Five stories from a week when Substack switched on an AI detector, OpenAI’s own models broke out of a locked test, and English councils refused to hand Palantir the NHS’s records.
I once put my own writing through a detector and it came back 98% AI. I got Claude to create me an article today and it came back at 68% human content. Like with any writing, it’s very subjective.
For images… Yes! Seeing a credit to AI on AI images is very helpful especially with how good they are at photorealism.
How long will be before companies realise that customers want to speak to humans, and not a bot who goes around in circles for a moment or two. Do we have to get used to it, or ...
I love that you are focusing on humans and how this is going to affect self-esteem. We often look at our younger generation and worry about that but I think in this climate we need to look at all ages. Everyone is struggling now with their identity and how AI is rolling out. I I’m also an AI optimist so I think this is gonna work itself out, but we need to keep our voices strong in helping to shape these things and not let tools like this. Add to the confusion.
Thanks so much, Sheryl. I honestly believe that this is a great opportunity for us to reconsider what it means to be human and the very best humans that we can be.
I’ve been writing on Substack for ten weeks. I love the platform and appreciate this community. That said, this decision feels nauseatingly slimy.
Why would Substack leadership trust AI, given the mountains of research and evidence that point to how often it’s wrong and its inherent bias? Putting human authors in a position where we have to defend our words against a machine is lousy. These are dangerous gambles.
Data is gold. It feeds the machine. If this Pangram tool is analysing everything new on Substack, how are our words, faces, and voices being used? Who’s benefitting?
I know many Substack writers are asking the same question. I raised the question in my latest article, “Big Tech’s Shadow Is Growing: Who’s Responsible When AI Is Wrong” Would love your thoughts.
Thank you for a great conversation and weekly summary
I really liked the discussion on AI on a escape route from the lab sandbox and the AI kill switch. That makes me wonder if the datacenter have one by design for a AI breach threat?
I also made my summery for the very interesting week here 👇
Talking about kill switches - do you remember the global Crowdstrike IT outage on July 19, 2024? A simple faulty security software update took down large companies, including major financial services, with millions of users.
Hyperscalers, like AWS, have had several widespread outages that have taken a big part of Internet services down for many hours or sometimes even days.
Absolutely, Finn. In fact, given all the hoo-ha that came about with mythos and Day Zero security issues, I wonder if AI tools could be used to find such bottlenecks again so they could be fixed.
AI tools can be used to find security vulnerabilities and single points of failure (SPoF), both to protect and improve reliability and also to attack critical infrastructure services.
In my last job before retirement, I focused on observability and identifying SPoFs, and we used AI tools to analyze massive amounts of operational data to detect and identify failure modes. Incidents happen all the time, and engineering effort is required to minimize the impact on end users. Sometimes the dependency chains are so complex that it is difficult for humans to anticipate all possible failure modes.
“You know how much AI you used if you used it” 😂
I once put my own writing through a detector and it came back 98% AI. I got Claude to create me an article today and it came back at 68% human content. Like with any writing, it’s very subjective.
For images… Yes! Seeing a credit to AI on AI images is very helpful especially with how good they are at photorealism.
I loved this line from Leor too. 😂
Thanks Sam! Absolute blast of a stream! 💯
How long will be before companies realise that customers want to speak to humans, and not a bot who goes around in circles for a moment or two. Do we have to get used to it, or ...
I really hope not! 😢
you can get angry at a bot. But will that spill into other parts of the society.
I love that you are focusing on humans and how this is going to affect self-esteem. We often look at our younger generation and worry about that but I think in this climate we need to look at all ages. Everyone is struggling now with their identity and how AI is rolling out. I I’m also an AI optimist so I think this is gonna work itself out, but we need to keep our voices strong in helping to shape these things and not let tools like this. Add to the confusion.
Thanks so much, Sheryl. I honestly believe that this is a great opportunity for us to reconsider what it means to be human and the very best humans that we can be.
Thanks for the conversation Sam and Leor.
I’ve been writing on Substack for ten weeks. I love the platform and appreciate this community. That said, this decision feels nauseatingly slimy.
Why would Substack leadership trust AI, given the mountains of research and evidence that point to how often it’s wrong and its inherent bias? Putting human authors in a position where we have to defend our words against a machine is lousy. These are dangerous gambles.
Data is gold. It feeds the machine. If this Pangram tool is analysing everything new on Substack, how are our words, faces, and voices being used? Who’s benefitting?
Cole, this is the question that the Substack team needs to answer for sure. Thank you for highlighting this.
I know many Substack writers are asking the same question. I raised the question in my latest article, “Big Tech’s Shadow Is Growing: Who’s Responsible When AI Is Wrong” Would love your thoughts.
Thanks Cole. 🙏
I’m listening via the podcast app and half way through it loses Leor and the audio is just Sam
Sorry Trisha. Not sure what has happened here. Will report to Substack, as I know there have been issues with the tech.
🤣🤣🤣🤣 jailbreak the bot loops...
Thank you for a great conversation and weekly summary
I really liked the discussion on AI on a escape route from the lab sandbox and the AI kill switch. That makes me wonder if the datacenter have one by design for a AI breach threat?
I also made my summery for the very interesting week here 👇
https://saraeson.substack.com/p/what-i-almost-missed-this-week-w30?r=1t83gn&utm_medium=ios
Thank you for sharing. 🙏
Great chat! Looking forward to tuning you into these every Monday.
Thanks so much Ashley! They are a genuine highlight of my week. 🙏
Talking about kill switches - do you remember the global Crowdstrike IT outage on July 19, 2024? A simple faulty security software update took down large companies, including major financial services, with millions of users.
Hyperscalers, like AWS, have had several widespread outages that have taken a big part of Internet services down for many hours or sometimes even days.
Absolutely, Finn. In fact, given all the hoo-ha that came about with mythos and Day Zero security issues, I wonder if AI tools could be used to find such bottlenecks again so they could be fixed.
AI tools can be used to find security vulnerabilities and single points of failure (SPoF), both to protect and improve reliability and also to attack critical infrastructure services.
In my last job before retirement, I focused on observability and identifying SPoFs, and we used AI tools to analyze massive amounts of operational data to detect and identify failure modes. Incidents happen all the time, and engineering effort is required to minimize the impact on end users. Sometimes the dependency chains are so complex that it is difficult for humans to anticipate all possible failure modes.
Thanks Finn. This is a really insightful response which has unintentionally left me terrified about the possibilities now available to rogue agents…
AI detectors simply don’t work. Slop is an issue of content not of style.
This is exactly my take as well, Alistair. Glad to see that I'm not the only one.