Suvash Sedhain

Core Ranking/Retrieval Lead @Tubi. ML Ph.D. from @anucecs. ex @twitter @adobe & @kobo.
Blog:

[Suvash Sedhain @suvsh]

RL/ML papers love equations before intuition. This post attempts to flip it: each idea appears only when the previous approach breaks, and every concept shows up exactly when it’s needed to fix what just broke. Reinforcement Learning for LLMs ("made easy") https://t.co/VMONTwPeXE

[Techikansh @techikansh]

Prompt: "Render a Black Hole"

Model: Claude Fable 5 (high)

** Anthropic cooked so hard, holy Fck **

[Claude @claudeai]

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

[Suvash Sedhain @suvsh]

@_arohan_ LLM is a vast map of all human knowledge. The prompt/prior is the compass that points it in the right direction.

[Suvash Sedhain @suvsh]

This you https://t.co/OnNMjFU4oP

[Dan Held @danheld]

I got offered to be the CEO of 3 Bitcoin treasury companies.

Declined all of them.

This outcome was largely foreseeable.

[Suvash Sedhain @suvsh]

Loved the name Rosalind. I’m guessing it’s after Rosalind Franklin, who arguably deserved more recognition from the Nobel committee.

[OpenAI @OpenAI]

We’re bringing new capabilities to GPT-Rosalind, a model series purpose-built for life sciences research at enterprise scale.

It brings GPT-5.5’s agentic coding and tool use together with stronger intelligence for drug discovery, analysis, design, and experimental workflows.

https://t.co/SrAJ3Mt7ka

[Suvash Sedhain @suvsh]

Nice but wen Qwen3.7 30bA3b or 27b?

[Qwen @Alibaba_Qwen]

👏👏 Introducing Qwen3.7-Plus — a multimodal agent model that unifies vision and language into one versatile agent foundation.

✅ Multimodal interactive hybrid agent: unified GUI & CLI operation across visual and text tasks
✅ Versatile coding agent & productivity assistant with full-modality input
✅ Visual Agent: perception, reasoning, grounding, and search-augmented QA
✅ Cross-harness generalization across diverse agent frameworks

One model. Sees, thinks, codes, acts.🙌🙌

Now available via API on Alibaba Cloud Model Studio. Try it — let us know what you build.😎

🔗🔗⬇️⬇️
Blog：https://t.co/pVYf0h3NNa
Qwen Studio：https://t.co/HUYgFW4cYf
API：https://t.co/viL0cXrMzW

[Suvash Sedhain @suvsh]

@_arohan_ @DougDoesOps @DougDoesOps can I get invite to see what’s going on :)

[Suvash Sedhain @suvsh]

Wow!!! Added to my to-read list.

[Pavlo Molchanov @PavloMolchanov]

🚀 Self-speculation brings 6.75x real speedup for LLM generation with SGLang inference!

Same model drafts future tokens in Diffusion mode → then verifies them in AR (causal) mode. One model and one KV cache. Just different attention masks.

Thanks to perfect alignment, we get 2× longer acceptance lengths than MTP techniques (Eagle-3, MTP, dFlash).

We run 2 forward passes… but the 2× higher acceptance means we break even - and with zero overhead from extra drafter, KV cache, or LM head that comes with MTP - those are not free.

Last week we released Nemotron-Labs-Diffusion + Tri-mode LLMs! We did continued pre-training on Ministral-3 models by switching attention patterns (block causal <> bidirectional). Result: one model that runs AR mode, Diffusion mode, and Self-Speculation.

Diffusion mode already shows high benchmark accuracy - excited to see what happens when someone beats left-to-right acceptance! 🔥

Github: https://t.co/Zqbw3KcAyF
Paper: https://t.co/rp86A7D0xJ
SGLang inference: https://t.co/uTgZPALEJl
Try the models on HF: https://t.co/1zStcCCWPi

[StepFun @StepFun_ai]

⚡️ Step 3.7 Flash is here: The new frontier is agent efficiency.

#1 ClawEval-1.1 (67.1), #1 SimpleVQA Search (79.2), #2 SWE-PRO (56.3), 95.3 on V* Python. Open weights under Apache 2.0.

Built for agentic, coding, search, and multimodal workflows — balancing speed, cost, and reliable execution.

- 400 TPS. 198B sparse MoE, ~11B active. 256K context, 3 reasoning levels.
- Understands UIs, charts, docs, images — then writes code or calls tools to act on what it sees.
- Web + visual search reaches further: more sources, deeper follow-up.
- Reliable tool use — less drift, fewer broken toolcalls. 98%+ on τ²-bench across all difficulty levels.
- Works with Claude Code, KiloCode, Hermes Agent, OpenClaw, and protocols like MCP.
- Runs locally on Mac Studio M4 Max, DGX Spark, AMD AI Max+ 395.

GitHub: https://t.co/kqlZkVIRHv
HuggingFace: https://t.co/qqceCrgPiw
GGUF: https://t.co/rR6XrnymWG
ModelScope: https://t.co/wney6Tzvqy
API: https://t.co/RvHWzRG7Fu
Blog: https://t.co/BxDiajiQ5G

[Suvash Sedhain @suvsh]

100% agree. AlphaGo raised the ceiling for Go; LLMs are lowering the floor for learning (math/engineering/science etc..). Curiosity is the new bottleneck along with, of course, the cost of tokens.

[Noam Brown @polynoamial]

After AlphaGo, the skill of human Go players noticeably improved. I suspect we will see a similar pattern in math.

[Suvash Sedhain @suvsh]

Here we go again!!

[Claude @claudeai]

Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors.

Available today at the same price.

[Liquid AI @liquidai]

Today, we're releasing LFM2.5-8B-A1B, a device-optimized model designed to power real-life applications on phones, laptops, PCs, robots, and fast & lightweight server-side use-cases.

> 8B MoE, 1.5B active
> Expanded 128K context
> LFM2.5 flagship hybrid MoE architecture
> Trained on 38T tokens + large-scale RL
> fast, reliable tool calling, punching above its weight, comparable to models with up to 4x its size
> customizable on a single GPU for any specialized task
> LFM2 open-weight license

🧵

[Suvash Sedhain @suvsh]

8/ The bigger lesson for me:

The value of recommendation foundation models is not just “bigger model = better recommender.”

It is that heterogeneous signals — watches, searches, sessions, surfaces, containers, time — can live in one shared sequence model and reinforce each other.

[Suvash Sedhain @suvsh]

How Tubi is using foundation models to power recommendations at scale: https://t.co/X6fdNjbc96.
This is one of the projects that got me fully LLM-pilled. @MikeTamir

[Suvash Sedhain @suvsh]

7/ The search result is especially interesting.

Search queries are often short, partial, ambiguous, or title-seeking. By combining query tokens with the viewer’s prior journey, TubiFM can use behavioral context when lexical evidence is sparse. That is where foundation-model-style recommenders start to feel very natural.