Zehua Cheng (@zehuac) profile picture

Chief [email protected] PhD@UniofOxford

Oxford, England
Joined November 2018
Photo by Zehua Cheng (@zehuac)Photo by Zehua Cheng (@zehuac)Photo by Zehua Cheng (@zehuac)Photo by Zehua Cheng (@zehuac)Photo by Zehua Cheng (@zehuac)
The visual effects of presenting both papers are quite interesting. We have more paper presented in CIKM 2026.
Our Chief AI Scientist @zehuac was in Rome to present six FLock papers in the @CIKM 2026 Full Research Track. Together, they map out what we need to make private AI training affordable, safe, and reliable at scale. 1. Cheaper training Most training updates the whole model at once, even though the parts affect each other. Gradient Interaction Graph Optimization updates them in a smarter order and reaches the same quality in about a third fewer steps. 2. Safer model merging Combining two models can work well or ruin both, and normally you only find out after running expensive tests. Curvature Graph Mergeability predicts the outcome in advance from a quick scan of each model. 3. Robustness to bad examples Showing misleading examples in a prompt can throw off the whole answer. In-Context Learning as Belief Propagation lets the model recognise the odd one out and trust it less, improving accuracy by about 7 points. 4. Keep structured outputs consistent When AI fills in a form, it often decides each field separately, so a junior job title can end up next to a senior salary. Bethe Consensus Decoding checks the fields against each other and returns a consistent answer, with no retraining. 5. Quantify uncertainty A model should be able to say “it’s one of these three” when it isn’t sure. Cavity-Calibrated Prediction Sets give guaranteed-reliable shortlists that are up to 58% tighter, so they stay useful. 6. Anonymise training text Real-world text is full of names and personal information that regulations keep off-limits. Graph-Driven Context-Preserving Anonymization removes 99.8% of it while keeping the text just as useful for training. Private AI becomes more critical as data becomes every organisation’s edge, and FLock is building at that frontier.
2
12
1,165
Zehua Cheng retweeted
Founders are so excited after speaking with FLock Because they know This / That gonna help them solve their problems and OWN the model! every company can be a model company. Just like everyone can be a 5min comedian. My dm’s open. Flock it.
1
2
3
24
4,274
Zehua Cheng 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
40,124
Zehua Cheng retweeted
We post-trained @Alibaba_Qwen's Qwen3.5 2B to build THIS / THAT Model 1.2, our open-source decision model, specialised for decision-making, routing, classification, and scoring. It scores higher than every hosted model we tested across 1,710 decisions and responds in 31.4 ms. That’s around 32x faster than Claude Opus 5. Training specialised models for specific business needs will become the norm.
19
6
2
54
4,523
Zehua Cheng retweeted
FLock joins @anam145official as the Main Partner for TRUST404 🇰🇷 From blocked transaction receipts to zero-trust approval logs and tamper-evident agent payments, 17 teams turned security ideas into working systems. Together we keep pushing for trust, verification and safer Web3 infrastructure.
1/ 🔐 Protecting the key is not enough You have to verify who asked for the signature. TRUST404 Web3 Security Hackathon 2026 Demo Day, September 29, Dreamplus Gangnam 322 people joined, and 17 teams brought working builds across 5 tracks. Trust Not Found. Build It.
6
5
31
4,225
哈哈哈感谢社区支持
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…
1
319
Zehua Cheng retweeted
truly amazing Paper on arvix preview landing soon.
7
20
981
Zehua Cheng 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
7,371
Congratulations to Moderna and Merck on their landmark Phase III results. This milestone underscores a broader shift in life science: data, models, and therapeutics are becoming deeply interconnected. We need to understand not only what a biomolecule is, but also how it moves, changes, and transitions between functional states. @flock_io latest new two papers accepted at #ICDM2026 explore exactly this direction in AI for protein dynamics: • Multi-timescale protein dynamics — disentangling fast local fluctuations from slow global conformational changes, enabling better use of abundant short MD simulations and scarce long trajectories. • Generative modeling of protein transition paths — moving beyond stable conformations to model rare, functionally important transitions between molecular states. We believe the next frontier of AI for biology lies beyond static structure prediction: toward dynamics, mechanisms, and ultimately more powerful foundations for drug discovery and precision medicine.
2
2
11
3,879
😄 Great success to our R&D team!
BREAKING: Six FLock research papers have been accepted to the CIKM 2026 Full Research Track. This is a major research milestone for FLock.
9
434
Zehua Cheng retweeted
Only FLock can do.
BREAKING: FLock is officially a @RedHat ecosystem partner. This is a major step in expanding enterprise-grade federated learning to the cloud-native infrastructure trusted by governments and Fortune 500 companies.
1
6
339
Zehua Cheng retweeted
So we got…. FLOCK papers published in ACL (San Diego), ICME (Bankok) and ICML (Seoul) These top AI conferences all happen in these two weeks… We really need more AI Agents to help the physical world matters :p
1
3
1
23
2,989
Zehua Cheng retweeted
FLock is headed to Geneva next week for a series of events convened by the @UN from 6-10 July. 🇨🇭🇺🇳 Our Founder and CEO, @0x7SUN, has been invited to attend the Global Dialogue on Artificial Intelligence Governance, established by the UN General Assembly under the Global Digital Compact. He will also be joined by our Head of EMEA, @TiffanyWang98, at the @AIforGood Global Summit and the WSIS Forum organised by @ITU. As sovereign and responsible AI move up the global agenda, we're looking forward to sharing lessons from our work on decentralised and privacy-preserving AI, from our sovereign AI collaboration with the Sarawak government to SDG-aligned initiatives with @UNDP. See you in Geneva.
8
13
3
49
6,217
Zehua Cheng retweeted
Next stop for CEO Forum Blockchain and AI for Good - Young Talent Edition: @UniofOxford | March 28 ​You will hear from: ​ 🎙️ @0x7SUN, founder & CEO of FLock, and University of Oxford alumnus. 🎙️ @helen_abfinance, founder of @abf_finance and @ChainforGood, former co-CEO of @Bybit_Official Join us for builder insights, explore what’s pushing the edge of blockchain and AI, and connect with the FLock & BGA teams. In collaboration with Oxford CSSA. If you're in town this Saturday, come through👇
76
52
1
79
3,649
Zehua Cheng retweeted
It's official: CEO Forum - Blockchain and AI for Good: Young Talent Edition is coming to @imperialcollege | March 25 Insider talks from founders and professors who are building and working at the intersection of blockchain and AI. 📢 Speakers: @0x7SUN, founder & CEO of FLock @helen_abfinance, founder of @abf_finance and @ChainforGood, former co-CEO of @Bybit_Official Ying-Ying Hsieh, assistant professor of Innovation and Entrepreneurship; Associate Centre Director, Imperial College Centre for Cryptocurrency Research and Engineering, @imperialcollege. If you're a student, builder, or just deeply curious about where purpose-driven tech is heading, this is your invite 🇬🇧
85
53
1
78
4,255
Zehua Cheng retweeted
🔥
Replying to @flock_io
7/ Read our paper "A Control Theoretic Approach to Decentralized AI Economy Stabilization via Dynamic Buyback-and-Burn Mechanisms." Co-authored by @zehuac, Wei Dai, Zhipeng Wang, @rui_sunn, @yyyzzgc, and @0x7SUN from University of Oxford, University of Manchester, and Newcastle University. arxiv.org/pdf/2601.09961
1
3
15
1,060
Zehua Cheng retweeted
January 15, 2026 is the FLock Day. Merry FLock!
Save the date - January 15. We’re unveiling the next evolution of FLock.
1
1
4
382