Deep Learning Weekly: Issue 466
Claude Opus 5, How Digibee Builds Prompts with Opik to Power Their AI-Native Integration Platform, a paper on Progress Reward Modeling for Robotic Learning, and many more!
This week in deep learning, we bring you Claude Opus 5, How Digibee Builds Prompts with Opik to Power Their AI-Native Integration Platform and a paper on Progress Reward Modeling for Robotic Learning: A Comprehensive Survey.
You may also enjoy FLUX 3, Evaluating Agents Beyond the First Prompt, a paper on StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents, and more!
As always, happy reading and hacking. If you have something you think should be in next week’s issue, find us on Twitter: @dl_weekly.
Until next week!
Industry
Introducing Claude Opus 5 \ Anthropic
Anthropic releases Claude Opus 5, taking state-of-the-art on Frontier-Bench and GDPval-AA and surpassing Fable 5 on several evals at half the cost, with its lowest misalignment score yet.
FLUX 3 - Real World Models: Towards Multimodal Flow Models as the Backbone of Visual Intelligence.
Black Forest Labs launches FLUX 3 in early access, a unified multimodal model trained jointly on image, video, and audio that generates 20-second clips with native audio and extends to robotic action prediction.
Fish Audio raises $52M seed to build AI voice models for creators and enterprises
Fish Audio raised a $50M seed led by Coreline Ventures and Capital Today, reaching $21M ARR and 8 million users a year after launching its open-source voice models.
OpenAI launches Health in ChatGPT for U.S. users, letting connected Apple Health and medical records ground conversations anywhere in the app, with that data excluded from model training and ad targeting.
Microsoft launches new in-house AI models it says cut costs up to 89% versus OpenAI | VentureBeat
Microsoft releases MAI-Image-2.5-Pro and MAI-Voice-2-Flash into public preview, displacing OpenAI models across Bing, PowerPoint, and Dynamics 365 with claimed GPU cost cuts of 84% and 89%.
MLOps/LLMOps/AgentOps
How Digibee Builds Prompts with Opik to Power Their AI-Native Integration Platform
A case study on how Digibee adopted Opik to bring evaluation-driven development into production, using traces and automated evaluations to catch regressions before they reach users.
Build your own Job Agent - Part 1
Part 1 of 3 of The Observable Job Agent series: A hands-on guide to building a LangGraph-powered job search agent that ranks real openings while instrumenting every step with Opik observability.
Learning
Teaching LLMs to Update Beliefs for Efficient Long-Horizon Interaction
A research post introducing ABBEL, which replaces recursive summarization with graded natural-language belief states, closing about half the gap to full-context agents on CollabBench in 50% fewer training steps.
Evaluating Agents Beyond the First Prompt
A breakdown of EvoCode-Bench, a 227-round multi-turn coding benchmark where pass rates fall from 46.7% at Round 1 to 7.7% by Round 10, with regressions rather than missing features driving failure.
Autonomous AI Intrusions Are Here: Lessons from the Hugging Face Compromise
A defender-focused analysis of the Hugging Face breach, in which an autonomous agent logged over 17,000 attack actions and commercial model guardrails blocked forensic work until responders switched to a self-hosted open-weight model.
Libraries & Code
An open-source AI observability tool used to debug, evaluate, and monitor LLM applications, RAG systems, and agentic workflows with comprehensive tracing, automated evaluations, and production-ready dashboards.
An agentic-first RL framework for research.
Papers & Publications
StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents
Abstract:
Computer-use agents are usually improved by strengthening perception: better models for reading a screenshot and choosing where to click. Yet a screenshot is only a lossy rendering of the underlying program state, e.g., the files, application backends, and DOM that hold the task data. Different states can produce the same pixels, while code can inspect and modify that state directly. StateAct is a code-first, multi-agent harness built around this distinction. Its main agent works directly with program state by using code, while a dedicated GUI subagent handles screenshot-and-click interaction on the few subgoals that need it, just 28 of 108 tasks and 1.1% of main-agent steps. The same direct access to program state also supports verification: an independent finish gate double-checks the saved result for structural failures, e.g., output that is missing, unsaved, or written to the wrong path. To stay on track over hundreds of steps, the main agent hands subgoals to fresh subagents, keeping its own context focused. On OSWorld 2.0, StateAct lifts Claude Opus 4.8 from 20.6% to 26.9% on binary success, and from 54.8% to 61.6% on partial success, at ~ 9x lower cost per task than the same model driven by screenshots alone; a code-only variant with no GUI subagent reaches only 45.9% partial, below that screenshot-based baseline’s 54.8%. In general, grounding action, verification, and memory in state, what we call state-grounding, shifts the main bottleneck from perception toward reasoning: failures depend more on what the agent thinks than on what it sees.
Progress Reward Modeling for Robotic Learning: A Comprehensive Survey
Abstract:
Robotic learning takes place in dynamic environments with large behavior spaces. A terminal success signal only tells the robot whether the task is completed. It does not explain whether the current behavior is making progress, remaining unchanged, or undoing earlier progress. For this reason, recent studies have increasingly explored progress rewards that provide feedback during task execution. However, the current literature lacks a shared framework. Existing methods use different observations, goal specifications, output signals, supervision sources, and evaluation protocols. This makes it difficult to compare them and understand what their results actually validate. In this survey, we provide a unified view of progress reward modeling for robotic learning. We organize the field in three connected steps. We first study the interface of a progress model. This defines the problem from the outside by asking what information the model receives and what form of progress signal it produces. We then move inside the model and study the methods used to construct this signal. This reveals the different assumptions and mechanisms behind progress estimation and reward generation. Finally, we examine the data and benchmarks that support these methods. This shows how progress supervision is obtained and what different evaluations actually measure. Together, these three perspectives connect what a progress model is, how it is built, and how its quality is validated. We further summarize the main limitations of current approaches and discuss future research directions.
 | A guest post by
|