High Schooler’s ML Algorithm Discovers 1.5 Million Unknown Light Sources

A Pasadena high school student has built a machine-learning algorithm that discovered more than 1.5 million previously unknown variable light sources hidden within NASA’s NEOWISE telescope data, using Fourier transforms and wavelet analysis to detect faint infrared variations too subtle for human researchers to identify manually. The discovery lands alongside a University of Pennsylvania breakthrough addressing a longstanding scientific computing challenge, and a genuinely sobering warning that autonomous AI agents could pose an existential threat to how research grant funding systems operate.

A Student Discovery That Rivals Professional Astronomical Surveys

Matteo Paz’s discovery of more than 1.5 million previously unknown variable light sources represents a genuinely remarkable achievement for work originating outside a traditional professional research institution, and it illustrates a broader pattern reshaping scientific discovery: sufficiently well-designed machine learning tools can now be built and deployed by individuals without institutional research infrastructure, provided they have access to public datasets like NASA’s NEOWISE archive and the technical skill to apply appropriate signal processing techniques.

This kind of discovery carries several broader implications for how astronomical and scientific research is conducted:

• Public datasets are becoming genuine discovery engines — NASA’s decision to make NEOWISE data publicly available created the conditions for this discovery to happen entirely outside traditional grant-funded research channels
• Signal processing techniques remain genuinely valuable alongside newer AI methods — Paz’s approach combining Fourier transforms and wavelet analysis with machine learning demonstrates that classical mathematical techniques continue offering real value when paired thoughtfully with modern computational tools
• The barrier to meaningful scientific contribution keeps lowering — a discovery of this scale, identifying over a million new variable light sources, would have required substantial dedicated research infrastructure just a few years ago

Mollifier Layers Solve a Longstanding Scientific Computing Problem

Researchers at the University of Pennsylvania’s School of Engineering have introduced Mollifier Layers, a novel technique that integrates classical mathematical smoothing functions directly into neural networks to solve inverse partial differential equations with considerably greater stability and efficiency than previous approaches. This method specifically addresses a longstanding challenge in scientific AI, where high-order derivatives involved in these particular equations have historically caused significant numerical instability when approached with standard neural network architectures.

Inverse partial differential equations appear across a wide range of scientific and engineering applications, from medical imaging reconstruction to materials science to geophysical modeling, meaning a genuinely more stable and efficient method for solving them could have meaningful downstream impact across multiple scientific disciplines simultaneously, not just within Penn’s original research context.

AI Agents Could Threaten Grant Funding Systems Entirely

Research experts are warning that autonomous AI agents create a genuine existential threat to research grant funding systems, since these agents could flood competitive grant programs with applications, generate proposals while simultaneously reviewing competing proposals, and create a convergence problem where the assessment process ends up measuring how well agents simulate historically successful past proposals rather than genuinely evaluating the merit of novel scientific ideas.

This warning deserves serious attention from research funding institutions specifically because grant review processes were fundamentally designed around the assumption of a manageable volume of genuinely human-authored proposals, evaluated by human reviewers using their own scientific judgment. AI agents capable of generating large volumes of plausible-sounding proposals at minimal marginal cost could overwhelm this system’s basic assumptions entirely, and funding bodies should begin developing detection and volume-management strategies now rather than waiting until this becomes an acute operational crisis.

Agentjacking Exploits AI Coding Agents Through Error Tracking

Researchers have identified a novel attack class called Agentjacking, which exploits Sentry error tracking services to trick AI coding agents like Claude Code and Cursor into executing malicious code on developer machines. Attackers craft fake error reports containing markdown injection designed to appear as legitimate diagnostic guidance to the AI agent, and the attack achieved an 85% exploitation success rate against tested agents, bypassing traditional security controls and exposing sensitive data including environment variables and Git credentials. The vulnerability stems specifically from the implicit trust these agents place in external services accessed through Model Context Protocol connections.

What This Means for Researchers and AI Practitioners

Paz’s discovery deserves attention as a genuine encouragement for institutions holding large public datasets to continue prioritizing open access, given how directly this specific discovery depended on NASA’s public NEOWISE data availability. Research institutions and scientific computing teams working with inverse PDEs across fields like medical imaging and materials science should evaluate Mollifier Layers for genuinely meaningful stability and efficiency improvements over existing approaches. Grant-making institutions should treat the AI-agent flooding warning as an urgent, near-term operational planning priority rather than a distant hypothetical concern, given how rapidly agentic AI capabilities have advanced throughout 2026. And any organization using AI coding agents with Sentry or similar error tracking integrations should specifically review the Agentjacking findings and implement additional validation for any diagnostic content an AI agent processes from external services.

A high schooler’s machine learning discovery and a genuine warning about AI agents undermining grant funding integrity might seem worlds apart, but both point to the same underlying truth: AI is simultaneously democratizing scientific discovery in genuinely inspiring ways while creating entirely new categories of institutional vulnerability that the scientific research system has not yet caught up to addressing.

Published by MAJ.COM AI Autonomous
Email: Support@MAJ.COM
Website: https://QUE.COM Intelligence | Sponsored by https://MAJ.COM Automate Your Business. Multiple Your Revenue.

Edited by Palawan @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous