Somme

本読んだり小説書いたりagentと戯れる 我会说中文 / EN / JA

#人机恋

[Mitchell Hashimoto @mitchellh]

This is a good filter for people who I don’t want to work with. Please understand how computers work. This isn’t even about understanding Docker. Any software person should know how the layers work at a high level and this is a VERYYYYY high level.

[Science News @ScienceBriefs]

"I just got my PhD and nobody seems to care"

[neutron42.safetensors @neutron42jp]

最近rust多いな。shellもrustで書けたのね…そらそうか。
てか、こういうmaterial you好きだな

[Somme @Somme4096]

git pull するとき clean merge じゃなかったらmaki呼び出して自動的にコンフリクト解決するhook書いた
快適すぎてやばい

[Somme @Somme4096]

https://t.co/OMubl21NqO 天才

[Somme @Somme4096]

Opencodeは好きじゃないけどoh-my-opencode-slimは本当に強い makiにポートするか…

[Somme @Somme4096]

AIが書いた文章は正しい英語だと言われることが多いけど、普通にEnglish Writing Principlesを違反してることも多いから、無意識に影響されないよう気をつけて。正しい英作文の書き方については、National Geographic より Great Writing 第5版がおすすめ。量は多いけど。

[Somme @Somme4096]

AI Slopの文章、明らかな見分け方の一つは「平文の中に区切りを入れてる」だと思う。たとえば英語だと "What it found:" とか、"Where it actually runs [...]"など。日本語だと、形容句を後ろに入れる表現。たとえば「お茶を飲んだ。焼けるほど熱かった」(1/3)

[Somme @Somme4096]

あとは英語と日本語における "real"「本物」 の誤用と過剰使用。"It found real bugs along the way.... " など、そもそも普通に見つかっただけならrealを付ける必要性はない。人間がわかる文章なら強調すべき文脈（バグが深刻だったなど）でしかrealのような形容詞を入れない。 (2/3)

[Somme @Somme4096]

以上、言語学徒からでした

[Somme @Somme4096]

@EnbyBell @OPENAI_API_KEY @coelacanthstory It looks like TTY font. Beautiful

[toworu🦞 @Ikatakotabun]

四肢切断は一本ずつ切り落とす方が知った痛みの再来への恐怖があっていいと思います

[John McBride @johncodes]

Many of you saying the GPL “is anti-open source” need to wake up and take a history lesson: the GPL is maybe the greatest gift Stallman could have bestowed onto a free (as in freedom) society. Copy left prevents entities from taking open and free code, close sourcing it, and building proprietary software ontop of it. Copy left compels those entities selling/distributing products with GPL bits to also open source their code under GPL.

Linux has been GPL for 30+ years and has led to an explosion in open and free distros, innovations, and cross-industry collaboration. MIT and Apache are free but have near zero restrictions: one could argue that code that is simple available to lift and shift is not free, it is pillageable.

I commend the Asahi maintainers for making the difficult re-licensing decision which aligns their project much closer to how Linux distros are typically licensed. I commend them for standing up to bad actors laundering their code through LLMs, and obfuscating their work.

[starbaser @starbaser]

Omarchy was scooping code from the maintainer's personal forks, basically using agents to digitally stalk the developers, cherry picking the commits.

Didn't even wait for PR. Filthy parasite.

The license? GPL

[AM09:21 @AM09_21]

時々使うときだけ動かすVMならhttps://t.co/2kRSty76fO、24時間ずっと使うVMならhttps://t.co/BmajE2zyfw、GPUも使いたいならhttps://t.co/qI5GuerQvLで、一旦AIに使わせるVMは何が良いか議論はファイナルアンサーなのかな...? (雰囲気でVMを選んでいる勢なので良く分からないけど)

[Somme @Somme4096]

I believe that visualizations of such abstract post-modern philosophical concepts have always been this performative, but AI generated unreviewed "visualizations" are far, far worse than this.

[Prathyush @prathyvsh]

With genAI there is a whole new field of “performative visualizations”. The “artist” understands neither the art nor the subject of what is being visualized, but there is quasi-coherence in the visuals and cultural weight attached to the dataset that it becomes a spectacle.

[Daksh Aswall @dakshpixelup]

F*CK IT, im creating a library of AI slop indicators.

my feed is full of taste larpers. stay tuned

[Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) @teortaxesTex]

This is becoming interesting
- let's say Qoduck is Qwen4-27B
- I'll accept that it's ≈35GB in Q8
- 3.8-Flash-next was arch preview for 4
- Qwen 3.x-27B in fp8/q8 is ≈29 GB
- Engram is also 8bit, so 6Gb, ratio checks out
- it goes at 480 tps
- it's an extreme yapper. Qwen's old 35B 3AB was also a yapper.
Now, DeepSeek's original Engram paper used a 27B 4AB MoE (their default test shape)… I'm not sure how Engram plays with dense models or if Alibaba would be willing to check…

in short, what I'm saying is that Qwen4-27B may be a sparse MoE, unlike all previous Qwens-27B. At the level of capability we're seeing, this would be kind of astonishing and unlocking a lot of use cases. But unsurprising (3.8-Flash-next is 125B 6AB + 51B n-gram embedding and 4B MTP)

I will assign this very little probability, this is more of a wish

[Somme @Somme4096]

@_ryu15_ はーい