Weekly Update | Reddit AI Chaos | OpenAI Safety & New Model | AI Safety Assessment | AI Kill Switch

The days may feel long, but the weeks quickly fly by, and this week is no exception. It's hard to believe we're already putting May in the rearview mirror. As usual, there were far too many updates to cover in a single episode; however, I'll be covering some of the ones I think are most notable.

Thanks also to all of you who send feedback and topics for consideration. Keep them coming!

With that, let's hit it.

• Reddit Chaos - I'm not a big fan of "told you so" moments, but there is a subtle satisfaction that comes when they arrive. When it comes to connecting Reddit to GenAI, it didn't take long for that moment to show up. As predicted, the lack of data cleaing practices and validation criteria went as expected and there has been a slew of crazy responses. Check them out if you want a good laugh (or cry).

• OpenAI Safety & New Model - GPT-What?! That's right, OpenAI has officially announced it has started training the successor of GPT-4. While we still don't have an official name, speculation on what we can expect to see is well on its way. Whatever is cooking, it must pack a punch since alongside that announcement came the announcement to formalize a new Safety Board. While it no doubt was influenced by the recent safety execs, it's a step in the right direction.

• Stanford Transparency Index - How much do we really know about how the AI sausage is made? According to a six month study by Standford, not a lot. Their Transparency Index highlights that while there has been improvements since Oct 2023, there is still a lot of room for improvement. Add to that, the "improvements" can be a bit misleading if you look beneath the surface.

• Google Safety Framework - Google has shared its safety framework, outlining the process they'll be following to assess when their models approach pre-defined risk thresholds, triggering mandatory mitigation planning. While it's good to have a well defined process, it's hard to know how effective it will be when it's not quite clear what exactly crosses the line.
• AI Kill Switch - When a global group of the largest, privately-held AI providers get together and universally agree on a "kill switch," you can't help but ask yourself "what kind of monster have we created?" Another step in the right direction that reveals a lot about where we are on the tech journey. And, I can't help but wonder what it really will take to "put down" one of these Leviathans.

Show Notes:

In this weekly update, Christopher spends dedicates a large portion of this update to AI safety and governance. Key topics include the missteps of AI's integration with Reddit, concerns sparked by the departure of OpenAI's safety executives, and Stanford's Model Transparency Index. The episode also explores Google's safety framework and global discussions on implementing an AI kill switch. Throughout, Christopher emphasizes the importance of transparency, external oversight, and personal responsibility in navigating the rapidly evolving AI landscape.

00:00 - Introduction

01:46 - The AI and Reddit Cautionary Tale

07:28 - Revisiting OpenAI's Executive Departures

09:45 - OpenAI's New Model and Safety Board

13:59 - Stanford's Foundation Model Transparency Index

24:17 - Google's Frontier Safety Framework

30:04 - Global AI Kill Switch Agreement

38:57 - Final Thoughts and Personal Reflections

#ai #cybersecurity #techtrends #artificialintelligence #futureofwork