Daily intelligence on AI, technology, and the decisions that follow.
Written for people who make decisions — not people who make predictions.
Get The Signal →Free to start. No spam. Unsubscribe anytime.
New evidence shows multi-hour agent task horizons and widening control gaps. It still does not establish recursive self-improvement or economy-wide job replacement.
Read the feature →Jalapeño is running GPT-5.3-Codex-Spark in the lab. OpenAI says a technical report is coming; until then, its efficiency claims remain unmeasured in public.
Read the briefing →The 975-billion-parameter mixture-of-experts model uses 41 billion parameters per token and carries an Apache 2.0 license.
A $1.505 billion Series D values the inference platform at $17.5 billion, with Nvidia among the participating investors.
Adversa AI's June disclosure names the gap between the command your agent's safety filter reads and the command bash actually runs.
On August 2, Commission enforcement powers over GPAI providers begin, and Article 50 transparency obligations start to apply.
The metric compares token throughput with power draw, and comparisons require a common workload, latency target, and system boundary.
DeepSeek V4 Flash lists at $0.14 per million input tokens, versus $5 for GPT-5.5, while Chinese models' share of US-company traffic on OpenRouter has remained above 30% since February.
The package includes a one-year pause on state permits for data centers above 20MW and a ban on surveillance pricing based on personal data.
A 919-person survey found full fleet approval was rare, while 59% reported confirmed incidents and 29% reported suspected incidents.
Muse Spark 1.1 lists at $1.25 per million input tokens and $4.25 per million output tokens through Meta's proprietary model API.
The July 1 policy statement targets undisclosed model steering and sets up a fight with state AI-bias rules.