HackerTrans
トップ新着トレンドコメント過去質問紹介求人

stealthcat

no profile record

投稿

The Judgment on Fiat Currency [pdf]

realmoney.com.my
1 ポイント·投稿者 stealthcat·2 年前·0 コメント

Show HN: Slope -- a small ML library with IREE and StableHLO MLIR backend

github.com
2 ポイント·投稿者 stealthcat·3 年前·0 コメント

Ask HN: Traning LLM directly on file bytes

2 ポイント·投稿者 stealthcat·3 年前·0 コメント

[untitled]

1 ポイント·投稿者 stealthcat·3 年前·0 コメント

コメント

stealthcat
·2 年前·議論
Marketed as “multimodal” but actually texts and images.

Multimodal dataset should be multimedia: text, audio, images, video, and optionally more like sensor readings and robot actions.
stealthcat
·2 年前·議論
This worldly life is no more than play and amusement. But the Hereafter is indeed the real life, if only they knew. [1]

Know that this worldly life is no more than play, amusement, luxury, mutual boasting, and competition in wealth and children. This is like rain that causes plants to grow, to the delight of the planters. But later the plants dry up and you see them wither, then they are reduced to chaff. And in the Hereafter there will be either severe punishment or forgiveness and pleasure of Allah, whereas the life of this world is no more than the delusion of enjoyment. [2]

[1] https://quran.com/29:64

[2] https://quran.com/57:20
stealthcat
·2 年前·議論
Unitree is better than Optimus.

Qwen et al is better than Grok.

BYD is better than Tesla.
stealthcat
·2 年前·議論
Did you have some small toy experiments to prove this?
stealthcat
·2 年前·議論
Should list most of the technical debt accumulated so far and rank them. At this stage, lots of corners have been cut.
stealthcat
·2 年前·議論
Are there decent wind models and motor models if you want to model drones
stealthcat
·2 年前·議論
Imagine all those millions of $$$ GPU of cloud credits for training, only to overlook this bug
stealthcat
·2 年前·議論
Consider applying PhD program to some ML labs in Europe other than UK and France, they pay livable salaries like junior engineers. Better if in Swiss since the salaries are much higher.

European ML labs generally have less prestige than US/China ML labs but they pay much more humane wages even for people with family.
stealthcat
·2 年前·議論
Eat lutein supplements and get as much sunlight and barefoot grounding
stealthcat
·2 年前·議論
Kotlin Multiplatform.
stealthcat
·2 年前·議論
In ML the problem is passing 0D scalar tensor as 1D 1-element tensor.
stealthcat
·3 年前·議論
How do you trust numbers to “quantify extremely smart people”?

Interviewers know to hire/no-hire within first 10 minutes of talking to candidates.

Similarly, the best LLMs nowadays are probed only with some prompts.

Just last time I see LLM midwit meme on quantifying best scores on all benchmarks vs idiots/geniuses who just prompt something for some while.

Do not try to put numbers on everything.
stealthcat
·3 年前·議論
More like GodotDesktop
stealthcat
·3 年前·議論
Fun exercise… but for such use cases, people use Casadi and DrJit

https://github.com/casadi/casadi https://github.com/mitsuba-renderer/drjit

DrJit is made by same author of pybind11 and nanobind.
stealthcat
·3 年前·議論
Isn’t this more of symptom of modern western values?

Try a book by Muhammad Asad (Leopold Weiss), like the Road to Mecca.

Every year during hajj in Mecca, there’s profound view of sea of humans wearing white from vast different backgrounds, wealth, health, ethnicities, continents. You can find Japanese, Africans, Australians, Russians, Indonesians, etc, variety of skin colors and builds, in one place.

Put the world in your hand, not in your heart.
stealthcat
·3 年前·議論
Kotlin multiplatform
stealthcat
·3 年前·議論
Quran
stealthcat
·3 年前·議論
Good code? Code reading is more like understanding different cultures. Read big tech codebases meta and google, to startups codebases like openai and comma.ai, you will see.
stealthcat
·3 年前·議論
Because you work in semiconductor industry.
stealthcat
·3 年前·議論
Software engineering is hard problem. How is Julia in multiple authors codebase? A language need to consider ergonomics of multiple users working on same thing. Reading >> Writing.

C++/Python still hold ground in this regards sadly. One side for low-level control, another side for glue code.

Related: https://kidger.site/thoughts/jax-vs-julia/