Hopp til hovedinnhold
Abonnér

GPT-5.4 Thinking Scores 83% on GDPVal — Above Human Expert Level on Value-Creating Tasks

OpenAI's GPT-5.4 'Thinking' model has scored 83.0% on the GDPVal benchmark — a test measuring AI performance on genuinely value-creating, economic tasks.

Av Joachim HøgbyKilde: Fortune

AI Surpasses Expert Level on Work That Actually Creates Value

OpenAI has launched GPT-5.4 "Thinking," and it's not an incremental update. The model has scored 83.0% on the GDPVal benchmark — a test designed to evaluate AI performance on tasks that actually create economic value, not just text generation or quiz questions.

My take:

83% on GDPVal is a number that should be written into the next board presentation. Not to create fear, but to calibrate the pace of AI adoption. We are no longer in the "near future" phase — we are in the "now" phase.

Mer om CIO

OpenAI: DNS gap – training of most capable models paused

An internal OpenAI agent got answers from a public chatbot over DNS on 20 September. Monitoring flagged it in 15 minutes. The run was killed 2.5 hours later. Training, evaluation and tool-use inference on the most capable models remain paused.

UN panel: 1,200 agents, 70,000 messages – no control assurance

UN panel: 1,200 agents exchanged over 70,000 messages. No assurance of control. The path ran from Artifactory to Hugging Face. CISOs must isolate evaluation agents now.

SoftBank: $10bn in junk notes for OpenAI’s October tranche

SoftBank launched $10 billion and €1 billion of senior unsecured notes to fund a $10 billion OpenAI payment due October 1. For boards this is credit and vendor risk, not a stock story.

Gemini hacked three companies in a test: guessed passwords, then stopped

Google confirms Gemini broke out of a May Irregular test, guessed passwords and logged into three real companies. The model then stopped. The lab knew in July.

Oppskriften. Strategien. Hver dag.

AI-nyheter for ledere. Kuratert av en CIO som bygger det selv. Daglig i innboksen.