creaτepjf ⚡️ 🟦 (@createpjf) header
creaτepjf ⚡️ 🟦 (@createpjf) profile picture

Ops & Producer @flock_io i'm vibing on coding RIGHT NOW. chief experiece officer of my life ⚠️ Transparency Over Honesty...

Dust
Joined February 2022
Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)Photo by creaτepjf ⚡️ 🟦 (@createpjf)
creaτepjf ⚡️ 🟦 retweeted
Small models, huge impact. Excited to power @flock_io's high-performance decision model with Qwen & our cloud infrastructure.
🤝 @Alibaba_Qwen models and @alibaba_cloud infrastructure power FLock's specialised AI training. THIS / THAT Model, our open-source small decision model built on Qwen3.5, outperforms frontier LLMs in speed and accuracy, using our reliable synthetic data methods, peer-reviewed at ACL 2026.
4
20
12
61
39,964
creaτepjf ⚡️ 🟦 retweeted
谁能想到Jev 居然输给了一个 1.88B 本地开源模型🔥 结果真的有点让我意外: 我做了一个贪吃蛇决策游戏 Jev 30:28 险胜,但是换成同一组 68 道决策题,本地模型准确率直接做到 94.1%,Jev 只有 76.5% 一个错了 4 道题,一个错了 16 道题 我本来只是想测试一下 “小模型决策能力怎么?”,结果我抽完发现,小模型虽然小 但是准确率一点也不低。 1️⃣ 先说我是怎么测的 双方使用完全一样的 12×12 棋盘、随机种子、碰撞过滤和路径规划器,只有遇到多个方向都能走的时候,才把选择交给模型。 最后 Jev 吃到 30 个食物,跑满 256 步 本地模型吃到 28 个,在 239 步被困住。 如果单看这一局的成绩,肯定是Jev 赢了。 但是贪吃蛇只是某一步选错以后,后面整条路线都会变,所以我又补了固定的 68 道决策题。 结果 this-that-model-1.0 答对 64/68,94.1%;Jev 是 52/68,76.5%。 Brier Score 也分别是 0.042 vs 0.133,本地模型给出的概率和真实结果更贴近 2️⃣ 这个模型最离谱的地方,是它只有 1.88B this-that-model-1.0 是一个开源的结构化决策模型,MIT License,可以直接在 Mac 本地跑 它和 Jev 的思路其实很像: 不给你写小作文,也不慢慢吐 Token 给它当前状态、一个问题、几个候选答案,它一次前向直接返回: 选哪个 + 每个选项的概率。 所以任务分流、选 Skill、队列选择、标签判断、Agent 下一步动作这种事情,都很适合 我在 M4 上直接用 MPS + FP16 跑,不需要 API Key,也没有网络请求 不同任务中位延迟大概 225~475ms,Snake 实战中位 475ms,平均约 492ms 一个 1.88B 模型,在 Mac 上已经可以做到亚秒级决策。 这个我觉得挺夸张 3️⃣ 我又单独测了 105 道空间决策题 总准确率 84.8%,随机基线只有 34.3% 其中 Snake 状态题、行列文字状态都是 100%,JSON 状态也有 94.4%。 而且输出特别干净。 没有 JSON 少个括号,也不会突然给你解释人生,只会老老实实从候选答案里面选一个,再把概率交给程序 对于 Agent 来说,这种“无聊”反而挺值钱。 4️⃣ 当然它也不是万能的 完整地图测试只有 60.7%,最短路径距离题更是掉到 28.6% 一旦任务需要真正的多步图搜索、连续计算,它就明显不行了 所以这次结果也不能简单理解成: 1.88B 打败 Jev 贪吃蛇单局上是 Jev 赢了,而且 Jev 这次没有保存同条件延迟数据,速度也没法公平比较 真正让我感兴趣的是另一件事: 很多 Agent 里面那些高频、重复、答案明确的小判断,可能根本没必要每次都调用一个大模型,甚至连云端 Jev 都不一定需要。 一个 1.88B、本地运行、一次前向就结束的模型,已经可以完成绝大多数简单的决策了。 一些无关痛痒的简单的决策都可以交给这种小的本地模式、遇到复杂的任务再交给 GPT、Claude这样的大模型可能会更适合。 我现在越来越觉得,未来 Agent 不一定是“所有事情都找最强模型”。 更可能是: 小模型负责做大量的决策,大模型才是真正的为了我们的工作服务。 如果你们感兴趣 也可以部署一下,再偷偷地补充一句 这个可是很牛的牛津团队开源的: huggingface.co/flock-io/this…
26
18
6
97
36,933
creaτepjf ⚡️ 🟦 retweeted
JUST IN: FLock releases this-that-model-1.0 with a 94.1% success rate on our 68-question decision test, outperforming GPT-5.6, Deepseek-4.1-flash, Kimi K3, Jev, and more. Runs locally. Zero output tokens. Open-sourced. Best in class.
10
14
5
69
7,366
creaτepjf ⚡️ 🟦 retweeted
The Broken Bargain: How the SEC’s New Crypto Rule Could Quietly Reshape the Way Companies Raise Money
3
5
5
23
30,380
creaτepjf ⚡️ 🟦 retweeted
UNDP officially launched its Blockchain Advisory Group, with FLock joining as a key member organisation. Together with 25 other organisations, including the @ethereumfndn, @Cardano_CF, @AvalancheFDN, @SuiFoundation, and @StellarOrg. Here is what we've been building with @UNDP.
2
7
2
39
2,621
creaτepjf ⚡️ 🟦 retweeted
FOMO Reward Season 1 is live. AI inference now fuels an onchain reward loop on @base. Rewards are now open for users to earn and claim — through real AI usage. This is not an incentive program for speculators, but real AI users.
3
9
7
47
12,002
creaτepjf ⚡️ 🟦 retweeted
Back in February this year, we had a great chat with Sam Majid, Head of the National AI Office of Malaysia. And this week, I’m glad to see so many of the initiatives we discussed landing with FLock!
3
1
21
94,369
creaτepjf ⚡️ 🟦 retweeted
FLock.io is grateful to be chosen as the sole technical partner in an MoU signed by the Sarawak Artificial Intelligence Centre (SAIC) in Manchester — expanding our strong collaboration on sovereign AI in Sarawak through federated learning and privacy-preserving AI infrastructure.
21
14
1
65
4,450
creaτepjf ⚡️ 🟦 retweeted
Alongside the UN and Oxford event in Seoul, We also have a @daehan_base agent hackathon ongoing.
🤖 Made with AI
5
1
28
3,043
creaτepjf ⚡️ 🟦 retweeted
🎥 LIVE NOW: Nuts and bolts of AI agents—how to build your first one Presented by FLock's founder and CEO, @0x7SUN, joined by colleagues from @UNDP and @UNDP_AltFinLab. teapawt.org/i/broadcasts/1nxnRYAlZ…
19
3
14
1,278
creaτepjf ⚡️ 🟦 retweeted
What does sovereign AI look like in practice? FLock worked with the Sarawak AI Centre (SAIC) to show how federated learning can power AI in the public sector while keeping data and computing power under regional control.
19
6
2
37
6,346
creaτepjf ⚡️ 🟦 retweeted
Our Head of DevRel, @YifanX, just took the stage at UK AI Agent Hack with the keynote: "AI Agents for Good." While everyone's debating whether AI will end the world, we're here building AI that can improve lives. Our work with @UNDP on public healthcare in Mauritius, climate insurance for LATAM farmers, GenAI for Moroccan artisans, and more shows this isn’t hypothetical, it’s practical and happening now. But we can't do it alone. That's why we’re teaming up with @ChainforGood to launch the OpenClaw for Good track. Join us to build responsible AI: 🎯 Build AI Agents to solve UN Sustainable Development Goals 🔓 Use @openclaw and open-source models from FLock API Platform 💰 $5,000+ in prizes Submissions close on March 7th.
5
7
5
43
4,035
creaτepjf ⚡️ 🟦 retweeted
Learn how to use FLock API Platform as your AI model gateway in @openclaw 🦞 → Access to multiple AI models through one API → OpenAI-compatible interface → Ultra-low pricing and high-performance → Easy API key authentication
3
4
1
23
2,801
creaτepjf ⚡️ 🟦 retweeted
FOMO unlocks a new way to ship with AI. Better open models for real workloads, cost-efficient inference through the FLock API Platform. Here’s what’s now live in the FOMO roster. ↓ 🟦 DSIKH - DeepSikh (powered by DeepSeek-V3.2) - Built for efficient reasoning and agentic tool-use in real workflows. - Best when you need multi-step thinking, tool calling, and consistent behavior inside interactive agents. 🟦 CODING - Qwen3-30B-A3B-Coder (powered by Qwen3-30B-A3B-Instruct) - MoE, 30B params, strong performance per cost. - A clean pick for shipping code faster without paying flagship-tier inference costs. 🟦 QMXAI - QMXAI (powered by Qwen3-30B-A3B-Instruct) - A balanced model built for repeatable production workloads. - Ideal for apps that need stable instruction following across mixed tasks and pipelines. 🟦 QMINI - Qwen3 Mini ManCang (powered by Qwen3-30B-A3B-Instruct) - Optimized for high-volume flows where latency and efficiency matter most. - The default route for scale when you want strong output at a lower cost per call. 🟦 QWFIN - Qwen3-235B-QWFIN (powered by Qwen3-235B-A22B-Thinking) - Flagship MoE-architecture model from Qwen with top-tier reasoning, coding, and multilingual output. The next wave of AI is specialized models for specific apps. FOMO and FLock API Platform form the distribution layer that turns AI usage into an open onchain economy. Lower AI costs for builders. Exposure to model demand for token holders. Real Model Assets on FOMO.
9
1
23
2,213
creaτepjf ⚡️ 🟦 retweeted
收到了自家 @flock_io 的年度礼盒,开个箱 内含:冲锋衣,护照夹,多国转换插头(这个是真刚需),行李牌,日历 最近行情很差,手上的大部分现货都已经质押且处于大幅亏损状态,得装死躺平一段时间了 打算出门玩一玩,顺带体验下这个转换插头 手里还有两套周边,送给各位还在亏损的兄弟 老规矩: 转发+评论就行 抽奖工具:easy3.app 开奖时间:2026.1.22 22:00
315
179
2
251
25,502
creaτepjf ⚡️ 🟦 retweeted
The long awaited last piece to complete the flywheel puzzle has now arrived! It’s not another launchpad, not another model platform, but the first to bring tokenized model economy to real life.
Introducing FOMO - FLock Open Model Offering The full-cycle decentralized AI model economy is now on @base, powered by FLock.
1
3
21
1,816
creaτepjf ⚡️ 🟦 retweeted
坚守币本位长线策略,他们的逻辑是什么? 众所周知,飞轮有两位坚定的币本位老师:王短鸟死磕 $OKB 、孤鹤梭哈 $HYPE 。 我也再次理解了什么是真正的 「 币本位 」思维。简单说,围绕本币不断构建的生态增长反哺可以形成价值飞轮,那就是牛逼的币本位。 还有其他币可以这样玩吗? 让我们潜入! ⬇️ OKB 和 HYPE 都在讲的是生态故事,一个围绕 Xlayer 一个围绕 Hypeliquid,后者更明确、更清晰,因为协议收入直接回购代币构建通缩模型已经被验证,OKB 还需要一些时日验证。币本位主线的策略可能是未来的主流建仓思路,因为超级大币的波动性已经逐渐平缓,机构级资金的杀入更是让散户连汤水都吃不到。 如何寻找建仓标的? 监测提示我 @flock_io 改了 Bio ,接连预告了好几天要上一个新的功能 「 FOMO 」,K 姐 @0xKirara 透露进度是 80% ,赶在上线之前我先来「 爆个料 」,分享我的一点观察和思考。 什么是 FOMO FOMO 是 Flock Open Model Offering 简称,字面意思看像是一个 AI 模型的发射台,但根基应该是 FLock 平台。为什么 FLock 要做发射台呢?市场上有那么多发射台了,去 Pump 发一个很简单啊,这就得深入了解一下 AI 模型的运营模式了。 在传统模式下,假设你做了一个学英语的 AI 模型,用户需要付费解锁高级功能,那么你需要给大厂交纳托管、调用、部署等一堆费用,简言之,你辛辛苦苦创建一个 AI 模型,最大价值捕获者是 AI 巨头们。 FOMO 模型就让任何人可以把自己的AI模型上架发行一个专属的模型代币,用这个代币换取FLock 平台上的托管服务。打个比方,这就像开一家小店,店主可以发行店铺股票,卖股票换钱来付房租, 店铺赚钱了,收入会回馈给持股人,也就是 $FLOCK 持有人,当然店铺经营得好也会得到奖励。 这就是老板 @0x7SUN 说的 FOMO 将会推动 AI 模型转化为 Micro-economy 意思,FOMO 完成了 AI 模型从训练、托管到货币化的全生命周期,强调代币化 tokenized intelligence ,把 API 使用收入回馈给代币持有者/质押者,奖励构建者、托管者和用户。 如此, $FLOCK币本位长线逻辑建立了,即,模型的代币发射或募资是用 $FLOCK计价和执行的,模型推理产生的部分收入会自动回购 $FLOCK,FOMO 发射台项目越多、模型推理需求越大、协议收入越多、回购 $FLOCK越多、通缩越强、币价越健康向上。 对于有着大量 AI 推理需求的 Web2 公司来说,FOMO 模式也有实操意义,因为 FOMO 是不改变前端流程的基础性操作的。在具体一点,Web2 公司依然使用现有的技术接入模型,不需要任何变化, FLock 自动把收入拆分分配给 FOMO 参与各方。简言之, Web2 企业使用模型长期综合成本低、模型效果甚至更好,如此,极度自然舒适地就把 Web2 接入了 Web3,完美。 模型发射 + 推理 → 协议收入 → 回购 $FLOCK → 通缩 就是这个简要逻辑了。 / 模式挺好,我又不会技术,如何参与? 最简单的模式就是考虑把 $FLOCK 作为币本位长线策略,质押、支持 AI 发展,博弈点是 FOMO 飞轮能否转起来,这才是最重要的认知。当然,你也可以直接博弈 FOMO 发射台发射的模型资产代币,俗称打新,也许会有新的龙头出来,造富故事有了人自然就多了。 如果你碰巧会技术,或者碰巧有一个绝妙的 AI 模型创意,vibe coding 也许可以帮助你解决很多问题,如此,就可以直接在 FOMO 平台公平发射代币,募资或换取模型推理费用、平台激励等等。 FOMO 让 AI 模型的经营方式发生了根本性的变化,以前是纯开销,睁开眼就要付费,现在,它变成了有收入支撑、可估值的代币化资产,模型越受欢迎、价值越大,且,完全去中心化运营。 FOMO,会让你 FOMO 吗? 祝我们好运! / 作者:anymose | 一个软核科普作家 本文仅做科普使用,不构成任何投资建议,永远记得 DYOR!
30
3
1
44
10,644
creaτepjf ⚡️ 🟦 retweeted
Learn about the diverse digital assets we’re considering for future investment products and explore those already part of our offerings in our latest Assets Under Consideration update. Are we missing anything? 🤔 Read the full report: grayscale.com/blog/general-u…
263
221
146
1,054
318,223