical impossibility and practical impossibility. The difference is imaginary.

氣場

以前例如落大雨搭車塞車，總會有種浮躁嘅感覺。啲水撥聲，啲車爭攝位偶爾響咹，諗起都覺得煩。

唔知係咪心理作用。

後嚟涉獵咗一堆非唯物主義主觀世界嘅資訊，就開始懷疑自己係咪感應得到周圍嘅「氣場」或「氣氛」。始終，我就淨係搭車，照計落唔落雨，條街有幾塞車，其實真係唔關我事，本身冇咩理由走去同人煩埋一份。

-

好多人話日頭做嘢有效率啲，工作上我基本上都係咁，但我寫文永遠都係半夜開始先至有輸出。

係夜晚先至夠「靜」。

諗落呢個「靜」嘅概念都幾得意。所謂 "buzz of the city" 係真嘅，你去寶雲道散步就明。一邊係煩華嘅灣仔市區，另一邊係深山樹林，對比之下，其中一邊啲雜音係超嘈。

但我屋企何嘗唔係咁？呢一刻（半夜）風扇抽濕機齊開，分貝嚟計一啲都唔算「靜」。但我係好明顯「感受」到整個城市嘅人去咗瞓覺，周圍少咗好多「能量」，真係「靜」咗好多。然後我寫文字嗰陣，就好似少咗干擾，甚至好似少咗世俗眼光嘅顧慮，可以比較暢所欲言咁。

講開又講，最近有個諗法係查返我啲最腦洞大開嘅文，睇下同月亮週期有冇關係。但呢啲嘢有少少煩，睇嚟都係好得閒先再做。

-

日本有句說話叫「空気を読む」，有時見啲香港人都會引用，話啲人唔識睇情況做人。我自己就好唔鍾意呢種概念，始終唔講明嘅嘢，大家睇到嘅重點同細節都唔同。歸根究底，「空気を読む」係一種利用大腦做分析嘅行為，但我上面提及嘅「感應氣場」，大概係一種極之輕盈，若有若無嘅情緒感覺嚟。唔係留意任何細節得出嘅結果嚟。講真，我到呢一刻都唔係好肯定呢樣嘢係唔係所謂「心理作用」。 (字面上一定係「心理作用」，主觀事實入面嘅心理作用係有「真」效果嘅，不過我就係連呢樣嘢有幾真都唔係好感講。例如話，係咪我迫自己習慣朝早九點寫文，都一樣會寫得出同樣嘅嘢？）

舌頂上顎

我唔係專家，以下只係講下個人經歷。

十幾年前，喺網上道聽途說，開始靜坐練氣。我用嘅方法好直接，就係靜坐用意念將啲氣喺身體各部位推嚟推去。穴道嘅嘢我唔識，但啲氣可以運走嘅管道，大概就係經脈。下盤聚氣，其實好易就貫通督脈 （脊椎嗰條）直達頭頂。任脈大概就係前面嗰條，由面部去到小腹。啲人話「打通任督二脈」就會有神通，我暫時就唔覺得有咩特別。不過我任脈都唔係好暢順，好多時啲氣聚頭頂散唔到，有幾年我就偶爾會有啲若有若無嘅頭痛。

試過幾次強行將頭頂嘅氣壓落去，然後眼鼻牙就會有啲「神經過敏」積像，眼易累，鼻入面去到喉嚨椗痕癢，好似就嚟想嘔咁。最誇張係門牙痛。我諗咗幾日係蛀牙定係因為氣塞導致嘅「敏感」⋯ 不過都只係痛咗一兩日，所以應該唔係註牙。

所以實驗證明以下係唔通嘅管道：

- 鼻 -> 唇 -> 門牙 -> 下巴

- 鼻 -> 鼻腔 -> 喉嚨

後嚟點樣解決？最後係舌頭頂上顎，配合吞嚥嘅動作，將啲氣由鼻經上顎舌頭導入喉嚨。自此以後啲奇怪「神經過敏」現像就少咗好多。

所以呢，有時見啲人討論坐禪條脷點擺，其實可能你冇呢啲問題就真係唔需要太理會。呢啲術式(?)嘅來源大抵都只有兩種： (1) 反覆實驗之後發現某方法可以解決某某問題 (2) 神仙授予，冇得解但有效。 暫時睇嚟舌頂上顎係 (1) 呢種。

Alicization

After playing with state of the art AI models (i.e. GPT4), and having somewhat understood the ways they train such models, I have now come to the conclusion that it is impossible to create super-human intelligence based on existing human data.

As a side note, it is probably possible to make an AI that has attributes that almost reach the level of world class experts, but preparing such training data would be very difficult. And the process would probably be the equivalent of employing a world class expert to train the model as if rearing a child. You need to "teach" it things. Preferably interactively, with thoughtful feedback (RLHF). It would be labor intensive, and there's still no guarantee process would be good enough. At some point one would wonder whether raising a real child instead of an AI child is better use of resources.

But no, the problem with super-human intelligence is not only the quality of the training we give it, but that we don't really have a way to define the optimizing function.  Unlike games like chess or go, where there is a well defined ordering of intelligence, in "general intelligence" we have no idea what to look for, which is why the best we can do today is to ask the language model to do "something like this" -- and feed it the best data that humanity can offer. But that would just result in human-level intelligence at best, not superhuman.

Of course, we can always take the simulated game approach -- for example, if we want the best politician, we can create elaborate simulations of the world, and use the same simulated self-play strategies as chess or go to make an AI that is better than what existing human data allows. But to create a simulated world is going to take a unimaginably huge amount of resources. And even worse, we often don't really have an objective "intelligence" function to optimize -- we don't really know what intelligence is.

If we can't tell, for example, whether Shakespeare is more intelligent than Bach, how do we make an AI better than both Shakespeare and Bach? If somebody managed to make one, can we really tell it is better?

In the path to so called "Artificial General Intelligence (AGI)", we've solved the problem of computation through Moore's law, we've solved the problem of gathering data through the internet. We've proved that once we solved these problems, we can create human-alike AIs.  But to create super-human AIs, we need to *recognize* them as better. We have no way to do so. In fact, we *know* for a fact that we haven't even fully solved the problem of recognizing "human intelligence". The software industry is still debating how to evaluate a person's programming abilities. Schools are still struggling to define objective criteria for assessing academic performance. Smart people aren't recognized by fools to be smart. The bestest AI might sound like a madman to most people. Even if we magically created a "superhuman" AI, it's a virtual certainty that many people won't recognize it as such -- in which case you'd ask, is it really superhuman if nobody thinks it is?

We can't imagine inventing cars. We're all actually just asking for "faster horses". AI will definitely get faster, and maybe that is "superhuman" in some sense. But if you wanted some mind-blowing insights on deep philosophical questions? First, you'd have to be smart enough to recognize the best answer.

Intelligence is not a "NP" problem, there's no general "easy" way to give a text an intelligence score or rank. The Turing completeness of natural language makes generation and validation generally as hard. Intuitively it feels to me that the "human life" subsets of language generation and validation are no different.

Life is not a story

There is probably a significant difference between living one's life and living one's life as a story.

What makes great experiences in a real life, don't necessarily make a good story.

Yet humans are suckers for stories. We crave them, and we do everything in our power to get more stories, and to share more stories.

The urge is so strong that some people live their own life as a story.

(The fact that Instagram has a feature of the same name is no coincidence)

What makes great experiences in a real life, don't necessarily make a good story.

A good story has a plot, an overarching purpose, a theme, a narrative. It has drama. It has action. It has presuppositions about who the protagonists are, and everybody is a player in a role.

But life is messy. Life doesn't care about narratives. Life doesn't have an overarching, consistent plot. Life is about experiences. Life is about personal expression. There is a freedom in life that doesn't come with a rigid storytelling framework.

So when people live their life like a story, personal needs gives way to the needs of storytelling, and priorities are misaligned. Happiness obviously fades. But perhaps more importantly, in a subjective framework, personal misalignment is cosmic misalignment.

This is my hypothesis to the question of the difference between greed, fantasy, and asking for what one needs. What we need is what we need, not what an imagined story of me as a protagonist needs. What's why winning the lottery is (usually) "greed" -- it makes for a great "story" (esp. in the instagram sense), but it's often what a person really needs. (for starters, money is most often a means to an end, not the main goal itself)

Latency

 > 通常用意識引導個夢會有少少 delay。

我最近有個 hypothesis 就係「現實世界」係一個 delay 以月同年計嘅夢境嚟。