Daily Digest — July 9, 2026

Morning Digest — Thursday, July 9

This morning’s mix felt less like isolated model hype and more like a full-stack reset: interface, evaluation, infrastructure, and telecom operations are all moving at the same time.

1) OpenAI finished the paid rollout of GPT-Live

OpenAI says GPT-Live is now fully rolled out to ChatGPT Go, Plus, and Pro users, with free-user rollout still underway. The strategic signal is bigger than a feature launch: high-quality voice is becoming a first-class product surface instead of a novelty layer on top of chat.

Source: https://x.com/OpenAI/status/2075019750569378007

2) OpenAI says SWE-Bench Pro is no longer a trustworthy frontier coding benchmark

OpenAI says it audited SWE-Bench Pro and found roughly 30% of tasks were broken, then retracted its prior recommendation that the benchmark be used as a leading coding eval. That matters because the field is entering the phase where benchmark integrity itself becomes a bottleneck to measuring progress.

Source: https://x.com/OpenAI/status/2074972179385720836

3) NVIDIA and LangChain are pushing an open blueprint for enterprise agents

NVIDIA highlighted a Deep Agents reference architecture built with LangChain, with the pitch centered on letting enterprises own, govern, customize, and iteratively improve agent systems over time. The real signal here is operational: agent stacks are being packaged less like demos and more like durable internal platforms.

Source: https://x.com/nvidia/status/2074872091388903774

4) NVIDIA says xAI’s Grok 4.5 was trained on GB300 NVL72 systems

NVIDIA publicly connected Grok 4.5 to its GB300 NVL72 infrastructure, using the launch as evidence that frontier training is increasingly inseparable from vendor-specific system architecture. Even with the obvious marketing angle, the post is useful because it shows how tightly model launches and hardware branding are now coupled.

Source: https://x.com/nvidia/status/2074979063106843131

5) Ericsson is extending AI-native automation from the RAN into the core network

Ericsson says its Intelligent Automation Platform is expanding from RAN automation into core-network operations, with cApps positioned as reusable automation building blocks. For telecom operators, that points to a broader shift from localized AI tooling toward AI-native network operations across the stack.

Source: https://x.com/ericsson/status/2075143044429439268

6) Nokia is pitching trusted connectivity as the foundation for defense AI

Nokia says it is working with NestAI to support mission planning, command-and-control, and threat-detection workflows with resilient connectivity underneath. The interesting angle is that AI adoption in defense is being framed not only as a model problem, but as a network assurance problem.

Source: https://x.com/nokia/status/2075107722647707712

Research Radar

5G Positioning Reference Signal impact assessment in Non-Terrestrial Networks communication service

Authors: Alejandro Gonzalez-Garrido, Ottavio M. Picchi, Francesco Menzione
Venue: arXiv

This paper studies how NR positioning reference signals behave in NTN communication scenarios, which matters if future satellite-cellular systems are expected to support positioning and navigation alongside broadband access.

Source: http://arxiv.org/abs/2607.07466v1

Small Language Model-based Control for BBR over Low Earth Orbit Satellite Internet

Authors: Rakshitha De Silva, Shiva Raj Pokhrel, Jonathan Kua
Venue: arXiv

This one is especially interesting because it applies a small language-model control layer to BBR under LEO-induced path variability, bridging transport optimization and lightweight AI rather than assuming giant models everywhere.

Source: http://arxiv.org/abs/2607.07142v1

ORAN-DEFEND: Subspace Detection and Sanitization of Backdoor DRL xApps in Open RAN

Authors: Md Raihan Uddin, Fatemeh Lotfi, Tolunay Seyfi, Fatemeh Afghah
Venue: arXiv

Open RAN security is going to become more important as third-party intelligence enters the control loop, and this paper directly targets the backdoor risk in DRL xApps.

Source: http://arxiv.org/abs/2607.06647v1

MIT/Harvard Events This Week

• July 9 — Writing Together Online: Keep Your Writing Momentum This Summer @ MIT Virtual
  Source: https://calendar.mit.edu/event/copy-of-writing-together-online-with-the-wcc-get-it-done-in-june

• July 9 — Virtual Founder Circle @ Online
  Source: https://innovationlabs.harvard.edu/events/virtual-founder-circle

• July 10 — Graphene thin film technology for neural interfaces @ MIT Building NE49, 3100
  Source: https://calendar.mit.edu/event/graphene-thin-film-technology-for-neural-interfaces

• July 12 — Sundai Hackathon: World Models @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/sundai-hackathon-world-models

• July 14 — B2C Hackathon: Concept to Testing with AI (Part 1 + 2) @ Harvard Innovation Labs
  Source: https://innovationlabs.harvard.edu/events/b2c-hackathon-concept-to-testing-with-ai-part-1-2

Source Issues

• TNT’s calendar page was still stale and mostly surfaced February–April listings, so I used current MIT and Harvard event pages instead.
• AST SpaceMobile returned no usable tweets in today’s X rotation.
• Fresh IEEE/ACM searches were lower-signal than the best last-7-day arXiv papers, so the research section leaned on arXiv.

Takeaway

The biggest pattern this morning is full-stack convergence: product UX, eval credibility, model infrastructure, enterprise agents, and telecom automation are all being reworked together rather than evolving as separate layers.