Forget retraining - AI can now tweak another AI's weights

Direct parameter adjustment method lifts benchmark performance without retraining

By GPU Data Hub Desk · GPU Data Hub·Editorial standards·3 Nov 2026, 08:15

Forget retraining - AI can now tweak another AI's weights

Researchers have introduced an approach dubbed Weight-Space Programming, allowing machine learning models to adjust another network's internal parameters directly. The method bypasses resource-intensive model retraining by tuning parameters to enhance evaluation benchmark results.

Original source

This summary was written by the GPU Data Hub desk from reporting published by The Register on 3 Nov 2026, 08:15. Read the full article at the original publisher.

Read Original Source →

Topics

model optimisation neural networks machine learning benchmarks

LinkedIn X

More from GPU Data Hub

Latest →

Anthropic S-1 update reveals $895.4bn in cloud and data center commitments

Artificial intelligence firm Anthropic has reported $895.4bn in cloud and data centre obligations within an updated S-1 filing. While the business registered top-line revenue gains, the pace of its commercial expansion has shown signs of deceleration.

AI Infrastructure·2h ago·Data Center Dynamics

Cities Worldwide Build and Scale AI Services With NVIDIA Physical and Agentic AI

Municipalities globally are deploying Nvidia's physical and agentic artificial intelligence technologies to convert operational data from sensors and cameras into actionable public services. The tools support the development of digital twins for infrastructure planning alongside autonomous vision and service agents. The initiative aims to help local authorities scale automated operational workflows across municipal networks.

AI Infrastructure·6h ago·NVIDIA Newsroom

Open models can do 90 percent of agentic work at a twentieth of the cost of a frontier LLM: Gartner

Research from Gartner indicates that open-source models are capable of executing 90 per cent of agentic workloads at roughly a twentieth of the cost of frontier proprietary systems. The advisory firm suggests that underlying models are becoming commoditised assets in enterprise deployments. Consequently, commercial value is increasingly migrating towards the operational harnesses and scaffolding that direct these models.

AI Infrastructure·7h ago·The Register

Google takes Dreambeans to India in bid to bring AI-generated personal stories to the masses

Google has launched its experimental storytelling project, Dreambeans, in India to deliver automated personal tales to broader audiences. The artificial intelligence initiative has been expanded to accommodate 13 regional languages across the country.

AI Infrastructure·10h ago·TechCrunch

OpenAI hires former U.S. Army Secretary Dan Driscoll for national security role

OpenAI has appointed former US Army Secretary Dan Driscoll to steer a newly formed national security division. The role is tasked with guiding the technology firm's contributions to modernising defence and state security systems.

AI Infrastructure·10h ago·CNBC Technology

Apple may try to increase revenue by doing…something with third-party AI apps

Apple has signalled that it may capitalise on third-party artificial intelligence services to generate fresh commercial revenue. Executive remarks during a recent financial results call pointed to prospective opportunities across external software integrations.

AI Infrastructure·11h ago·TechCrunch