ServiceNow AI Research
@ServiceNowRSRCH
Followers
18K
Following
4K
Media
446
Statuses
2K
Unlock work experiences of the future. Join @ServiceNowRSRCH as we advance the state-of-the-art in Enterprise AI. #ServiceNowResearch #LifeAtNow #Hiring
MontrΓ©al, QuΓ©bec
Joined July 2016
SLAM Labs presents Apriel-1.5-15B-Thinker π An open-weights multimodal reasoning model that hits frontier-level performance with just a fraction of the compute.
15
79
337
ServiceNow AI Research presents PipelineRL β one of the most impactful efficiency tricks in modern RL training. An elegant solution to a noisy, expensive problem. Worth the read π
Don't sleep on PipelineRL -- this is one of the biggest jumps in compute efficiency of RL setups that we found in the ScaleRL paper (also validated by Magistral & others before)! What's the problem PipelineRL solves? In RL for LLMs, we need to send weight updates from trainer to
0
10
21
In-flight weight updates have gone from a βweird trickβ to a must to train LLMs with RL in the last few weeks. If you want to understand the on-policy and throughput benefits hereβs the CoLM talk @DBahdanau and I gave:
1
29
139
𧩠Down the Rabbit Hole of Backdoors in the AI Supply Chain As AI agents become trusted digital coworkers, their security is more critical than ever. In our latest research, @ServiceNowRSRCHβusing @nvidia DGX Cloudβ’ on AWSβuncovered how data poisoning can secretly implant
servicenow.com
Find out how the very process used to make AI agents smarter can also make them vulnerable, based on research from ServiceNow and NVIDIA.
1
1
6
Had a great time talking about how to train LLM agents at @thealliance_ai, awesome questions from the crowd! π·:
π Our paper βπ»ππ€ π‘π πππππ πππ’π πΏπΏπ πππ π΄ππππ‘: π΄ ππ‘ππ‘ππ π‘ππππ π·ππππππ ππ β got an π¨π«ππ₯ at next weekβs πππ π πͺπΌπΏπΈππ΅πΌπ½ πΌπ» ππΌπΊπ½πππ²πΏ π¨ππ² ππ΄π²π»ππ! π₯οΈπ§ We present the ππ’π«π¬π π₯ππ«π π-π¬πππ₯π
0
6
14
π Internship Opportunities in AI Agents! At @ServiceNowRSRCH, we have 4 internship positions on AI Agents β exploring robustness, privacy & collaboration. π§ Applicants must be registered students at a Canadian university. π Thread with details & links to apply:
2
14
82
What a night! π§ Last week, @ServiceNowRSRCH & @LawZero hosted a special social event on AI safety & security alongside the #CoLM2025 conference in Montreal. Key takeaways: β’ System-level design matters β’ Real-world > benchmark risks β’ Balancing short- & long-term safety
0
1
2
π Releasing DRBench, an Enterprise-Grade Deep Research Benchmark Paper! π Paper: https://t.co/meWM3Qj77w π» Code: https://t.co/RfhQ1mWayc Weβre excited to introduce DRBench, the first benchmark designed to evaluate deep research agents on open-ended enterprise research tasks,
0
19
41
Very excited to be presenting Pipeline RL this afternoon at CoLM. Join us if you are interested in fast on policy RL training for LLMs π
We did lots of good work since PipelineRL release in May: βοΈ higher throughput, seq parallel training, multimodal, agentic RL π white paper with great explanations and results: https://t.co/F3YsIbNRUy We'll present today at CoLM EXPO, room 524C, 1pm!
0
8
20
π’ Donβt miss it tomorrow at #COLM2025! Our researcher Arjun Ashok will present 𧩠βContext is Key: A Benchmark for Forecasting with Essential Textual Informationββ³ π 4:45 PM at the XTempLLMs 2025 workshop π https://t.co/P63UMb0REr
#AIresearch #ServiceNowResearch #XTempLLMs
0
1
7
π’ Excited to share our COLM 2025 paper: BigCharts-R1 π We introduce a new dataset and training framework that push the boundaries of chart reasoning in VLMs. ποΈ Join us at Poster Session 4 on Wednesday, 4:30 PM! 𧡠Thread below π https://t.co/aJLDFZYRom
arxiv.org
Charts are essential to data analysis, transforming raw data into clear visual representations that support human decision-making. Although current vision-language models (VLMs) have made...
3
8
20
β¨ What if we could tune Frontier LLM agents without touching any weights? Meet JEF-Hinter, an agent capable of analyzing multiple offline trajectories to extract auditable and timely hintsπ‘ In our new paper π , we show significant performance gains on downstream tasks β‘ with
1
14
24
π§ Donβt miss it at #COLM2025! SOCIAL: βReasoning LLMs β Tips & Tricks Discussionβ π
Tue, Oct 7 | π 1:00β2:30 PM | π Room 517BC Co-presented by ServiceNow AI Research + NVIDIA with Torsten Scholak (ServiceNow) & Olivier Delalleau (NVIDIA). Join the discussion on advancing
1
1
7
π Itβs CoLM week! The Conference on Language Modeling (CoLM 2025) kicks off tomorrow in MontrΓ©al π¨π¦π Proud that ServiceNow AI Research is a main sponsor β and that our team will present 5 papers on: π Multimodal reasoning π Unified AR & diffusion models π Dense retrieval
0
4
8
Very excited to see vLLM supports Pipeline RLβs in-flight weight updates! It allowed our team to quickly and reliably train Qwen base 7B to reason from scratch! Want to hear more? Join us at our Pipeline RL expo talk at CoLM this Thursday 1PM room 524C.
π The RL community keeps pushing boundaries β from better on-policy data and partial rollouts to in-flight weight updates that mix KV caches across models during inference. Continuing inference while weights change and KV states stay stale sounds wild β but thatβs exactly what
1
10
26