SK (@Samking207) header
SK (@Samking207) profile picture

technology, gaming and Music

Maryland, USA
Joined December 2020
Photo by SK (@Samking207)Photo by SK (@Samking207)Photo by SK (@Samking207)Photo by SK (@Samking207)Photo by SK (@Samking207)
Introducing Gemini 4 Carbon, it's a big jump over Argon which we released just 2 weeks ago. Gemini 4 Carbon is SOTA on cyber, long horizon tasks, ranks #1 on Artificial Analysis V5, significantly improves on usability in domains like coding, and is more aligned.
1,223
1,156
855
15,860
1,224,731
PROMPT Build a single-file HTML page with three.js (ES module from jsdelivr, [email protected]) that renders a photoreal, fully procedural human eye in the browser. No textures, no 3D models, no audio files - everything generated in shaders and code. SCENE - Eyeball: unit sphere with a custom ShaderMaterial. - Sclera: warm off-white, pinker toward the edges, thin red veins from ridged fbm noise, pink caruncle in the inner corner. - Cornea: refract the view ray (IOR 1.376) onto a recessed iris plane so the iris shows parallax and depth. - Iris in polar coords: radial fibres (multi-octave noise), crypts (dark lens-shaped holes), a wavy collarette ring, a dark limbal ring, a pupillary ruff and pigment freckles. Use the iris height field for bump lighting, and add a caustic on the side opposite the light. - Wet surface: a rectangular softbox reflection on the cornea, a tiny sharp glint and Fresnel environment reflection. - Skin patch: an irregular blob-shaped sheet with a rounded thick rim, built from rings that run from an almond-shaped lid opening out to the edge, and rebuilt every frame on the CPU. - The lids hug the eyeball (smooth-max against the sphere), with an upper lid crease and a wet pink lid margin. - Skin shader: per-channel wrap lighting for subsurface scattering (red wraps furthest), pores from voronoi, fine crinkles parallel to the lids, crow's feet at the outer corner, freckles, mottling, an oily specular and a peach-fuzz rim. - Lid shadows on the eyeball computed analytically from the same almond function, plus striped lash shadows and a tear meniscus highlight along both lid margins. - About 500 eyelashes as camera-facing tapered ribbons: curled, clumped into groups, 2–3 rows, longer in the middle. Upper lashes rotate downward when the eye closes. - Background: warm brown gradient with soft diagonal light streaks and a contact shadow under the patch. BEHAVIOR - The eye follows the mouse cursor. When the cursor is idle, it makes random saccades with micro-tremor. - The upper lid follows vertical gaze. Natural random blinks every 2–6s (fast close, slower open). - Pupil hippus (slow breathing of size); it constricts after a camera flash and slowly recovers. - Clicking the eye grows a new random iris that crossfades from the old one. There are 8 base palettes (amber, hazel, green, blue, grey, teal, brown, ice) with jitter, and an occasional golden central heterochromia. POST - EffectComposer with 4x MSAA, BokehPass focused on the cornea, a subtle UnrealBloom on highlights, and a final pass with chromatic aberration, vignette, film grain and a light sharpen. Tone-map with ACES. UI (HTML overlay, monospace, uppercase, warm cream text with a yellow accent) - Top-left "eye exam" panel: iris swatches + "new" button, pupil slider with auto toggle, 6 skin tones, lash length, framing (macro / portrait / wide / space - in space the patch floats and rotates), focus (shallow / sharp), blink and squint buttons, volume. - Corner crosshair marks, slowly rotating thin dashed HUD rings, side readouts for gaze in degrees and pupil in mm, top-right resolution / fps / iris name. - Snapshot button: white flash, the pupil constricts, and a thumbnail of the frame appears. - Drag to orbit, scroll to zoom. SOUND (Web Audio API only, starts on the first click) - Ambient: a soft low drone (D3, A3, D4) plus a generative pad of sine notes picked from the D major pentatonic every 2–5s, with long attacks, convolution reverb (generated impulse response) and a feedback delay. - UI tick on every action: a 15ms high-passed noise click plus a short pitched sine blip from the scale. Quieter ticks on hover. - Blink: a soft low pitch-drop thump. New iris: a 4-note rising arpeggio plus a filtered noise sweep with tremolo. Framing change: a tremolo noise swell. Snapshot: a two-stage shutter click. - Footer sound on/off toggle with a small live level meter. Keep it 60fps on a normal laptop, make it work at phone width, and put everything in one
1
2
339
732
1,417
1,123
20,438
2,257,093
Replying to @alannnfx
still not close to opus
3
1
1
96
Gemini 4 Carbon checkpoint in the Arena under the name Gemini 4 Argon: >A freestyle swimmer doing the front crawl using Three.js (single HTML file)
I got few exceptional outputs from Gemini 4 Carbon checkpoint in Arena under the name ‘gemini 4 argon’ >Matching Opus 5.5 closely,even edging it out in a few cases.(as of now)
2
2
49
2,549
🚀 DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API. 🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model. 1/4
1,626
4,127
2,851
37,185
8,176,444
推荐一下 dsh-plugin-upgrade-skill 这个帮助 DSH 插件作者升级自己插件、适配 DSH 新版本 的 skill。 当然,长远来看,DSH 插件生态不能依赖于插件作者每次都要修改自己的代码来兼容 DSH 的新版本。这一方面需要 DSH 逐渐减少破坏性变更,另外也需要一些技术手段,让插件 API 升级后,面向旧 API 编写的插件仍然可以通过兼容层使用,类似于面向旧的桌面或手机操作系统版本编写的程序仍可在新版操作系统中运行。 dsh-plugin-upgrade-skill 这个项目是 oh-my-dsh 组织出品。oh-my-dsh 以及其相关联的 omdsh-dev 组织是最早建立的 DSH 插件作者的独立开发者社区,上面有挺多高质量 DSH 插件项目。 omdsh-dev 组织: github.com/omdsh-dev (这个页面上有精选项目的目录) 插件升级 skill: github.com/oh-my-dsh/dsh-plu… (oh-my-dsh 及 omdsh-dev 是独立开发者社区,与 DeepSeek AI 之间互相不存在隶属或授权合作关系。)
67
37
7
583
70,731
Congrats to the Claude Code team on shipping Mods! A cute way to make your agent harness extensible and customizable. It's a nice convergence of ideas: in Claude Code, mods are just plugins; and @DeepSeekHarness has been built around rethinking the concept of plugins from the very beginning. While a subset of Claude Code features are made available as mods, in @DeepSeekHarness everything is a plugin: models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and UI are ALL implemented as plugins, and can be mixed, matched, replaced, and extended. In @DeepSeekHarness's Creator Mode, you can prompt the agent to inspect the live runtime, write plugins, and persist them, in order to change the UI, modify agent-loop behavior and mechanisms, and add tools or new ways of orchestration — simply everything about the agent harness can be modified on the fly and persisted. So we've been thinking about the term Plugin.Engineering for a while… Or should it be Mod.Engineering now? 😂
Mods are absolutely insane. You can now customize Claude to work and look the way you want by just prompting it. Each person works differently, so there's no reason why everyone should have an identical Claude experience. Make Claude your own, and share mods as plugins so others can try your mods too.
56
44
12
741
124,440
NVIDIA IS GIVING 4 STRONG AI MODELS FOR FREE 🤯 no credit card required you can use: • DeepSeek V4.1 Flash • GLM 5.3 • GLM 5.3 Flash • Kimi K3 get access: build.nvidia.com base URL: integrate.api.nvidia.com/v1 hope this helps
FREE OPEN SOURCE CRYPTO TRADING BOT 🤯 someone built Cabbage, a trading bot you can: • backtest • paper trade • live trade • check before trading with Doctor Astra handles the reasoning Jev makes the BUY HOLD or SELL decision built with RSI and EMA signals, hard risk rules and SQLite state Apache-2.0 and completely free link: github.com/sopersone/cabbage… not financial advice, test it before using real money
88
211
19
2,944
459,221
🔥DeepSeek Harness 官方桌面端下载地址来了! 不用再装 Node、不用开终端、不用开浏览器,双击就能直接用官方 DSH! Windows 64位:
download.deepseek.com/dsh-de… Mac Apple Silicon:
download.deepseek.com/dsh-de… 官方说明文档:
github.com/deepseek-ai/deeps… 已经等了好久的朋友可以冲了,版本号带今天日期,新鲜出炉! #DeepSeek #DeepSeekHarness #DSH桌面端 #官方客户端 #AI编程助手 #崔添翼 #梁文峰
🚨重磅:DeepSeek Harness 官方桌面端已正式现身!Github官方打包好已可下载! 版本 0.1.7-rc.1.20260924.1,对得上官方仓库 tag dsh-v0.1.7-rc.1。欢迎页、本地工作区、账户余额、关于面板都是真界面。 要点如下: 🔹本地文件夹直接当工作区,搜资料、做表格、写代码、排障,不用守着浏览器标签。 🔹官方 Desktop 已在仓库 apps/desktop:Electron 壳 + 内置 Node/pnpm/Python,不先装环境也能开。 🔹上手三步:开始设置 → 选工作区 → 充额度。Token 按实际执行扣,不跑任务不花钱。 🔹安装包还没大规模官宣,社区已从 GitHub 挖到可装版本。认准 deepseek-ai/deepseek-harness,别下不明来源包。 Agent 终于能双击干活了。 #DeepSeek #DSH #DeepSeekHarness #AIAgent #桌面客户端 #梁文峰 #崔添翼
9
37
5
210
59,366
A classic mathematical theorem. 6,000+ lines of Lean code. Verified by the kernel. Xiaomi MiMo 2.6 Pro assisted researchers in fully formalizing the original main theorem of Li–Yorke’s classic paper Period Three Implies Chaos in Lean 4. Guided by a research-designed exploration strategy, multiple Subagents collaborated on theorem formulation and proof formalization. After revision and integration, the project spans 6,000+ lines of Lean code, verified by Lean’s kernel with no unfinished proof placeholders. Notably, Xiaomi MiMo 2.6 Pro was not specifically post-trained for Lean. #XiaomiMiMo #XiaomiMiMoV26 #AIforScience #XiaomiAI
7
42
10
594
58,910
MiMo-V2.6: The Hard Road to Scaling Up RL MiMo-V2.6 is very likely one of the largest single RL runs, by compute, that any open-source model team has undertaken to date. In an era when compute is brutally scarce, we still chose to dedicate a team of several dozen people to one goal over an extended period: scaling up RL. That takes more than research conviction. It takes a vision for AGI, respect for the unknown, and the nerve to walk straight into the hardest problems. The result is a model whose potential was built through mid-training and unlocked through heavy RL. Today, it is the number one open-source model. I strongly recommend reading the technical report. I believe it will become one of those papers that Agent RL practitioners keep reopening and discovering something new in each time. In my view, the research innovations and engineering challenges behind it surpass those of DeepSeek R1, which I was partly involved in. Some will ask: why MixRL instead of MOPD? First, they are not competing choices. We ran MixRL on verifiable tasks of moderate difficulty, including code and related agentic tasks, and found that the resulting models generalize remarkably well. Second, tasks that are difficult to verify, extremely long-horizon, or simply too challenging to include in a joint RL run are trained separately. Including them would substantially reduce rollout efficiency or introduce significant rollout staleness. We then merge the resulting capabilities through MOPD. Games, 3D tasks, and tasks with subjective evaluation signals all fall into this category. There is also a third, slightly cheeky answer. Our team is flat enough and free enough of organizational silos that MixRL simply is not difficult for us. More importantly, everyone enjoys working this way. People from different domains come together every day, driven by the pursuit of AGI and intelligence that can continuously improve itself, to confront and resolve the RL bottlenecks in each field. I will always remember the RL daily update meetings from this period. They were intense and dense, with intelligence emerging in real time. To help the open-source community focus on solving real Agentic RL problems, we have released a Qwen model distilled from MiMo RL trajectories as a stronger starting point for RL, along with 7K diverse environments and a complete RL training framework. We hope these resources will help move Agentic RL research forward. MiMo-V2.6 is only the beginning. In an era when intelligence is easy to replicate, we still choose the hard road toward self-improvement and AGI. Much of what lies ahead remains unknown. But we are willing to keep investing the time, compute, and passion required to take on one hard problem after another and work each of them all the way through, until intelligence crosses into a new regime.
238
354
131
4,319
676,663
Introducing Xiaomi MiMo-V2.6 — Pro & Flash. Frontier intelligence, all the modalities, built in public. 🔹 Two omnimodal models, advancing through scaled reinforcement learning 🔹 Pro performs on par with Claude Opus 5 and GPT-5.6 Sol across most agent benchmarks 🔹 Pro scores 46 on the Artificial Analysis Intelligence Index — the highest among open-source models 🔹 Stronger coding, computer use, 3D reasoning and creative capabilities 🔹 Open model weights, technical report, RL environments and training code Blog:mimo.xiaomi.com/mimo-v2-6
383
868
485
8,830
1,528,596
Nearly half a year of silence. We spent it studying one problem: how far RL can scale. MiMo-V2.6 is in the middle of its RL run right now. Three things we scaled: compute (~2B tokens per step, 1568 prompts × 16 rollouts, fully async), environments and harnesses (multi-task agentic RL, mixed across multiple harnesses in one run), and grader compute (agentic in-group credit assignment, with test-case and rubric-based rewards). We'll open-source the details piece by piece over the coming weeks. Streaming the run: mimo.xiaomi.com/rl/
423
977
555
9,888
3,136,259
Been using Deepseek v4.1 Flash via OpenCode. I've noticed that adding this simple line after my prompt drastically improves the final output: "Upon completing your work, spawn a subagent to do a full and thorough quality pass through it for any remaining bugs, issues or QoL improvements" The code is much more mergable and needs fewer review-bot rounds.
46
45
3
1,025
51,197
Replying to @farzyness
Grok 4.7 needs a few more days to cook. We might have penalized response length too much (or something) in RL, as it still gives up on hard tasks (that it can do!) too early and isn’t yet sufficiently rigorous in checking its work.
1,550
957
851
19,903
2,146,891
🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6
1,110
3,081
1,745
28,781
7,179,866