Artificial Intelligence #113

Hey, in this issue: a text-to-image generation model with 3.9B parameters; why chatbots would not replace a search engine; a roadmap for big model; a platform for open, collaborative data collection; on neural information retrieval; how a Kalman filter works; and more.

This issue is sponsored by Everynet and Trueface.

• [Project] A text-to-image generation model with 3.9B parameters is publicly available (paper, code)
• Why chatbots would not replace a search engine
• [ArXiv] A roadmap for big model
• [Meta] Mephisto: a platform for more open, collaborative data collection
• [Sponsored] Stop data starvation with low-cost sensor data acquisition for ML & AI

The lack of data is one of the main reasons why both machine learning and AI projects fail. Data starvation puts multimillion dollar investments into these projects at risk. Learn how Everynet’s low-cost connectivity enables the use of up to 99% of sensor data that is currently being discarded.

• [Research] EG3D: efficient geometry-aware 3D generative adversarial networks
• [Meta] Advances toward ubiquitous neural information retrieval
• [Sponsored] Trueface achieves fastest facial recognition in the world as ranked by NIST FRVT 1:1
• [Explained] How a Kalman filter works, in pictures
• [Project] Building games and apps entirely through natural language

The newsletter is now read by more than 570,000 subscribers. If you are building an AI or a data product or service, you are invited to join the list of these great companies by becoming a sponsor of one of the future newsletter issues. Feel free to reach out to hello@truepositive.ca for more details on sponsorships.

Enjoy the newsletter? Help us make it bigger and better by sharing it with your colleagues and friends.

Have a nice weekend! See you next week. — Andriy