Sooner or later, an AI agent couldn’t solely recommend issues to do and locations to remain on my honeymoon; it might additionally go a step additional than ChatGPT and guide flights for me. It might keep in mind my preferences and funds for lodges and solely suggest lodging that matched my standards. It may additionally keep in mind what I preferred to do on previous journeys, and recommend very particular issues to do tailor-made to these tastes. It’d even request bookings for eating places on my behalf.
Sadly for my honeymoon, at the moment’s AI techniques lack the type of reasoning, planning, and reminiscence wanted. It’s nonetheless early days for these techniques, and there are a variety of unsolved analysis questions. However who is aware of—possibly for our tenth anniversary journey?
Deeper Studying
A strategy to let robots study by listening will make them extra helpful
Most AI-powered robots at the moment use cameras to grasp their environment and study new duties, but it surely’s changing into simpler to coach robots with sound too, serving to them adapt to duties and environments the place visibility is restricted.
Sound on: Researchers at Stanford College examined how rather more profitable a robotic may be if it’s able to “listening.” They selected 4 duties: flipping a bagel in a pan, erasing a whiteboard, placing two Velcro strips collectively, and pouring cube out of a cup. In every job, sounds offered clues that cameras or tactile sensors wrestle with, like realizing if the eraser is correctly contacting the whiteboard or whether or not the cup accommodates cube. When utilizing imaginative and prescient alone within the final check, the robotic may inform 27% of the time whether or not there have been cube within the cup, however that rose to 94% when sound was included. Read more from James O’Donnell.
Bits and Bytes
AI lie detectors are higher than people at recognizing lies
Researchers on the College of Würzburg in Germany discovered that an AI system was considerably higher at recognizing fabricated statements than people. People often solely get it proper round half the time, however the AI may spot if an announcement was true or false in 67% of instances. Nonetheless, lie detection is a controversial and unreliable expertise, and it’s debatable whether or not we must always even be utilizing it within the first place. (MIT Technology Review)
A hacker stole secrets and techniques from OpenAI
A hacker managed to entry OpenAI’s inner messaging techniques and steal details about its AI expertise. The corporate believes the hacker was a non-public particular person, however the incident raised fears amongst OpenAI staff that China may steal the corporate’s expertise too. (The New York Times)
AI has vastly elevated Google’s emissions over the previous 5 years
Google mentioned its greenhouse-gas emissions totaled 14.3 million metric tons of carbon dioxide equal all through 2023. That is 48% greater than in 2019, the corporate mentioned. That is principally on account of Google’s huge push towards AI, which can doubtless make it more durable to hit its purpose of eliminating carbon emissions by 2030. That is an completely miserable instance of how our societies prioritize revenue over the local weather emergency we’re in. (Bloomberg)
Why a $14 billion startup is hiring PhDs to coach AI techniques from their residing rooms
An attention-grabbing learn concerning the shift taking place in AI and information work. Scale AI has beforehand employed low-paid information staff in nations equivalent to India and the Philippines to annotate information that’s used to coach AI. However the huge growth in language fashions has prompted Scale to rent extremely expert contractors within the US with the required experience to assist prepare these fashions. This highlights simply how necessary information work actually is to AI. (The Information)
A brand new “moral” AI music generator can’t write a midway first rate music
Copyright is without doubt one of the thorniest problems going through AI at the moment. Simply final week I wrote about how AI companies are being forced to cough up for high-quality coaching information to construct highly effective AI. This story illustrates why this issues. This story is about an “moral” AI music generator, which solely used a restricted information set of licensed music. However with out high-quality information, it’s not capable of generate something even near first rate. (Wired)