🥟 Chao-Down #310 How students are likely writing millions of papers with AI, TikTok to introduce AI avatars that can make ads, OpenAI makes ChatGPT 'more direct, less verbose'

How bad is AI plagiarism getting in education?

About a year ago, the website Turnitin rolled out an AI writing detection service trained on a dataset of student-written and AI-generated papers. After going through 200 million submissions, the site found that 22 million student papers likely incorporated AI directly in their work. 

Of these submissions, Turnitin found that at least 11% of these papers had AI-generated content make up 20% of the work while 3% of the submissions had as much as 80% or more of the actual writing be AI-content (which would literally be almost a complete copy-paste).

We know that AI detection services can be biased among non-native English speakers, and as such they’ve had a shaky rollout with universities stepping back from fully implementing them. But the stats don’t lie and we’re witnessing that in just one year’s time, teachers and parents need to face a new world as they grapple with students embrace of AI tech.

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

OpenAI makes ChatGPT ‘more direct, less verbose’ (TechCrunch)

Apple Plans to Overhaul Entire Mac Line With AI-Focused M4 Chips (Bloomberg)

Students Are Likely Writing Millions of Papers With AI (WIRED)

TikTok may add AI avatars that can make ads (The Verge)

Ready or not, AI is coming to science education — and students have opinions (Nature)

iOS 18 May Feature All-New 'Safari Browsing Assistant' (MacRumors)

Always be Learnin’ 📕 📖

Will Search Be Generative AI or Blue Links? Actually, It’s Both. (bigtechnology.com)

Building reliable systems out of unreliable agents (Rainforest QA)

Algorithmic Progress in Language Models (Epoch)

Projects to Keep an Eye On 🛠

penpot/penpot: Penpot: The open-source design tool for design and code collaboration (Github)

Filimoa/open-parse: Improved file parsing for LLM’s (Github)

OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments (os-world.github.io)

The Latest in AI Research 💡

Direct Nash Optimization: Teaching Language Models to Self-Improve with General Preferences (arxiv)

No "Zero-Shot" Without Exponential Data: Pretraining Concept Frequency Determines Multimodal Model Performance (arxiv)

CantTalkAboutThis: Aligning Language Models to Stay on Topic in Dialogues (arxiv)

The World Outside of AI 🌎

Home Prices to Jump 5% As Housing Market Looks Tighter Than in 2023: Economist (Business Insider)

America’s Chip Renaissance Needs Workers (WSJ)

Where Does the Best Innovation Happen? Not in Stand-Alone Labs (WSJ)

Dollar stores are shutting down across America. They did this to themselves (CNN Business)

Many cancer drugs remain unproven 5 years after accelerated approval (NBC News)

What's the best time to exercise? (qz.com)

One Last Bite 😋

Total eclipse of the Internet: traffic impacts in Mexico, the US, and Canada (Cloudflare)