Viven (@okaybroda) header
Viven (@okaybroda) profile picture

Software Engineer. Machine Learning enthusiast. LFC fan. 🇲🇾

Toronto, Ontario
Joined August 2009
Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)Photo by Viven (@okaybroda)
Gemini 3.8 Flash is still the agentic model of choice. Right balance of quality and cost
60
We're giving away a CONTROL Resonant custom wrapped NVIDIA GeForce RTX 5080 to celebrate its official release with #RTXON. To enter: 🟢 Share this post 🟢 Comment #RTXON T&Cs: nvda.ws/4z24Gcq
37,029
30,542
2,813
33,141
3,114,181
Took ideas from this PR and made a rudimentary Jev-like app using DiffGemma. It gets its probabilities from a one-pass read of yes/no logits. DiffGemma is multimodal, so image input is supported and tested.
I ran some real, live evals on Jev vs DiffusionGemma-as-Jev (my patch for vLLM!) DiffusionGemma comes out as the winner, I think. Headlines: Is Jev faster than DiffusionGemma? No ❌ (API vs DGX Spark) Is Jev smarter than DiffusionGemma? No ❌ (they're roughly tied!)
1
119
Next I’m testing it with a Subway Surfers–style demo, like the Jev demos people have been building, to see whether it’s actually usable for visual control.
1
29
Chinese researchers did it again! Moonshot AI proved that 95% of attention computation is completely wasted.. They’ve open-sourced the exact attention mechanism powering Kimi's long context.. and it's 16x faster. It’s called MoBA (Mixture of Block Attention). Every time you feed an AI a massive prompt, every token desperately tries to look at every single token that came before it. The math scales quadratically ($O(N^2)$). It’s like reading a 500-page book and re-reading the entire thing from word one every single time you want to understand a new sentence. Moonshot looked at this waste and said: No. Instead of forcing full, brute-force attention across everything, MoBA applies the principles of Mixture of Experts (MoE) directly to the attention layer. How it works: • It chops the context history into discrete blocks. • A lightweight dynamic gating mechanism lets each query token instantly "vote" and route only to the specific blocks that actually matter. • It achieves up to 95% sparsity on long contexts without sacrificing performance. The results are staggering: It maintains full-attention accuracy on 1-million-token contexts while slashing massive compute overhead. We are watching the structural efficiency barrier of long-context AI get smashed in real-time. While Western labs argue about how to afford million-token context windows, Moonshot open-sourced the blueprint to make them radically cheap.
52
342
33
2,361
106,866
Might be a hot take but I think software engineering hasn't changed much. Yes, LLMs are non-deterministic but we have been building for non-deterministic users for decades - humans! We have built QA processes around users with safeguards, guardrails to prevent unwanted outcomes.
29
Viven retweeted
Holy moly. 🤯 Qwen 3.8 27B just scored higher than: GPT 5.6 Terra GLM 5.2 DeepSeek V4 Pro Muse Spark 1.2 Claude Opus 4.8 on the Artificial Analysis Agentic Index. And you can run this on a single RTX 3090/4090. Go show your GPU some respect. 🫡
124
201
46
2,804
396,582
Viven retweeted
Fast H3 implementation for Metal. Enjoy, modify, and so forth: github.com/antirez/h3.c Contains code from @liuliu which is welcomed in taking back whatever parts he likes for @drawthingsapp in case there are H3 plans there.
51
175
44
2,164
565,328
Viven retweeted
Voice mode now runs on Claude's more capable models and reaches the tools you've connected mid-conversation. Talk through the hard problems out loud, in many more languages.
709
777
433
11,519
1,772,643
Viven retweeted
Attack is the best form of defense. ♟️
934
18,495
1,265
135,572
3,134,126
Viven retweeted
We love you, @MoSalah. ❤️
307
12,008
564
124,466
1,359,630
Jürgen Klopp led a furious backlash against Fifa president Gianni Infantino, who now faces calls to resign after overturning Folarin Balogun’s World Cup suspension. “This is our sport, not theirs. If Donald Trump and Gianni Infantino really sorted this out between themselves, it is madness; it calls everything into question.” @ben_rumsby, @TelegraphDucker and @Tom_Morgs have more ⤵️ telegraph.co.uk/football/202…
482
4,321
341
30,272
1,556,666
Most text embedders match a query to a doc in a single forward pass. But reasoning-heavy retrieval takes a few hops of reasoning. So I tried a recurrent architecture instead.
1
1
78
Where it hits a wall: knowledge. The base was pretrained on a deliberately thin slice of text (HRM-Text is efficiency-first), so it's weak on knowledge-heavy domains. The reasoning engine's there, the base is the bottleneck.
1
29
Nothing but pride for @CANMNT_Official today.   An incredible run and a sign of what’s to come.   Thank you to our team, the fans, and everyone who made this World Cup showing our best yet.
249
900
27
10,281
147,495
Our World Cup journey comes to a close. Canada’s football journey is only just beginning. #CANMNT
742
886
351
6,447
399,481