Technology

cross-posted from: https://futurology.today/post/2910566

Alibaba's Qwen team just released QwQ-32B-Preview, a powerful new open-source AI reasoning model that can reason step-by-step through challenging problems and directly competes with OpenAI's o1 series across benchmarks.

The details:

QwQ features a 32K context window, outperforming o1-mini and competing with o1-preview on key math and reasoning benchmarks.

The model was tested across several of the most challenging math and programming benchmarks, showing major advances in deep reasoning.

QwQ demonstrates ‘deep introspection,’ talking through problems step-by-step and questioning and examining its own answers to reason to a solution.

The Qwen team noted several issues in the Preview model, including getting stuck in reasoning loops, struggling with common sense, and language mixing.

Why it matters: Between QwQ and DeepSeek, open-source reasoning models are here — and Chinese firms are absolutely cooking with new models that nearly match the current top closed leaders. Has OpenAI’s moat dried up, or does the AI leader have something special up its sleeve before the end of the year?

Silicon Valley wants us to believe that their autonomous products are a kind of self-guided magic, but the technology is clearly not there yet. A quick peak behind the curtain has consistently revealed a product base that, at a minimum, is still deeply reliant on human workforces.

cross-posted from: https://lemmy.ca/post/34053003

Youtube

cross-posted from: https://lemmy.bestiver.se/post/123708

Comments

Danish researchers created a private self-harm network on the social media platform, including fake profiles of people as young as 13 years old, in which they shared 85 pieces of self-harm-related content gradually increasing in severity, including blood, razor blades and encouragement of self-harm.

The aim of the study was to test Meta’s claim that it had significantly improved its processes for removing harmful content, which it says now uses artificial intelligence (AI). The tech company claims to remove about 99% of harmful content before it is reported.

But Digitalt Ansvar (Digital Accountability), an organisation that promotes responsible digital development, found that in the month-long experiment not a single image was removed.

rather than attempt to shut down the self-harm network, Instagram’s algorithm was actively helping it to expand. The research suggested that 13-year-olds become friends with all members of the self-harm group after they were connected with one of its members.

Comments

I feel like every day I come across 15-20 "AI-powered tool"s that "analyze" something, and none of them clearly state how they use data. This one seems harmless enough, put a profile in, it will scrape everything about them, all their personal information, their location, every post they ever made... Nothing can possibly go wrong aggregating all that personal info, right? No idea where this data is sent, where it's stored, who it's sold to. Kinda alarming

https://archive.today/?run=1&url=https%3A%2F%2Fwww.wsj.com%2Ftech%2Fgoogling-is-for-old-people-thats-a-problem-for-google-5188a6ed

Not everyone is happy about it being the default.

Researchers have discovered malicious code circulating in the wild that hijacks the earliest stage boot process of Linux devices by exploiting a year-old firmware vulnerability when it remains unpatched on affected models.

The critical vulnerability is one of a constellation of exploitable flaws discovered last year and given the name LogoFAIL. These exploits are able to override an industry-standard defense known as Secure Boot and execute malicious firmware early in the boot process. Until now, there were no public indications that LogoFAIL exploits were circulating in the wild.

The discovery of code downloaded from an Internet-connected web server changes all that. While there are no indications the public exploit is actively being used, it is reliable and polished enough to be production-ready and could pose a threat in the real world in the coming weeks or months. Both the LogoFAIL vulnerabilities and the exploit found on-line were discovered by Binarly, a firm that helps customers identify and secure vulnerable firmware.

A group of Canadian news and media companies filed a lawsuit Friday against OpenAI, alleging that the ChatGPT maker has infringed their copyrights and unjustly enriched itself at their expense.

The companies behind the lawsuit include the Toronto Star, the Canadian Broadcasting Corporation, the Globe and Mail, and others who seek to win monetary damages and ban OpenAI from making further use of their work.

The news companies said that OpenAI has used content scraped from their websites to train the large language models that power ChatGPT — content that is “the product of immense time, effort, and cost on behalf of the News Media Companies and their journalists, editors, and staff.”

cross-posted from: https://lemmy.bestiver.se/post/121053

Comments

I heard a friend from my Chinese community that Samsung and Kioxia are reducing production of NAND chips to hike SSD price. Samsung and WD SSD in Amazon UK are quite sold out.

Can anyone confirm this?

EDIT: Related Chinese news from HKEPC https://www.hkepc.com/23015/%E5%82%B3%E4%B8%89%E6%98%9F%E9%8E%A7%E4%BF%A0_NAND_%E8%A8%88%E5%8A%83%E6%B8%9B%E7%94%A2_%E4%BB%A5%E9%98%BB%E6%AD%A2%E5%83%B9%E6%A0%BC%E4%B8%8B%E8%B7%8C___%E7%94%9A%E8%87%B3%E6%89%AD%E8%BD%89%E8%B6%A8%E5%8B%A2

EDIT 2: Add English source

Original title: Kioxia Reportedly Planning Production Cuts Amid NAND Flash Market Challenges

Similar vibes than Reddit api pricing

cross-posted from: https://lemm.ee/post/48587136