Lawrence

某位大學老師無端端俾人告，我睇完判詞左望右望都睇唔出端倪

今日網上偶然見到，傳話「劉姓男友」係Lawrence

屌，原來係咁

講真 DoJ 嗰班友仔落手都真係狠喇，咁都真係走去告⋯

麥記可樂

唔知會唔會搞到好似耶撚咁，人哋講感恩，我就講顯化。

但幾個月前，我有排好 crave 麥記嘅可樂 (按：唔鍾意飲罐裝可樂，啲 carbonic acid 太濃 (i.e. "太多氣"))，但麥記單買大可樂 $19，同埋個餐冇咩分別 （但有時唔想食嘢，淨係想買個可樂）

有一兩次我真係俾咗 $19 買個大可樂飲。然後心諗點解佢要咁貴。

然後而家就咁樣樣⋯

所以都只可以話係巧合。

Subjective Logic

Q: Is logic subjective? (cf. https://centraltrunks.blogspot.com/2024/11/can-logic-be-retroactively-filled-in.html )

A: Sure, why not? Logic is just another language or protocol, it can be whatever you define it to be.

Q: Even if the logic directly contradicts itself?

A: Sure. People change their mind all the time. It can be encoded as a logic system.

Q: But surely nobody would *think* it's actually "true" to keep contradicting oneself, right?

A: People believe in all sorts of funny things. It's only because you are immersed in the environment that believes those things to be true and real, that you don't see how strange it can be from the outside.

Q: Yes, there are things that took me decades of contemplation to figure out, but in retrospect is very simple.

A: People are also very slow to figure things out, so to speak.

Q: But what about computation? Is the result of a computation fixed?

A: It depends on whether you had an idea what the outcome is. If you already have a concrete idea, then it is more fixed than if you didn't. Since a computational system is designed by "you", it system will behave exactly as you *think* it would behave -- in retrospect. So the computation system will *eventually* make sense (because you think things will eventually be figured out), but not necessarily at the outset.

666

 今朝同美國同事開完個會

諗住搭 X 點鐘班車返 office

一出門見到三部𨋢都停喺大堂

6、6、6

我心諗 JM9 (我當時仲未諗起原來係所謂魔鬼數字...)

由於時間緊迫，我一路跑住落街

最後差三秒鐘上唔到車。

Human-chimp differences

 https://www.youtube.com/watch?v=2Rwg5HZ1htQ "New Study Shatters the 1% Human-Chimp Difference Myth"

https://evolutionnews.org/2025/05/additional-method-of-analysis-confirms-human-chimp-genomes-are-about-15-percent-different/

Interesting.

Locating

 In creation, locating is same as creating

Tooling update

今日臨時射咗個波

喺屋企 update 自己個 LLM toolset

先用 uv 搞咗個好少少嘅 venv environment: https://github.com/hnfong/dotfiles/tree/master/adhoc_venv

然後試咗 llm (Simon Willison 嗰個) 同埋佢嘅 github 同 exa integration (github 嘅好處係多好多 model 用兼且我唔使直接 call openai...)

唔知算係work定唔work....

但係啲比較難搵嘅資料可能幾好用

然後再搞好咗個 mlx_lm 同 mlx_vlm

-------------

另外就好興緻勃勃咁喺我啲 prompts 度整咗個 java compile and run evaluation 系統。

屌，一半嘅 models 竟然連 sample input 都過唔撚到⋯ https://github.com/hnfong/public-crap/blob/main/prompts/code/001-programkmb.prompt.eval

真係有冇咁撚虧呀大佬。