NEO 1X Robot, OpenAI chips, The AI Scientist, and the future of prompt engineering

Will prompt engineering ever die? In Episode 19 of Mixture of Experts, host Tim Hwang is joined by Kaoutar El Maghraoui, Kate Soule and Shobhit Varshney. Today, the experts chat the future of prompt engineering, a new paper released about The AI Scientist, NEO 1X’s humanoid robot, and OpenAI’s in-house AI chips. Will AI takeover scientific discovery? Will everyone have at home AI assistants? Why is OpenAI investing in chip production? Tune-in for our expert’s takes.

0:00 - Intro
1:17 - Future of Prompt Engineering
11:18 - NEO 1X Robot
21:56 - AI for Scientific Discovery
31:48 - OpenAI's in-house Chips

The opinions expressed in this podcast are solely those of the participants and do not necessarily reflect the views of IBM or any other organization or entity.

[nasaofficial]

This was such a fun episode! Every MOE brings out something special, and i really loved the chemistry and discussion that happened in this episode!

[raymobula]

@4:10 funny, I’m playing around with DPSy and in this context asked Gemini if it runs a similar algorithm in the background. It answered that it does, to better interpret the prompt and user intention.

[Herbert-cn]

The best promt what you have to learn is to be humble 😊 and magic will happen

[calebweintraub]

Likely many decades away from ubiquitous use of at home robots but even once deployed en masse the embodied humanoids would serve in an executive role only assuming tasks appropriate for the form factor… would delegate tasks to specialized devices such as vacuums and pool skimmers and other tasks for which its design would be seen as suboptimal

[jasonrhtx]

Specialized applications/services (or modularization/mixture of experts) seems the most likely deployment of AI, from robotics to NLP/LLM. The cost-efficiency of a dishwasher points to keeping specialized functions separate—combining a dishwasher with an oven results in opposing design requirements. Think of the carplane, which neither drives well as a car nor flies well as a plane. 😂

[JikeWimblik]

Today to get a bot you adapt a bot already there for the task or you train your own. In the future you'll use a pre optimized extensible model that can shrink and grow relative to task and rather than train a model you'll feed in instructions and data into software that produces a second data set full of logic and neural relationships packets that the based model draws from.