Pinned Tweet
Agentic post-training needs data that targets the model's failure modes, so you teach it what it doesn't know yet. One failure tells you something, but training needs many feasible, verifiable tasks that exercise the same capability in different situations.
We built AutoSynthData to turn those capability gaps into training data. It uses a target model's failures and a stronger teacher's successes to decide what the model should learn next, then generates and validates new tasks that exercise those capabilities. As the model improves, the curriculum shifts toward what it still finds difficult.
On EnterpriseOps Gym (Hybrid), 2,000 generated tasks raised Gemma 4 26B's Pass@1 by 7.2 points (35% relative).
Read our full blog here: huggingface.co/blog/ServiceN…
w/ Shruthan, @den_run_ai, @sagardavasam
Vignesh Raja retweeted
🚀 Meet Apriel Voice 27B Preview: an open-weight voice AI built to get work done.
Beyond conversation: adapt to dynamic instructions, reason across modalities, and orchestrate multi-step tool calls to complete real tasks.
See Apriel Voice in action. 👇
Vignesh Raja retweeted
I think I found a planet nobody knew existed.
It's been hiding in NASA telescope data for seven years.
At the end of this thread, you can fly to it yourself.
Vignesh Raja retweeted
SFT is not dead! 🥳
We found a way to make SFT rival current prevailing posttraining methods, often generalizing better and forgetting less than RL and OPSD. 🤯
Following our prior work on reasoning with sampling, we now introduce sampling to the posttraining stack.
1/n
Vignesh Raja retweeted
i gave astra a robot, a paint brush, and a camera then asked it to paint the golden gate bridge in real life!
it figured out how to control the robot, and progressively got better throughout its attempts. the timelapse is sick
Vignesh Raja retweeted
StarHarness: Evolve the harness, not the model
ServiceNow's new method searches for the optimal agent scaffold around a fixed model, boosting performance by 20–35 points across enterprise benchmarks while cutting inference cost up to 53%.
Vignesh Raja retweeted
Introducing Atlas:
The world's first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D.
Model the world, move the camera, and simulate space & time.
On ITBench, Qwen3.5-27B with an evolved harness beats GPT-5.5 on the baseline harness by +19.2 points.
More capable models do not always make better enterprise agents.
With the right harness, a smaller open-weight model can outperform a larger frontier model running the default scaffold.
In our new preprint, we present StarHarness, a method for directly searching for an environment-optimal harness around a fixed model.
Across three enterprise benchmarks, the evolved harness improved performance by 20–35 percentage points and reduced inference cost by 17–53%.
Our results suggest that some of the capability gap we attribute to models may actually come from how well the surrounding harness fits the environment.
Read the full preprint here: arxiv.org/abs/2608.24804
w/@den_run_ai, Vikas Yadav, @RajeswarSai, @patricebechard, @siris_krishna, @sagardavasam
Vignesh Raja retweeted
LLM review weirdness...
Just renaming an uploaded pdf from "paper.pdf" to "paper_final_draft_pdf_ready_for_review.pdf" boosts average scores
(gpt-5.6-terra)
Vignesh Raja retweeted
We're open-sourcing StarShell: our bare-bones agent with only a shell + filesystem that beats MCP and browser-use agents on real enterprise tasks — at 10x lower cost.
Paper also just got a big update with new experiments.
📄 arxiv.org/abs/2604.00073
💻 github.com/servicenow/starsh…
🗂️ huggingface.co/collections/S…
🌐 servicenow.github.io/StarShe…
@Vikas_NLP_UA @sagardavasam @RajeswarSai @ServiceNowRSRCH
Vignesh Raja retweeted
In 1969, we landed on the moon and flew Concorde. Half a century later, we could do neither.
Blake Scholl (@bscholl) founded @boomsupersonic (YC W16), the startup building America’s first supersonic airliner, to change that.
At Startup School 2026, he shares how a cardboard mockup with Office Depot seats became XB-1, the first independently developed jet to break the sound barrier, and why founders have to build for both the worst day and the best day.
00:08 — Why the Future Stopped Moving Faster
03:14 — Why I Started Boom Supersonic
05:57 — Building a Supersonic Jet From Scratch
08:33 — The Worst Day and the Best Day
10:14 — How We Changed US Law
11:36 — How 50 People Built a Supersonic Jet
15:05 — Designing Hardware Like Software
17:54 — Financing a Multi-Billion-Dollar Startup
19:30 — Great Ideas Are Hiding in Plain Sight
24:05 — How AI Is Changing Hardware
27:16 — Working With Regulators
30:55 — Build Something You Love
36:27 — How to Build Confidence
39:35 — Learning Hard Things From First Principles
44:17 — When Should You Start a Company?
Vignesh Raja retweeted
Live in the future, then build what's missing.
Excited to kick off day 1 of Startup School 🚀
“Apparently, if you’re here, you’re going to make it. So I’m happy I’m here.”
- Jensen at @ycombinator AI Startup School
Vignesh Raja retweeted
we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this.
openai.com/index/hugging-fac…
Vignesh Raja retweeted
We're partnering with @NVIDIA to make sure every AI agent your enterprise deploys can be seen, measured, and governed—from the desktop to the data center.
Today at Knowledge 2026, we're introducing Project Arc: an autonomous desktop agent that thinks, writes code, and executes complex work across your enterprise tools. Every action it takes runs inside NVIDIA OpenShell's sandboxed environment and is governed by AI Control Tower—so it's fully auditable before it ever reaches production.
This is what enterprise AI should look like: autonomous, but never unsupervised.
Get the details: newsroom.servicenow.com/pres… #Know26
Vignesh Raja retweeted
Super Apriel: one checkpoint, many models ⚡️
You can choose the speed ↔ quality tradeoff at serving time, not at release time.
Huge congrats to the entire team.. no more shipping one architecture and regretting it later. 🙌
@tscholak @ostap__alex @SathwikTejaswi @sagardavasam
