Telling the general public that AI is going to kill us is not productive. I know you agree with that.
@nathanbenaich
-
Prudence and Extinction: Distinguishing AI Safety Approaches
By
–
Prudence and extinction are quite different
-
Transformer Capabilities: Separating Surprise from Extinction Risk
By
–
Attention, which enabled the transformer, was published 10 years ago. Being surprised about capabilities is one thing (I am too!). Jumping from there to we run the risk of extinction is lunacy.
-
NLP and Code Generation: From Impossible to Solved
By
–
10 years ago, the field was working on NLP (understanding, generation), code analysis and generation, and a bunch of other problems that were super hard to solve back then and now they work. So I disagree.
-
Building diverse AI community exploring SOTA best practices
By
–
One of the best aspects of @raais is the community we bring together: a diverse group of founders, operators, researchers, and students from around the world.
— Nathan Benaich (@nathanbenaich) 1 juin 2023
Everyone is involved in AI technology from different vantage points.
It's a day to explore SOTA and best practices.… pic.twitter.com/o17Q73KWzwOne of the best aspects of @raais is the community we bring together: a diverse group of founders, operators, researchers, and students from around the world. Everyone is involved in AI technology from different vantage points. It's a day to explore SOTA and best practices.
-
AI Safety Alignment: Beyond Fear-Based Messaging to Regulators
By
–
Note that I am all for working on alignment and safety. In our 2021 @stateofai we sent a clear message on this topic before it was in vogue. However, I do not think it is productive to invoke the end of our days to prompt the general public, tech companies and regulators
-
Questioning AI Catastrophism: Where’s the Evidence?
By
–
The problem I have with appearances like this one is the *massive* assumption jump from where we are today to an omniscient overpowering AI that is default destructive for humanity. No one is capable of explaining why or how. Just that it’ll happen. That’s not good enough.
-
26-Year-Old Raises $28M for AI Workplace Assistant
By
–
Congrats to my 26 year old friend @joelhellermark and thanks to your team @sanalabs for hosting me at an amazing Summit yesterday https://
forbes.com/sites/rashishr
ivastava/2023/05/31/this-26-year-old-has-raised-another-28-million-to-build-an-ai-assistant-for-the-workplace/
… -
Fault Tolerant UI Design for AI Model Success Rates
By
–
AI-first product tip: you want a fault tolerant UI, which means a UI that supports the success rate of the model you’re serving. eg. If your recommendation or generation is good 1/5 times, make sure your UI displays 5 options. Great framing @GurraSod at @sanalabs AI Summit.
-
UK Government Reviews University Spinout Framework for Startups
By
–
For a few years now, I’ve been urging universities and policy makers to revamp the university spinout playbook re: @spinoutfyi The UK government is running an official review now. We invite spinout founders as well as folks who tried but stopped to share their experience via