MIT scientists have boosted AI’s reasoning capabilities with a revolutionary hybrid architecture! Yes, the geniuses at MIT have developed three frameworks that will give AIs the power to reason like humans.
Imagine you have a friend who is a little limited in terms of judgment (we all have a [insert the first name of your closest colleague] in our entourage 😅). He understands what you’re saying and can even respond, but as soon as he has to think a little, he lags and sometimes crashes. Well, today’s AIs are a bit similar!
They’re super good at processing tons of data and generating text, but in terms of understanding context and complex reasoning, they’re not quite there yet. That’s where our friends at MIT come in with their treasure trove of natural language abstractions. Basically, they have created libraries that will help AIs better understand their environment and solve problems in a more human way. We’re talking about three frameworks with cute little names: LILO, Ada, and LGA.
LILO is the developer’s asset. It will help AIs synthesize, compress, and document code like a pro. No more AIs creating unreadable code pads; make way for well-structured and commented programs!
Ada is the group’s strategist. It will allow AIs to plan actions flexibly. Instead of stupidly following a recipe, the AI will be able to adapt and improvise according to what it finds in its fridge. Ada even showed a 59% improvement in cooking simulation tasks and an 89% improvement in bed construction tasks.
Finally, LGA is the ace of perception. It will give robots a big boost in understanding their environment, a bit like giving them a good pair of glasses. No more robots getting caught in the carpet; they will be able to navigate your apartment like a fish in water (or almost 😅).
With these three frameworks that mix neural networks and classical logical approaches, the researchers hope to take AI to the next level. Thus, we will finally have chatbots that understand your bad jokes, robots that bring you a beer without spilling half of it in the process, and maybe even AIs that help you code your next video game while you laze on the couch!
For those who want to go further, here are the publications on arXiv detailing the advances made:
I can’t wait to see it all work for real!
