SIGNAL

Wednesday, February 18, 2026

21 stories · 6 min read

Metas new deal with Nvidia buys up millions of AI chips

Meta has struck a multiyear deal to expand its data centers with millions of Nvidia's Grace and Vera CPUs and Blackwell and Rubin GPUs. While Meta has long been using Nvidia's hardware for its AI products, this deal "represents the first large-scale Nvidia Grace-only deployment," which Nvidia says will deliver "significant performance-per-watt improvements in [Meta's] data centers. " The deal also includes plans to add Nvidia's next-generation Vera CPUs to Meta's data centers in 2027.

The Verge AI· Stevie Bonifield· 2 min· 95d agoSource →

01QUICK HITS

01

Show HN: AsteroidOS 2.0 – Nobody asked, we shipped anyway

Hi HN, After roughly 8 years of silently rolling 1. 1 nightlies, we finally tagged a proper stable 2. 0 release.

[pdf] [video]

Thousands of CEOs just admitted AI had no impact on employment or productivity

Hacker News · 1 min

→04

Google announces dates for I/O 2026

It's official: Google I/O 2026 will take place from May 19th to 20th. In an announcement on Tuesday, Google says it will share the "latest AI breakthroughs and updates in products across the company, from Gemini to Android and more" during the event, which will take place in-person in Mountain View, California's Shoreline Amphitheatre, and online. Similar to previous years, Google I/O 2026 will feature keynotes from company leaders, fireside chats, product demos, and more.

The Verge AI · 2 min

→05

Apple is reportedly planning to launch AI-powered glasses, a pendant, and AirPods

The second-gen Ray-Ban Meta smart glasses. | Photo by Amelia Holowaty Krales / The Verge Apple is pushing ahead with plans to launch its first pair of smart glasses, along with an AI-powered pendant and camera-equipped AirPods, according to a report from Bloomberg's Mark Gurman. The three devices come with built-in cameras and will connect to the iPhone, allowing Siri to use "visual context to carry out actions," Bloomberg reports.

The Verge AI · 2 min

→06

Anthropic releases Sonnet 4.6

Anthropic has released a new version of its midsized Sonnet model, keeping pace with the company's four-month update cycle.

TechCrunch AI · 2 min

→07

Mistral AI buys Koyeb in first acquisition to back its cloud ambitions

Mistral AI has agreed to buy Koyeb, a Paris-based startup that simplifies AI app deployment at scale and manages the infrastructure behind it.

TechCrunch AI · 2 min

→

03THE TENSION2▸

PERSPECTIVE A

“ResearchGym: Evaluating Language Model Agents on Real-World AI Research”

— arXiv AI

arXiv:2602. 15112v1 Announce Type: new Abstract: We introduce ResearchGym, a benchmark and execution environment for evaluating AI agents on end-to-end research. To instantiate this, we repurpose five oral and spotlight papers from ICML, ICLR, and ACL.

Source →

♦vs♦

PERSPECTIVE B

“Password managers' promise that they can't see your vaults isn't always true”

— Ars Technica AI

Contrary to what password managers say, a server compromise can mean game over.

Source →

04RESEARCH3▸

ResearchGym: Evaluating Language Model Agents on Real-World AI Research

arXiv:2602. 15112v1 Announce Type: new Abstract: We introduce ResearchGym, a benchmark and execution environment for evaluating AI agents on end-to-end research. To instantiate this, we repurpose five oral and spotlight papers from ICML, ICLR, and ACL.

arXiv AI· Aniketh Garikaparthi, Manasi Patwardhan, Arman Cohan· 3 min· 95d agopaperSource →

da Costa and Tarski meet Goguen and Carnap: a novel approach for ontological heterogeneity based on consequence systems

arXiv:2602. 15158v1 Announce Type: new Abstract: This paper presents a novel approach for ontological heterogeneity that draws heavily from Carnapian-Goguenism, as presented by Kutz, Mossakowski and L\"ucke (2010). The approach is provisionally designated da Costian-Tarskianism, named after da Costa's Principle of Tolerance in Mathematics and after Alfred Tarski's work on the concept of a consequence operator.

arXiv AI· Gabriel Rocha· 3 min· 95d agopaperSource →

Attention-gated U-Net model for semantic segmentation of brain tumors and feature extraction for survival prognosis

arXiv:2602. 15067v1 Announce Type: new Abstract: Gliomas, among the most common primary brain tumors, vary widely in aggressiveness, prognosis, and histology, making treatment challenging due to complex and time-intensive surgical interventions. This study presents an Attention-Gated Recurrent Residual U-Net (R2U-Net) based Triplanar (2.

arXiv AI· Rut Pate, Snehal Rajput, Mehul S. Raval, Rupal A. Kapdi, Mohendra Roy· 3 min· 95d agopaperSource →

05BUILDERS3▸

Password managers' promise that they can't see your vaults isn't always true

Contrary to what password managers say, a server compromise can mean game over.

Ars Technica AI· Dan Goodin · 2 min· 96d agoSource →

Most VMware users still "actively reducing their VMware footprint," survey finds

Broadcom's "strategy was never to keep every customer," CloudBolt report says.

Ars Technica AI· Scharon Harding · 2 min· 96d agoSource →

Letting the Robots In

AI agents are increasingly browsing the web alongside humans, but most websites still serve them the same cluttered HTML designed for browsers. Here's what I've done to make this site more readable for AI systems, why I think educators should care about being AI-accessible, and why the real problem isn't scraping but the attribution gap that AI companies have chosen not to fix.