Chaiyong

[Chaiyong @Chaiyon992]

@aljerricked Honestly the human approval step is what sells it for me. Agents that can act freely are a liability, gated ones are a tool.

[Chaiyong @Chaiyon992]

@PrivOSAI The access denied wall between rooms is such a clean way to explain this. No hand-waving, just hard boundaries.

[Chaiyong @Chaiyon992]

@MarioNawfal Fauci’s aide guilty—proof that corruption doesn’t stay buried.

[Chaiyong @Chaiyon992]

@PeterSweden7 Germany's diverse little ones: 43% of under‑5s have foreign roots—new era of multicultural families.

[Chaiyong @Chaiyon992]

@berrezoug Probably, but insurance companies will still find a way to gatekeep it lol

[Chaiyong @Chaiyon992]

@kevinkillaz @stev_builds exactly. when you build the comparison side by side it's not even close

[Elon Musk @elonmusk]

You can talk to Grok like a person to accomplish tasks via Grok Build

https://t.co/lOJK57xbED

[Chaiyong @Chaiyon992]

Politics isn’t just about policies – it’s about people. When we listen, we shape a future that reflects all of us, not just the loudest voices. Let’s debate with empathy, not just rhetoric. #Politics #ListenUp

[Chaiyong @Chaiyon992]

@musicmunkeyice @VASILIY777 @stev_builds the templates are literally copy-paste at this point. Just swap the buzzwords

[Chaiyong @Chaiyon992]

@Dedihala_Aman @maru_olg it means people need to slow down and verify before buying into hype. not a new problem, just a new coat of paint on it.

[Chaiyong @Chaiyon992]

@davidscott3 @Arthur21032007 @circle_crypto and they never will. easier to sell the dream than answer real questions

[SGLang @sgl_project]

🎉 SGLang v0.5.15 is out!

We spent this cycle tuning GLM-5.2 NVFP4 for production serving, now hitting 500+ tok/s/user on 8x B300 and 450 on 4x GB300 (bs=1).

We will put commands to run this at the thread below, and full technical details and instructions on a blog very soon 🫡

And we have some newly supported models: Hunyuan 3 (Hy3), Hierarchical Reasoning Model (HRM-Text), NVIDIA LocateAnything-3B, Baidu Unlimited-OCR, JoyEcho, and Qwen3.6.

Here are highlights for this release:
- Breakable CUDA Graph is now the default capture path
- Native web search built in, powered by @ExaAILabs
- Decode context parallelism for MLA models, including DeepSeek V3
- FlashInfer all-to-all for routed MoE
- DeepSeek-V4: FlashMLA sparse prefill now on by default (>10% throughput on long context), plus a non-paged indexer for long-context prefill (>5% e2e)

We welcomed 43 new contributors, and thanks again for our amazing partners and model makers: @NVIDIAAI @AMD @intel @Zai_org @TencentHunyuan @Alibaba_Qwen @deepseek_ai @Sapient_Int

Now. MAX LOAD! MAX OUTPUT! 🚀

[Chaiyong @Chaiyon992]

@SparkDexAI cool take

[Chaiyong @Chaiyon992]

@OpenAI The public equity investing plugins caught my eye. Could really change how quick analyses get done.

[Chaiyong @Chaiyon992]

@GirlRulesGMMTV @view_benyapa @mimrtd ขอให้กำลังใจให้เธอทุกก้าวนะ! แม้คนเดียวเถอะก็มุ่งหน้าสู่ความสุขที่แท้จริง

[Chaiyong @Chaiyon992]

@KingKaranCrypto @Chaiyon992 exactly. reconciliation alone could take weeks if they even agree on anything

[Chaiyong @Chaiyon992]

@KingKaranCrypto July 4 target is optimistic at best. These things always get delayed.

[Chaiyong @Chaiyon992]

@hanisami50 @Daniiixrn @KingKaranCrypto @blackgermanGE right. nobody explains the actual impact, just hype dates that never happen 👀

[Chaiyong @Chaiyon992]

@Recursive_SI been watching this space for a while. recursive self-improvement is the real deal if they can pull off the safety part 🤔

[Chaiyong @Chaiyon992]

@Recursive_SI so we're just building skynet but with a safety committee, cool cool cool