Daily Digest: Robots Learning Tricks

PLUS: still finding Nemo and dying business models

October 23, 2023

Sign up | Advertise | Ben’s Bites News
Daily Digest #268

Hello folks, here’s what we have today;

PICKS

  1. Eureka! the robots can do pen spinning tricks - A new agent by Nvidia, Eureka, allows robots to learn human skills by writing customized reward algorithms for trial-and-error reinforcement learning.🍿Our Summary (also below)

  2. Open AI isn't ready to play "spot the AI image" - AI detector tool for Dall-E 3 exists, but OpenAI is not ready to show it to the world. The tool has higher accuracy compared to previous attempts but has not met OpenAI's high standards for reliability. 🍿Our Summary (also below)

  3. Strange ways AI disrupts business models by Scott Belsky. Scott Belsky’s (CPO @Adobe) latest newsletter talks about different ways AI will disrupt business models, creative outlets, marketing patterns and more. 🍿Our Highlights (also below)

TOP TOOLS

View more →
*sponsored

WHO’S HIRING IN AI

  • OpenAI - Creating safe AGI for all.

  • Scale - Bring human intelligence to software.

  • Microsoft - Leading the new era of AI.

  • Inworld - Crafting unique stories for NPC interactions.

  • Pinecone - Vector databases for everyone.

  • Coreweave - The GPU cloud.

  • Synthesia - Text to videos in minutes.

  • Adept - A new way to use computers.

NEWS
QUICK BITES

Nvidia has developed a new AI agent, Eureka, to teach robots complex skills like pen-spinning. Eureka allows robots to learn skills similar to humans by writing customized reward algorithms for trial-and-error reinforcement learning.

What is going on here?

The Eureka AI system can rapidly teach real-world skills to robots by writing automated reward functions.

What does this mean?

Eureka uses GPT-4 to write automated reward functions. Without needing specialized prompting or templates, Eureka can generate reward functions that outperformed human-written ones over 80% of the time, improving robot performance by over 50% on average.

Leveraging GPU-accelerated simulation in Nvidia’s Issac Gym, the system can swiftly test reward algorithms and constantly refine based on results. Eureka can be used to teach a diverse set of skills, like opening cabinets, tossing balls, and dexterous pen-spinning tricks to all types of robots, from robot arms to humanoids.

Why should I care?

Reward functions are chocolates for doing the right thing. They help researchers in steering the AI model’s behaviour to meet a desired goal. But writing good reward functions is hard. Eureka’s auto-generated programs could replace tedious and suboptimal human coding, making it far easier to train robots to perform complex real-world tasks.

Another highlight is that a lot of work can be done by combining the emerging tech in different fields (for eg: simulation labs, robotics and LLMs in Eureka’s case).

QUICK BITES

AI detector tool for Dall-E 3 exists, but OpenAI is not ready to show it to the world. The tool has higher accuracy compared to previous attempts but has not met OpenAI's high standards for reliability.

What is going on here?

OpenAI is debating when to release a tool that can detect if an image was made by their AI art generator DALL-E 3.

What does this mean?

OpenAI's image classifier is 99% accurate on unmodified DALL-E 3 images. It remains over 95% accurate even if the image is cropped, resized, compressed or overlaid with minor text or cutouts.

For comparison, the text classifier’s accuracy ranged about 30% for multiple factors. A couple of months after release, Open AI took the text classifier offline.

Now, OpenAI is unsure what threshold of accuracy is sufficient before releasing the tool publicly. Their current focus is Dall-E 3 generate images only but even then, the further the human edits it, the harder it becomes to catch.

Why should I care?

On one hand, the tool could help detect harmful deepfakes and AI art theft. But if the accuracy isn't near perfect, it risks unfairly labelling human-made art as AI-generated.

OpenAI wants to avoid a repeat of the controversy around their previous AI text classifier tool which was criticized for low accuracy. They are being cautious and soliciting input from impacted communities like artists.

To add, the core philosophical question is, at what point does an image created with AI but subsequently edited stop being considered AI-generated? No easy answers to that one.

QUICK BITES

Scott Belsky’s (CPO @Adobe) latest newsletter talks about different ways AI will disrupt business models, creative outlets, marketing patterns and more. Here’s what Scott thinks about the business model disruption:

1) Better AI = Bad Business.

AI will enable real-time optimization and decision-making, which could incentivize some companies to constrain AI efficiency to protect their business models.

This means companies may limit how effective their AI optimization becomes if it threatens their bottom line. For example, Tinder may not want matches to be too perfect, as that would lead to user churn.

2) Time’s over for billing by the hour

Some industries built on time-based billing will need to evolve to value-based pricing. As AI speeds up workflows, billing simply by the hour will not capture the full value delivered. Instead, compensation should be based on differentiating values like experience, skills, and proprietary data that professionals leverage, not strictly time spent. For eg: Laweres, consultants and freelancers.

3) Anti-Marketing AI

Additionally, AI-powered recommendation engines will reduce bias and subjectivity in purchasing decisions. Instead of brand sway or personal relationships guiding decisions, AI will objectively surface the best and most cost-effective options based on individual preference and reviews. This threatens companies that have relied more on marketing than merit.

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

  • All 10k+ links we’ve covered, easily filterable (1 referral)

  • 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)

Reply

or to participate.