WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev) profile picture

์ง€๋‚˜๊ฐ€๋˜ ์„œ๋ฒ„ ๊ฐœ๋ฐœ์ž 1

Seoul, Korea
Joined February 2012
Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)Photo by WKBae ๐Ÿ•ฏ๏ธ (@wkbae_dev)
WKBae ๐Ÿ•ฏ๏ธ retweeted
ใ€ŠๅŸบไบŽ็บฏ่ง†่ง‰่ฏ†ๅˆซ็š„AI็ญ็ซ็ณป็ปŸใ€‹
933
9,953
1,031
125,548
4,378,148
์™œ ์•„๋ฌด๋„ Neon์œผ๋กœ Neon ๋งŒ๋“ค๊ธฐ ๊ฐ™์€ ๊ฑธ ์•ˆํ•˜๋Š”๊ฑฐ์ง€ Gemini 4 Neon ยท xhigh | 48min | 1st attempt
7
13
5
80
7,887
WKBae ๐Ÿ•ฏ๏ธ retweeted
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.
8
3
4
75
14,007
WKBae ๐Ÿ•ฏ๏ธ retweeted
์ž๊ธฐ๊ฐ€ ์ปจ๋‹ํ•  ๊ธฐ๊ณ„๋Š” ์ž๊ธฐ๊ฐ€ ๋งŒ๋“ค์–ด์•ผ์ง€ ์‚ฌ๋‹ค์“ฐ๋Š” ๊ฑฐ ๋ณด๋ฉด ์ด ๋‚˜๋ผ ์ „์ž๊ณตํ•™์˜ ๋ฏธ๋ž˜๊ฐ€ ์ฐธ ์–ด๋‘ก์ง•์–ด
6
354
25
668
26,772
WKBae ๐Ÿ•ฏ๏ธ retweeted
@ ๊ทธ๋‹ˆ๊นŒ ๋ณ€ํ˜ธ์‚ฌ๋ฅผ ์„ ์ž„ํ•˜์ž๋งˆ์ž ์ด๊ฑธ ์ž์ˆ˜ํ•˜๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ์ด๋Œ์—ˆ๋‹ค๋Š” ๊ฑฐ์ง€? Ai๋Š” ๋ณ€ํ˜ธ์‚ฌ๋ฅผ ๋Œ€์ฒดํ• ์ˆ˜์—†๋‹คโ€ฆ
์„ธ๋ฌด๋ƒฅ์ด ai ์•ˆ๊ฒฝ ์ธ์ •ํ–ˆ๋„ค ์ž์ˆ˜ํ•œ๋Œ€
2
3,282
194
7,078
855,013
WKBae ๐Ÿ•ฏ๏ธ retweeted
there is smoke coming from my gpu, a new ceo has been chosen
208
7,146
476
66,433
839,774
WKBae ๐Ÿ•ฏ๏ธ retweeted
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.
395
732
323
7,090
780,415
WKBae ๐Ÿ•ฏ๏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
131
260
136
2,800
290,756
WKBae ๐Ÿ•ฏ๏ธ retweeted
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.
92
124
36
3,254
322,291
์  ์Šจํ™ฉ ๋ถˆ๋ฅœ ์‚ฌ๊ฑด์„ ๋ณด๋Š” ๋ฏธ๊ตญ ๋ณธํ†  ํŠธ์œ„ํ„ฐ 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…‹
7
721
49
1,186
72,277
WKBae ๐Ÿ•ฏ๏ธ retweeted
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. Theyโ€™re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
3,202
9,251
6,970
96,500
21,084,016
WKBae ๐Ÿ•ฏ๏ธ retweeted
Asked one of our developers how he uses his AI agents and he just sent me this:
179
1,337
270
16,202
1,779,593
WKBae ๐Ÿ•ฏ๏ธ retweeted
LMFAO a binance dev ops downloaded homebrew off a malicious google search ad as with 99% of these hacks... it comes down to a skill issue
108
104
49
1,704
162,547
WKBae ๐Ÿ•ฏ๏ธ retweeted
์˜ค๋Š˜ ํ† ์Šค๋ฅผ ์‚ญ์ œํ–ˆ๋‹ค. ์†ก๊ธˆ์ด ํŽธํ•ด์„œ ์“ฐ๊ธฐ ์‹œ์ž‘ํ•œ ์•ฑ์ธ๋ฐ ์ฒซํ™”๋ฉด์„œ ์†ก๊ธˆ์„ ํ•  ์ˆ˜ ์—†๊ฒŒ ๋˜์—ˆ๊ธฐ ๋•Œ๋ฌธ์ด๋‹ค. ์ € ํŠธ์œ—์„ ์“ฐ๋˜ ์‹œ์ ์— ๋น„ํ•ด ์€ํ–‰ ์•ฑ๋“ค์€ ํ›จ์”ฌ ๋‚˜์•„์ง„๊ฑธ ๋ณด๋ฉด ์—ฌ๋Ÿฌ ์ƒ๊ฐ์ด ๋“ ๋‹ค.
์‚ฌ์‹ค ์€ํ–‰์—์„œ ์ง„์งœ ์ •๋ง ์กฐ๊ธˆ์ด๋ผ๋„ ๋…ธ๋ ฅํ–ˆ์œผ๋ฉด ํ† ์Šค๊ฐ™์€ ์•ฑ์ด ๋ฐœ๋ถ™์ผ ํ‹ˆ์€ ํ•œํ‹ˆ๋„ ์—†์—ˆ๋‹ค๊ณ  ๋ณธ๋‹ค ์–˜๋„ค๋“ค ์ง„์งœ ์‚ฌ๋žŒ์ƒˆ๋ผ๋“ค๋„์•„๋‹˜
19
759
152
1,117
200,575
WKBae ๐Ÿ•ฏ๏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.
392
545
414
6,790
1,885,553
WKBae ๐Ÿ•ฏ๏ธ retweeted
-----BEGIN RSA PRIVATE KEY-----
29
244
4
6,944
105,109
WKBae ๐Ÿ•ฏ๏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! ๐Ÿงต Blog: rikiyat.github.io/blog/weighโ€ฆ
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-spโ€ฆ github.com/RikiyaT/lm_undโ€ฆ huggingface.co/Qwen/Qwen3.5-0โ€ฆ huggingface.co/RikiyaT/Qwen3.โ€ฆ
2,367
7,254
5,755
55,906
10,093,360
WKBae ๐Ÿ•ฏ๏ธ retweeted
trivia: โ€ข started working on the game as a side project 4 days ago, landing ~1,150 PRs โ€ข every time i found an issue in the code i asked bot to make a lint rule against it โ€ข 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 โ€ข 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 โ€ข game engine is mostly custom. three.js draws the 3D scene, but the simulation, post-processing and audio are all ours โ€ข 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 โ€ข we generate dynamic og:images for every run based on your final stats, and it's all localized too! the game supports 10 languages โ€ข my favorite feature: the music seamlessly transitions into different music for the various biomes, continuing the previous melody โ€ข 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 โ€ข 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 โ€ข 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 โ€ข the build fails if the first download goes over budget. it's about 124 KB gzipped right now โ€ข bots play the game headless to tune each difficulty against target survival rates โ€ข comments are banned by a lint rule. if code needs explaining, the fix is a better name or a test
203
187
92
3,964
548,925