🔮🗞 Publishing in the platform; globalisation; Google's deep learning chip; panpsychism, elephants, poo & graphene++ #108
Platforms and publishers. How a hoax spread. Uber’s psychological tricks to keep drivers working. TPU chip saves Google from building new data centres. New federated learning approach. Is matter conscious? Algorithmic ethics problem. The state of globalisation. Eating humans and living with elephants.
🚀 Forwarded this from a friend? Sign-up to Exponential View here.
📰 Oldskool RSS lover? We are on Feedly.
Dept of the near future
👩🏾⚕️ The platform press by Emily Bell & Taylor Owen. Traditional publishers and Silicon Valley platforms switch roles, as the latter take over the editorial switchboard. “If the speed of convergence continues, more news organizations are likely to cease publishing—distributing, hosting, and monetizing—as a core activity.” Long read, worth it. (Also, fascinating analysis by the Digital Forensic Research Lab on how the Alt-Right got #syrianhoax to spread.)
🗄 Google’s dedicated deep-learning chip, the TPU, is 80x faster and 30x more efficient than using GPUs and has saved the company from building a dozen new data centres.
Dept of artificial intelligence
Google’s new federated learning approach is cool. It allows for wide-scale privacy learning on multiple mobile devices. The approach keeps all personal training data on your device while propagating updates in a privacy-safe manner for the global model to improve. The approach is nicely power efficient as it uses cycles on your device while it is idling. (We talked about Apple using differential privacy in some of its on-device ML efforts. See EV#67.)
The tradeoff every AI company will make is balancing the need for real world training and error in the wild, argues EV reader, Joshua Gans. Comment: I think synthetic data sets and simulated environments could help address this gap. Many autonomous vehicle companies are doing that.
🤖 🖌️ A pretty comprehensive set of resources about algorithm-driven design.
[video] Girl hugging what she thinks is a robot gives us clues to how our kids will deal with embodied AIs.
Self-driving cars have a serious bicycle problem. Bikes complicate the actor-state space for autonomous vehicles. Building an AV for Arizona will be easier than for Amsterdam.
OpenAI’s experiments with unsupervised learning on text yield a single neuron that predicts sentiment. INTERESTING and highly accessible.
Profile of secretive self-driving taxi firm, Zoox.
Racial discrimination in auto-insurance pricing in the US. Algorithmic ethics problem.
Small morsels to appear smart at dinner parties
🌎 Getting over globalisation. EXCELLENT CSFB report on the challenges to globalisation and the switch to multipolarity.. (Haven’t read all of it.)
Perhaps globalisation isn’t dead. Container prices are spiking.
🚰 Graphene sieve purifies seawater.
Tesla sets a new record: passes 25k sales in a quarter.
Climate could hit a state not seen for 50m years: “temperatures were up to 18°F (10°C) warmer, ice was almost nowhere to be seen, oceans were higher than they are now”. 😟⌛
Aspiring entrepreneurs, don’t forget that Tesla Elon Musk was once x.com Elon Musk.
If eaten, the human body would provide 125,000 to 140,000 calories. (hint: not particularly nutritious)
Young poo makes aged fish live longer. FASCINATING
More rivers granted the status of “living human entities”.
Inside the elephant-human communion: a complex attempt at coexistence. 🐘
🥂 Sharing reaches new heights as scientists teleport lemonade over the internet.
🇵🇹 I am away for a few days taking a short family break. We are travelling to Lisbon, my first trip there, so excited to see what the city has to offer.
As a result, Exponential View is a bit abbreviated this week. Hope it inspires some excellent conversations.
P.S. 🔥 Our twitter handle is on fire. You should follow it.