Throughout my first 2.5 years at OpenAI, I labored on the Robotics staff on a moonshot thought: we wished to show a single, human-like robotic hand to unravel Rubik’s dice. It was a tremendously thrilling, difficult, and emotional expertise. We solved the problem with deep reinforcement studying (RL), loopy quantities of area randomization, and no real-world coaching knowledge. Extra importantly, we conquered the problem as a staff.
From simulation and RL coaching to imaginative and prescient notion and {hardware} firmware, we collaborated so intently and cohesively. It was an incredible experiment and through that point, I typically considered Steve Jobs’ actuality distortion subject: while you consider in one thing so strongly and carry on pushing it so persistently, in some way you can also make the inconceivable doable.
For the reason that starting of 2021, I began main the Utilized AI Analysis staff. Managing a staff presents a special set of challenges and requires working model adjustments. I’m most happy with a number of initiatives associated to language mannequin security inside Utilized AI:
- We designed and constructed a set of analysis knowledge and duties to evaluate the tendency of pre-trained language fashions to generate hateful, sexual, or violent content material.
- We created an in depth taxonomy and constructed a robust classifier to detect undesirable content material in addition to the explanation why the content material is inappropriate.
- We’re engaged on numerous methods to make the mannequin much less more likely to generate unsafe outputs.
Because the Utilized AI staff is training one of the best ways to deploy cutting-edge AI methods, reminiscent of massive pre-trained language fashions, we see how highly effective and helpful they’re for real-world duties. We’re additionally conscious of the significance of safely deploying the methods, as emphasised in our Constitution.
