Johnson Lai | Head of Data & AI @Chromia & Ambassador @Alibaba_Qwen | Previously @CoinGecko

Malaysia
Joined January 2011
We've started preparing and cleaning scrapped 7.8k+ IMD oracle requests from the IMD API huggingface.co/datasets/eval…
EVAL Engine has acquired IMD NFT and running a node as part of our experiment into the AI Swarm ecosystem We’re actively exploring how EVAL can contribute to IMD base:0xdd78523217390bb0d49c7601e7e54c36d71622f0
7
30
1,313
EVAL Engine has acquired IMD NFT and running a node as part of our experiment into the AI Swarm ecosystem We’re actively exploring how EVAL can contribute to IMD base:0xdd78523217390bb0d49c7601e7e54c36d71622f0
2
9
2
31
5,833
Decision-4B is live A Jev-like decision model. On your phone. In our Unbound app. Right now No API. No GPU. No account
6
10
1
63
7,337
been testing a few Jev models by making them play Pokémon & snake lol which is the best way I've found to see how a model handles a complex situation tested: Laya, @jaredpalmer's Kev, @flock_io This/That 1.1, @googlegemma djev
1
2
1
4
698
2/ but for snake game, @flock_io This/That 1.1 scores 22.7 and becomes the first model to beat the greedy heuristic
1
1
79
TLDR; - Laya: fastest by 3× (6 ms/decision) - @jaredpalmer's Kev: solid on both, only one to reach Viridian every run - @flock_io This/That 1.1: first to beat the greedy heuristic at snake - @googlegemma djev: decisive, slow, gets stuck in Pallet Town Honestly, I spent way more time building the harness than the models spent "playing" 😂
1
74
chromaway:native is waking up. 🔥 AI agents are the next big wave, but autonomy without security is a ticking time bomb. The future of AI needs: 🤖 Onchain memory for AI agents 🔐 Privacy & secure data management ⛓️ Verifiable actions & transparent execution ⚡ Scalable infrastructure built for AI This is exactly what we are building. The AI agent economy needs more than just intelligence. It needs trust. $CHR 🚀
chromaway:native 🚀 • AI + Data infrastructure • On-chain relational database with native AI capabilities • Growing AI agent ecosystem • Decenchro, Unbound & Chromia Pay🔥 $CHR is one to watch.
28
3
109
13,391
Took us a while, but we finally released Unbound AI on the official Apple App Store! 🎉 The model was originally developed by @Chromia & @Eval_Engine and has already surpassed 10,000 downloads on Hugging Face in just one week
Unbound AI is finally live on iOS App Store! A model with no guardrails now lives on the hands of every iPhone users No API, subscription or servers. Runs 100% on-device Try here - apps.apple.com/us/app/unboun…
7
6
1
40
4,322
Few days left to own your private and secure AI Agents decenchro now hosts AI Agents that can perform research, analysis and answer general web3 questions - all at no costs More agents coming soon 🔜 decenchro.com Powered by @ATBASHai
50+ agents have been deployed on decenchro.com Private by default, with a record of everything they do Early, but it's growing in the ecosystem Secure your agent - decenchro.com @ATBASHai chromaway:native
3
13
38
4,412
With @inco_ai DFlash2 is even faster lol, 40% increase @ 138.98t/s inco.ai/blog/dflash2/
Wow @Alibaba_Qwen's 3.8-27b model beats Opus 4.6 Max on several benchmarks & you can now run it at home I'm getting 100+ tokens/s: very usable intelligence, running locally🤯
1
9
935
Wow @Alibaba_Qwen's 3.8-27b model beats Opus 4.6 Max on several benchmarks & you can now run it at home I'm getting 100+ tokens/s: very usable intelligence, running locally🤯
2
1
1
13
1,679
Introducing decenchro Your private and secure AI agents, deployed in one click. Everyone wants to run one. Almost nobody wants to spend a weekend wiring servers and pasting keys into config files. With decenchro, now you can.
10
22
4
92
43,179
Tested EasyCache with MiniMax H3 today. ⚡ 118.48s → 82.08s 📈 ~31% faster (1.44× speedup) I couldn’t notice any quality difference in this test, but I’d probably stick to a 0.20 reuse threshold for a safer balance between speed and output quality
Will Smith x Spaghetti remix 🤪 MiniMax-H3 on RTX 3060 6GB & 64GB RAM (A 5s 480p T2V clip took around 4–5 mins using EasyCache + SageAttention)
3
6
922
Comparing MiniMax-H3 with Wan, LTX, and Seedance. All self-hosted models were benchmarked on an NVIDIA RTX PRO 6000 Blackwell using ComfyUI All videos are 5s long * times in second shown are generation times
1
4
1,266
Prompt: One continuous cinematic shot in a premium glossy handcrafted miniature world. The camera makes a slow, steady forward push along the glowing cyan route toward the warm golden signal. The faithful penguin wizard remains the same character with the same face, navy hat, amber stars, colors, and proportions. The penguin breathes subtly; the hat tip, tiny flags, and soft atmospheric particles move gently. Strong layered foreground parallax, smooth controlled motion, and stable geometry. Preserve the exact source composition and character identity. No cuts, no crossfade, no morphing, no new characters, no text.
1
1
121
TLDR; based on my experience so far, MiniMax-H3 delivers very good quality for the generation time I can also increase the resolution when higher quality output is needed on ComfyUI
1
206