WKBae ๐ฏ๏ธ retweeted
์ ์๋ฌด๋ Neon์ผ๋ก Neon ๋ง๋ค๊ธฐ ๊ฐ์ ๊ฑธ ์ํ๋๊ฑฐ์ง
Gemini 4 Neon ยท xhigh | 48min | 1st attempt
WKBae ๐ฏ๏ธ retweeted
ํธ๋ผํ, ๋ฏธ๊ตญ ๋ด ์์คํ
์ฅ์ ์์ถ์ "๊ด์ฐฎ๋คโฆ๊ทธ๊ฒ ์ธ์"
yna.co.kr/view/AKR2026111802โฆ
The harness is a crucial lever on model behavior. Compared with a Bash-only toolkit, native harnesses generally raise both solve rates and costs.
However, this effect isn't uniform. In Codex, 6.1 Terra and 6 Luna gain 10 points, while the newer 6.2 models see little change in solve rate. Grok and DeepSeek make big gains from their harnesses, while Muse stays flat. Unlike others, Fable 5.2 is much cheaper in its native harness.
WKBae ๐ฏ๏ธ retweeted
์๊ธฐ๊ฐ ์ปจ๋ํ ๊ธฐ๊ณ๋ ์๊ธฐ๊ฐ ๋ง๋ค์ด์ผ์ง ์ฌ๋ค์ฐ๋ ๊ฑฐ ๋ณด๋ฉด ์ด ๋๋ผ ์ ์๊ณตํ์ ๋ฏธ๋๊ฐ ์ฐธ ์ด๋ก์ง์ด
Meet Gemini 4 Hydrogen.
It brings cutting-edge intelligence to everyone, with speed and efficiency for everyday coding and agentic work.
Across 4 thinking levels, it stands on its own or teams up with Carbon and Argon โ 2.5X faster than 3.8 Flash, outpacing competing models in its class.
WKBae ๐ฏ๏ธ retweeted
introducing the Gemini Decisions API
powered by Gemini Decision 1, our new multimodal decision model
get fast, calibrated decisions on up to 1000 questions at once across text, images, audio, video, and PDFs
try it today in AI Studio and via the Gemini API ai.studio
Anthropic quietly dropped a fascinating finding in the new Claude Fable 5.2 prompting guide.
When you give Fable 5.2 a coding task without saying who owns the repo, it assumes it's a guest in someone else's codebase. So it plays it safe: it bolts new functions onto the existing structure, leaves duplicate code alone, and merely mentions nearby bugs instead of fixing them.
Adding this single sentence to your system prompt or CLAUDE.md completely changes its behavior:
"This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it."
In Anthropic's internal tests, adding that one line made Fable 5.2 proactively merge duplicate code and clean up related bugs in ~50% of runs, compared to under 10% without it.
WKBae ๐ฏ๏ธ retweeted
์ ์จํฉ ๋ถ๋ฅ ์ฌ๊ฑด์ ๋ณด๋ ๋ฏธ๊ตญ ๋ณธํ ํธ์ํฐ
dogdrip.net/733938412
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
ใ
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
Theyโre capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
WKBae ๐ฏ๏ธ retweeted
่กๅ็ใซไฝใฃใฆใใพใฃใ...
ใใใฏใฌใฃใฆ200่ฒใใใญใใฒใผใ
hachiware-iroate.pages.dev/
#ใกใใใใใกใณใขใผใ
This Post is from a suspended account. Learn more
WKBae ๐ฏ๏ธ retweeted
์ค๋ ํ ์ค๋ฅผ ์ญ์ ํ๋ค. ์ก๊ธ์ด ํธํด์ ์ฐ๊ธฐ ์์ํ ์ฑ์ธ๋ฐ ์ฒซํ๋ฉด์ ์ก๊ธ์ ํ ์ ์๊ฒ ๋์๊ธฐ ๋๋ฌธ์ด๋ค. ์ ํธ์์ ์ฐ๋ ์์ ์ ๋นํด ์ํ ์ฑ๋ค์ ํจ์ฌ ๋์์ง๊ฑธ ๋ณด๋ฉด ์ฌ๋ฌ ์๊ฐ์ด ๋ ๋ค.
WKBae ๐ฏ๏ธ retweeted
We built Pando, a new Git backend for our monorepo, to handle more pushes and fetches without changing how our engineers and agents use Git.
Pando cut push latency by ~80% and now serves nearly a petabyte of Git data a day. We can grow Git capacity alongside agent usage.
WKBae ๐ฏ๏ธ retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! ๐งต
Blog: rikiyat.github.io/blog/weighโฆ
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-spโฆ
github.com/RikiyaT/lm_undโฆ
huggingface.co/Qwen/Qwen3.5-0โฆ
huggingface.co/RikiyaT/Qwen3.โฆ
trivia:
โข started working on the game as a side project 4 days ago, landing ~1,150 PRs
โข every time i found an issue in the code i asked bot to make a lint rule against it
โข used TLA+ heavily to find bugs in the game's various state machines: the story, cutscenes, the game's systems. there are many formal verification skills, one for each major system
โข i put context into the filepaths: code that runs every frame or every tick gets a .hot.ts extension, and a lint rule bans allocating in it (no object literals, closures, new, etc.), so the game loop creates almost no garbage and GC doesn't cause stutters
โข game engine is mostly custom. three.js draws the 3D scene, but the simulation, post-processing and audio are all ours
โข the simulation is fully deterministic, so every run replays exactly from its inputs. when you post a high score, the server replays your whole run to check it isn't cheated
โข we generate dynamic og:images for every run based on your final stats, and it's all localized too! the game supports 10 languages
โข my favorite feature: the music seamlessly transitions into different music for the various biomes, continuing the previous melody
โข no audio files. all the music and sound effects are synthesized live in the browser by a custom Game Boy-style chip (two pulse voices, a wave voice, noise) plus a synthwave layer. the songs are written as text in code
โข all the pixel art is code. every sprite is a file drawn with one shared palette and packed into a sprite atlas at build time
โข it runs inside an X post, from the 245px card up to full screen. the bar is 60 fps with 500 enemies on a mid-range phone
โข the build fails if the first download goes over budget. it's about 124 KB gzipped right now
โข bots play the game headless to tune each difficulty against target survival rates
โข comments are banned by a lint rule. if code needs explaining, the fix is a better name or a test














































