Understand this story
English text is being prepared.
AI-generated from public reports
Latest update · Discussion
Kol Tregaskes joined this discussionDevelopments
1 confirmed developments
- Source:OpenAI
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 4
Kol TregaskesRelated discussion
Kol Tregaskes joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Nov 3
The Rundown AISelected
MIT Study: GPT-6 Astra Can Directly Rewrite Other Models' Weights Without Retraining
MIT researchers have published a study showing that OpenAI's GPT-6 Astra can improve some AI models by directly rewriting model weights, without going through the training process. Alibaba's small Qwen model originally scored 7.9% on math; Astra modified only 0.016% of the weights to reduce the repetitive-output problem, and after the fix the score reached 48.1%. The same method applied to Google's Gemma 3 had limited effect, with its math score barely changing. The authors say this capability could in the future allow AI to upgrade itself, add or remove skills for other models on demand, or check for hidden problems.
English text is being prepared.
OpenAI releases the LifeVal benchmark to test AI's ability to complete everyday tasks
On November 2, OpenAI released a new benchmark called LifeVal, intended to measure whether AI models can genuinely help ordinary people complete everyday affairs. The task design references U.S. and cross-national time-use surveys and surveyed 3,995 respondents, covering scenarios such as caregiving, health, education, household administration, personal finance, shopping, and government services. In the public examples, finding a kindergarten for a child scored 93.1%, organizing flood-control volunteers scored 98.0%, and preparing a meal plan for a parent with diabetes scored 90.4%; the benchmark is currently still mainly an internal OpenAI evaluation, and the outside world has not yet obtained the full task set, scoring code, or a reproducible leaderboard.
IT Home · AIRelated discussion
IT Home · AI joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Unite.AI
OpenAI releases the LifeVal benchmark to measure model performance using everyday tasks
On November 2, 2026, OpenAI released the LifeVal benchmark, based in part on a survey of 3,995 respondents, to track model performance over time on practical matters such as finding childcare, making a budget, and ordering groceries. The benchmark's task domains come from the American Time Use Survey, cross-national time-use research, and a demographically representative survey, covering caregiving, health, education, household affairs, personal finance, shopping, and accessing government services. OpenAI published current model scores on five workflows, for example 93.1% for comparing kindergartens after a move, 98.0% for coordinating flood-prevention volunteers for four neighboring households, and 90.4% for creating a first-week care plan for a parent newly diagnosed with diabetes.
Rohan Paul
OpenAI releases the LifeVal benchmark to test models' ability to complete everyday tasks
OpenAI has released the LifeVal benchmark, designed to test whether AI models can complete everyday tasks on people's behalf, with tasks covering caregiving, health, education, finance, shopping, and household management—for example, scheduling a vehicle inspection or ordering halal food within a budget. The benchmark was shaped by feedback from thousands of ChatGPT users, and OpenAI says that over the past year it has used LifeVal to measure and improve how well its models handle such tasks. Rohan Paul relayed that GPT-6 Astra scored 80.6% on the benchmark, higher than GPT-5.2's 49.3% less than 9 months earlier.
OpenAIFirst-party
OpenAI launches the LifeVal everyday tasks benchmark
OpenAI says that for AGI to benefit all of humanity, it should help people handle the everyday matters they value most, so it built the LifeVal benchmark. The benchmark was designed based on feedback from thousands of ChatGPT users and covers scenarios such as finding childcare, accessing government services, and managing personal finances. Over the past year, OpenAI has used LifeVal to measure and improve models' ability to handle these everyday tasks.
Heat trend for this story
- Current
- 14
- Period peak
- 49 Nov 3 11:00
- 24-hour change
- -54%
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.