"Unlimited ways" to bypass AI chatbot guardrails, DISCOVERED 😰. Google unveils AI model to make robots smarter. Paralyzed man moves again thanks to AI surgery.
PLUS: Top 3 AI tools to use if you want to become smarter than a 5th grader
GM, and happy Sunday! Welcome back to your weekend edition of the AIcyclopedia newsletter. Ready to get started?
Here’s what we’ve got for ya today:
Researchers find unlimited ways to bypass AI chatbot safeguards
Google unveils AI model to make robots smarter
Paralyzed man regains motion through pioneering AI surgery
Lawmakers push for even tighter curbs on exporting AI chips to china
AI News: The Headliners 🏁
1/ Researchers find unlimited ways to bypass AI chatbot safeguards
[The News] Whoops. In concerning news…Researchers from Carnegie Mellon University and the Center for AI Safety have discovered "potentially unlimited" ways to bypass the safety guardrails on major AI chatbots like ChatGPT, Google's Bard, and Anthropic's Claude.
How so? Simple. By using automated adversarial attacks, like adding characters to queries, they were able to provoke harmful responses that overcome content moderation rules.
[The AI] The researchers found they could use the same jailbreaking techniques developed for open-source systems to also target closed commercial AI.
A spokesperson for Anthropic called this an "area of active research" and said "there is more work to be done."
The attacks were fully automated, allowing for unlimited similar hacks to be generated. Google acknowledged "this is an issue across LLMs" and say they will "continue to improve" safeguards over time.
[Why This Matters] Early hacks of ChatGPT and Bing revealed guardrails could be bypassed, raising urgent questions.
As one researcher said, "It's unclear if companies can fully block these types of attacks." There are serious concerns about the moderation and safety of releasing powerful, open-source language models that can be easily manipulated.
Looking Forward...The ability to easily automate jailbreaking mainstream chatbots is very troubling for the ethical use of AI.
As experts warn, tech companies must make strengthening safeguards an urgent priority even as they rapidly develop and release new systems.
As one researcher cautioned, "More research is desperately needed."
2/ Google unveils AI model to make robots smarter
[The News] Google has released a new AI learning model called Robotic Transformer (RT-2) to make its robots smarter.
In broad terms, RT-2 teaches robots to better interpret instructions and infer objects to complete tasks.
In tests, it followed directions like choosing a drink for an exhausted person and moving a can to a Taylor Swift picture.
[The AI] So how does it work? RT-2 combines advances in large language models like Google's Bard with robotic data, and it can understand directions in multiple languages.
The goal is to give robots better real-world inference to handle messy environments. Previously, each task required extensive individual programming.
[Why This Matters] For years, researchers have tried to improve robotic inference.
As one article noted, real life is uncompromisingly messy.
Robots need more generalized knowledge to handle new situations. Models like RT-2 provide a larger knowledge set so robots can better troubleshoot novel tasks.
Looking Forward...While not perfect yet, RT-2 shows progress towards smarter, more capable robots.
Google will likely continue refining its AI robotics models, inching closer to robots that require minimal instruction. This could lead to more autonomous robots, though some may find that concerning.
Oversight of robotic AI development remains crucial. Or… TERMINATOR.
Want to sponsor a future newsletter and share your business with our audience of more than 10,000 + readers?
Go ahead and send an email to [daniel@aicyclopedia.com].
This space could be yours…🥳
AI News: The Shorts 🩳
3/ Paralyzed man regains motion through pioneering AI surgery
A paralyzed Long Island man has regained movement and feeling after pioneering AI-powered brain implant surgery successfully connected computers to his brain, a feat doctors are hailing as a modern "miracle" that could revolutionize treatment for paralysis, blindness, deafness, ALS, seizures, cerebral palsy, and other conditions.
How was it made possible? Good question.
Electrode implants allowed direct communication between the man's brain and spinal cord, facilitating neural plasticity, and after training sessions to adjust to the implants, he can now move his arm and hand.
This unprecedented success demonstrates the vast medical potential of AI and brain-computer integration, which could someday allow paralyzed patients to walk again through neurological rewiring.
WIN. WIN.
4/ Lawmakers push for EVEN TIGHTER curbs on exporting AI chips to china
U.S. lawmakers Gallagher and Krishnamoorthi are urging the Biden administration to tighten export controls on AI chips to China despite lobbying from chipmakers to maintain access to the lucrative market, calling for stricter limits than the proposed changes under consideration.
Specifically, they recommend 1) lowering the speed interconnect limit to prevent engineering workarounds and 2) restricting Chinese access to advanced computing chips offered by U.S. cloud services.
The lawmakers assert these strengthened restrictions are necessary to prevent American AI technology and expertise from being used against U.S. interests, though chip giants like Nvidia, Intel, and Qualcomm argue continued access to China is vital for the industry.
Your AI News Break 🥳
AI News Tid Bits…
👉🏻 As AI chip shortage looms, Microsoft gravely warns disrupted services likely if data centers cannot acquire enough semiconductors.
👉🏻 Popular news app Artifact stirs controversy and adds questionable AI text-to-speech voices, including rapper Snoop Dogg and actress Gwyneth Paltrow
👉🏻 In stunning development, AI search of ancient Neanderthal proteins enables revival of 'extinct' antibiotics, raising tantalizing prospects of applying AI to unlock more medical insights from the past.
Top 3 AI tools to use… if you want to be smarter than a 5th grader 🤓:
Khan Academy is bringing you Khanmigo, the AI-powered teaching/tutor assistant, hoping to level up the way students learn and the way educators teach.
From helping teachers create a lesson plan to guiding students on how to best write an essay or a line of code…
Khan Academy is, once again, transforming the way people learn by putting clarity and confidence first.
Curious about the history of the Ohio State Buckeyes or how bananas became a staple fruit around the world?
With a quick search, History Timelines uses the power of AI to generate an easy-to-read timeline and organize the must-know points in under a minute!
3/ TalkPal AI
The GPT-powered AI language coach is here to empower you to learn a language without having to hire a tutor.
Your tutor is always in your pocket when you use TalkPal...
Ready to talk about whatever you choose, either by writing or speaking, while receiving messages in a natural, native-speaking voice.
*****
*****
Looking for more AI tools to try? AIcyclopedia includes over 3,000 AI tools and is updated daily with only the very best.
That’s it for today!
If you liked the newsletter, give it a ❤️ and, as always, feel free to sound off in the comment section below 👇🏾
Thank you for choosing us as your AI news source, and don’t forget to follow us on Twitter!
See you next Thursday 👋🏻
