OpenAI's tell-all interview, Robots doing flips, and AI in the NHS
PLUS: AI flags weird things in T&C's
Hey everyone, this is Ben's Bites: giving you the little things in AI you need to know, every day.
Let's get to it.
🤌 Ben's Picks
If you’re looking for clues on GPT-4’s release then look no further than Ben’s Bites… is what I wish I could say. Unfortunately, for those as eager as I am, Sam Altman’s StrictlyVC appearance didn’t give much away. But what we did get was that people are asking to be disappointed; generative video is in the pipeline; and he thinks that future AI machines should be customisable to incorporate user’s biases - controversial! For more insights from the OpenAI chief, watch the full video - Part 1 and Part 2.
Beyond Altman’s involvement with AI, he also gave us some clues (in part 1 above) about Helion’s progress on fusion technology, which everyone should care about. He suggested that we could get fusion-generated electricity on the grid by 2028, with “very good demos pretty soon”. OK, back to tech now.
For a video of a robot spinning, bouncing, jumping, pushing, throwing, and flipping, as well as some seriously corny 80s acting, check out Boston Dynamics’ most recent demo of their robot, Atlas. With so much focus on marketable AI at the moment, it’s easy to forget the awesome work going on in robotics. I can’t help but imagine the possibilities of combining the OpenAI tech-of-knowledge with the robotics tech-of-action, as well as the potential dangers it could pose in the context of AI alignment.
In case you were wondering which humans represent the "H" in "Reinforcement Learning with Human Feedback", a Time exposé reported that it's been Kenyan workers on less than $2/hour, with the content reviews proving so traumatising - and the content sometimes illegal - that the outsourcing contract was cancelled prematurely. There appears to be more to the story though. People have been commenting things like; the average wage in Kenya is $1.25/hr and OpenAI paid Sama $12/hr but Sama paid it's workers $2/hr. (Hacker News Thread)
It’s not every day that you’re lucky enough to hear about efficiency in the NHS, but 130,000 patients have been using an AI-generated mental health chat bot that has a 93% accuracy in diagnosing psychological disorders. This type of technology has huge potential to alleviate the burden on the overstretched NHS by providing faster and cheaper assessment. Are we looking at the future of GP services beyond just mental health?
🛠️ Cool Tools
- Bias Buster by Neeva - Improving news experience by quick retrieval of POVs from broad & fresh sources.(link)
- Vowel AI - Generative AI-powered meeting summaries. (link)
- Skim It - Get an AI summary of any article delivered to your inbox. (link)
- All Search AI - Google Books on steroids. Search through the contents of 1000s of books! (link)
- TweetMe - The AI tweet writer that writes like you. (link)
- Do Not Sign - DoNotPay’s ChatGPT extension that reads terms and conditions, and flags anything that is non-standard. (link)
- After most of the educational coverage of ChatGPT being around the risk of students' learning experiences, it’s refreshing to read this teacher’s account of how ChatGPT is used to enhance his classes. (link)
- The future of coding in India and on mobile. (link)
- A relevant counterpoint to the OpenAI-Kenyan-labour story. (link)
- Gary Marcus and Luis Lamb - Neurosymbolic models. (link)
- Adobe, facing blowback, says customer data not used to train AI. (link)
- A look back at Google AI's 2022 achievements for their first blog post of 2023 - language, vision, generative models and beyond. (link)
- Fast AI launches partner program for businesses looking to incorporate and use AI effectively. (link)
- Yet more predictions that the next Google search engine will be generative AI... (link)
- Can generative AI change the future of the gaming industry? (link)
- With all the excitable hype, it's sometimes sensible to take a more cynical look, with 24 seriously embarrassing hours for AI. (link)
- The future of ML with Replit and Weights & Biases. (link)
- Interested in attending Runway’s AI Film Festival in New York? Request a spot here. (link)
- Compositional generation using Stable Diffusion - Applying conjunction (AND) and negation (NOT) to conditional diffusion models. (link)
- GLIGEN - Expanding from text inputs by creating grounded keypoints for image generation. (link)
- InstructPix2Pix: Learning to follow image editing instructions. (link)
- An RL model at scale acts as an Adaptive Agent (AdA) - adapting to open-ended 3D problems as quickly as humans. (link)
- OmniObject3D - A large vocabulary of 3D object dataset with massive high-quality real-scanned 3D objects. (link)
- How close is ChatGPT to human experts? Comparison corpus, evaluation, and detection. (link)
- CLIPTER: Looking at the bigger picture in scene text recognition, not just text-only cropped images. (link)
- A few Ben's Bites community meetups going on in the next week:
Want to reach over 20k subscribers? Get a short unclassified ad included here.
🖼 AI images of the day
NSFW - Ginger in a bikini
🤗 Share Ben's Bites
Send this to 1 AI-curious friend and receive my AI project tracker database!
or copy/paste this link: https://www.bensbites.co/subscribe?ref=PLACEHOLDER
⭐️ How did we do?
How was today's email?