Co-Inventors of the Groundbreaking Transformer AI Architecture Launch a Tokyo-Based Initiative to Chart a New Course in Generative AI
Top ex-Google Brain researchers start AI research company in Tokyo
Two big shots from Google just set up their own AI gig in Tokyo. David Ha and Llion Jones, the brainiacs behind some major AI advances, are calling their new venture Sakana AI. Llion was part of the team that gave us the “transformer” tech that’s behind wild stuff like the ChatGPT chatbot. Instead of just supersizing that tech,
Sakana AI wants to try something new – like having a bunch of smaller AI brains working together. All the main people behind that big “transformer” idea have now left Google and they’re pulling in some serious cash for their own projects.
A Letter Prompted Talk Of AI Doomsday, Many Who Signed Weren’t Actually AI Doomers
In March, an open letter from the nonprofit Future of Life Institute, signed by nearly 35,000 AI researchers, technologists, entrepreneurs, and concerned citizens, urged a “pause” on AI development. The letter raised concerns about the capabilities of AI programs like ChatGPT and potential risks to humanity.
However, two MIT students, Isabella Struckman and Sofie Kupiec, found that the motivations behind the signatures were quite diverse. Their investigation revealed that few of the signatories actually saw AI as an imminent threat to humanity. Instead, many were worried about the fast-paced competition among tech giants like Google, OpenAI, and Microsoft. They feared that rushing to release experimental algorithms without proper scrutiny could lead to disinformation, biased advice, or increased power and wealth for already powerful tech companies. Others were concerned about AI displacing workers at an unprecedented rate or drawing public attention to the rapid advances in AI technology.
While some signatories may have disagreed with parts of the letter, they signed it to draw attention to the various risks associated with AI. However, the letter’s emphasis on doomsday scenarios may have unintentionally made it harder for concerns about less-than-superintelligent machines to be noticed or inspire action.
Researchers are helping robots teach themselves to open dishwashers and doors
Researchers at ETH Zurich are making strides in helping robots teach themselves how to open doors and dishwashers with minimal human guidance. Instead of using remote control or pre-programmed actions, their new model works in three steps: the user describes the scene and action, the system plans a rough route, and then refines that route into a viable path.
The process takes into account the robot’s movement, forces, limbs, and when and where it should interact with the object. This research breaks down the tasks into two categories: object-centric (like opening doors) and robot-centric (like moving around objects).
They demonstrated their findings on a four-legged robot called ANYmal from ANYbotics, a spin-off from ETH Zurich. This research could pave the way for fully autonomous robots that can perform tasks without human help.
AI Could Soon Help Us Talk to Animals, But There Is a Problem
Neurobiologists Yossi Yovel and Oded Rechavi from Tel Aviv University propose the “Doctor Dolittle challenge” for AI in a new essay in Current Biology. They suggest that for AI to communicate with animals, it must use the animal’s own signals, use those signals in various contexts, and elicit a response as if from another of the same species, not a machine.
This has been partially achieved with bees via a robotic bee mimicking the waggle dance, but only in a single context. Yovel and Rechavi also point out the difficulty of understanding complex animal communication. Wittgenstein argued that even if a lion could speak, we might not understand it.
Arthur releases open source tool to help companies find the best LLM for a job
Arthur, a startup into machine learning stuff, has dropped a new free tool called Arthur Bench. Its job? Help companies figure out which fancy AI model (they call ’em LLMs) is the best fit for their data. Picture this: You’ve got a bunch of these AI models, and you wanna know which one’s the champ. Arthur Bench lets you test ’em all out, especially with the kind of questions or prompts your users might throw at them. So, you could check how one model from Anthropic stands up against one from OpenAI based on your user’s needs.
The CEO, Adam Wenchel, says most companies are flying blind, not knowing which AI tool is the best. That’s why they made this. There’s also a paid version coming if you want more features or don’t want the hassle of the free one. And, just a heads up, they also launched Arthur Shield earlier, which acts like a security guard for these AI models, keeping out the bad stuff.
4 in-demand freelance A.I. jobs—one can pay $100/hour: ‘There’s incredible demand for people who are creative’
The world of AI is booming, and creative folks who know how to tap into this technology are in high demand. That’s the word from Julia Pollak, the chief economist at ZipRecruiter. Freelance marketplace Fiverr has also seen a surge in AI-related job searches, says Yoav Hornung, the head of verticals at Fiverr.
Here are four AI gigs that are making waves:
AI Consultant: They help identify AI-powered solutions for unique problems and guide them through AI-centric projects. Searches for AI consultants skyrocketed by over 650% from January to July 2023 on Fiverr. Some are raking in a cool $1,000 for just 10 hours of work.
AI Video Editor: They can make still images come to life, tailor backgrounds, and even include AI-generated voiceovers. The demand for AI video editors shot up more than 625% between January and July 2023. Some are charging as much as $1,120 per project.
AI Prompt Engineer: These folks help tailor prompts for tools like ChatGPT to create powerful blogs, articles, do research, and fact check. Some are charging a sweet $100 for just five prompts.
AI Content Editor: These editors help tidy up AI-generated content like ChatGPT blog posts. One editor is making up to $625 per project.