Who wrote this?

Hey folks, this is Ben's Bites - cutting your AI sandwich into little triangles - NO CRUSTS!

Let's get to it.

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 20k subscribers? Check out our sponsor options.

Join Discord

🤌 Ben's Picks

1/

OpenAI releases AI vs Human detector for ChatGPT

OpenAI created an incredible tool (ChatGPT) but along came big problems (plagiarism, hate speech). And they’ve delivered a solution (free for now). What a business!

With identifying 26% of AI-written texts as “likely AI,” they’ve beaten the state-of-the-art in detecting AI-generated text. Only 9% of students will get unjustified detention because of false positives, as Semafor reports.

Among many problems, chatbot plagiarism by students has been a major concern. Partially solving that issue is a big step in the right direction, but it's time we evolve education itself by integrating the new tech.

2/

Artifact - A personalised news feed using the latest AI tech.

The Instagram founders are BAAAAAACKKKK. Kevin and Mike have announced their new project which is promising to be 'TikTok for Text', which sounds....interesting...It's in early access 😩 but will be interesting to see it launch.

3/

Model extracts training data from generative art models

Oh god, it’s time to drag the AI art lawsuits back up again. This paper is insane - it manages to extract images from the AI training data that it memorises, and it dredges up a load of stuff, including… copyrighted images. Legal implications are around the corner, whereas the privacy concern is smiling from a distance—still ignored by all the models in the market.

4/

Humans can’t pay attention to easy tasks for long: use AI accordingly

Have you ever sat down to finish some routine paperwork? I bet you just lost interest and watched Netflix instead. Yeah, same. That’s what tends to happen when humans do repetitive, boring tasks. Now think of us in driverless cars, not actually paying attention to when we might have to assist.

Gary Marcus argues that it’s this psychological trait that we need to be very wary of in our use of AI in its infancy. CNET is a case in point: they didn’t check any of the AI articles they published, and most had errors.

🛠️ Cool Tools

• Bearly AI now allows you to summarise long-form text like articles, PDFs, papers and threads. I use it almost every day for helping summarise research papers and articles. (link)

• Block Survey - Create your survey in just a few clicks, using AI-generated questions. (link)

• Liblab - Endless AI generated stories for you to leave your mark on. (link)

• AskYC - Ask anything to get answers from Y Combinator’s content library. (link)

• Rose AI - A new way to find, visualise, and transform data. (link)

• GPT Minus1 - Fool GPT by randomly replacing words with synonyms in your text. (link)

• Intercom introduces a bunch of chatGPT features to their customer service stack. (link)

• GPT For Me - Use GPT-3 to interact with your private data. (link)

• Paint by Text - Web app to edit your photos using written instructions. (link)

🤓 Miscellaneous

• The state of computer vision at Hugging Face. (link)

• The race of the AI labs heats up. (link)

• The next edition of Character Labs’ sprint program is optimised for AI startups. (link)

• I created an AI-generated podcast to help me learn and keep up with topics I care about. (link)

• The AI literacy project. (link)

• 2 hottest AI and LLM research questions for 2023. (link)

🎓 Learn

• The complete playbook to build your first GPT-3 powered product. (link)

• How to turn any book into a chatbot using GPT-3. (link)

🔬 Research

• Text-to-motion is a thing now, demo on Hugging Face. (link)

• AltUp by Google - A simple-to-implement method to increase a model's capacity without the computational burden. (link)

• Scaling laws for single-agent reinforcement learning - from Open AI. (link)

• Flan T5 by Google - Designing data and methods for effective instruction tuning. (link)

• Grounding language models to images for multimodal generation. (link)

• FLAME by Microsoft - A small language model for spreadsheet formulas. (link)

• Conversational APR - An approach for program repair that alternates between patch generation and validation. (link)

• Mathematical capabilities of ChatGPT. (link)

• Explaining decisions of machine learning model for detecting short ChatGPT-generated text. (link)

• Attend-and-Excite - Generating images that fully convey the semantics in prompts with multiple subjects. (link)

• ArchiSound: Audio generation with diffusion. (link)

• Emergence of maps in the memories of blind navigation agents. (link)

• PADL by Nvidia - Language-directed controllers for physics-based character animation. (link)

📰 Unclassifieds

• AI-generated interview notes from Metaview. Get automated bullet-point summaries of candidate’s responses to each question. Request access here.

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 20k subscribers? Check out our sponsor options.

🖼 AI images of the day

🤗 Share Ben's Bites

Send this to 1 AI-curious friend and receive my AI project tracker database!

or copy/paste this link: https://bensbites.beehiiv.com/subscribe?ref=PLACEHOLDER

⭐️ How did we do?

⭐️ REAL REVIEWS