SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Study Claims Chinese Frontier AI Models Bear Traces of Export-Controlled GPUs

0
·1 views

A technical analysis published on unemploymentlabs.com argues that leading Chinese frontier AI models show signs of having been trained on hardware subject to US export controls. The author examines model checkpoints for forensic traces linked to specific GPU characteristics. The claim is discussed on Hacker News, where the post drew 39 points and 6 comments at the time of capture. The article's full methodology and evidence were not included in the source text provided, so the strength of the conclusions cannot be independently assessed here.

Read the full story at Hacker News

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Why Software Engineering Still Matters for Web Apps Amid the Rise of Vibe Coding

A DEV Community article argues that building web applications by relying heavily on AI-generated code, known as vibe coding, can be risky. It explains that AI tools such as ChatGPT, Claude and Cursor can produce application code from simple prompts, often without developers fully understanding or reviewing the results. The piece outlines core software engineering stages, including requirements, design, implementation, testing and maintenance, as essential for dependable applications. It highlights security, scalability, reliability and maintainability as key concerns for web apps that handle real user data and serve many users at once.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Five-Person Usability Test Can Surface Most App Problems for About $30

Developers who have built an app often can't see its onboarding clearly, since they already know what each button and step means. A small usability test with five first-time users, each working through a few tasks, is presented as a low-cost way to uncover where people get lost. The approach draws on Jakob Nielsen's argument that five users reveal most usability problems, with the advice to run several small rounds of testing and fixes rather than one large study. The article cautions that the method is for discovering issues rather than measuring conversion, and that distinct audiences should each be tested separately. It recommends choosing a single clear user path, framing tasks around user goals instead of step-by-step instructions, and recruiting testers who resemble the real audience.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Compromised GitHub accounts planted a secret-stealing payload in thousands of repositories

Security firm Aikido Security reports that about 4,837 compromised GitHub accounts opened test pull requests across 61,764 repositories beginning November 10. The payload reportedly executed in at least 1,806 of them, and SafeDep identified 8,093 repositories that received malicious commits before its search ended. The commits were made under real users' identities and, in some cases, collaborator or bot accounts, so fork-related safeguards do not stop them. The malware collects Actions secrets, tokens, and cloud credentials, and can create AWS keys or add Azure federated credentials that persist even after rotation. Affected maintainers are advised to check commit history and workflow runs for signs of the attack and to rotate and audit credentials.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

GiTRay: Free Tool Scans GitHub Repositories for Suspicious Code Before Running

A developer has released GiTRay, a free web tool designed to help users identify potential malware in GitHub repositories before installing or running them. Users paste a repository URL, and the tool scans the contents without cloning, building, or executing any code. It flags items such as install-time or auto-run scripts, code that accesses credentials or sends data elsewhere, obfuscated download-and-execute patterns, and dependencies known to be malicious. The creator notes the tool is still in development and may miss threats or flag legitimate code, so a clean result is not a guarantee of safety. The developer is requesting user feedback, particularly on confusing results and false positives.

0 comments Read more at DEV Community