Saw this in my way home from the gym this morning. Top comment on the news article: “You can’t park there” Photo credit, the Sun Times News.
I know I'm like four years behind, but I just watched Stranger Things Season 4 "Chapter Four: Dear Billy" for the first time last night. It really stressed me out! Some catharsis at the end but damn if my heart rate didn't come back down for about two hours.
Content provenance technologies are becoming more important than ever. Just ask my pet bear cub. (This video was 100% AI generated)
I read somewhere that we were lucky to live during the brief period of human history after the invention of the photography/video and before it was trivial to mimic reality with AI. Now enjoyment of any cool image or video is tainted by the concern we’re being tricked. I think about that a lot
Thanks, I hadn’t seen that yet! Another vector for false negatives, and why content provenance is a better solution
And, content could show as positive for SynthID if someone runs a real image through Nano Banana and asks it to change nothing, or make a minor/invisible change. Of course, the original image would still scan clean, but people could use false positives to sow doubt.
Content could show as negative for SynthID and still be AI generated, if it was generated by any AI other than Google. It will likely never be possible to definitively tell if content was AI generated, which is why content provenance for REAL content is so important.
I'm excited to see Google's SynthID getting traction. It's an invisible watermark to make it easier to tell if an image/video was AI generated. But a reminder that SynthID detection can yield both false positives AND false negatives for AI content:
The elf is back. Little dude likes to play with fire #dontblink #doctorwho #weepingangel
A message from my daughter to my Whovian friends.
In honor of TARDIS Day I'm resharing my daughter's bass guiTARDIS. Still trying to convince her to pose with it wearing my 12th doctor cosplay. #TARDISDay #doctorwho #TARDIS
I thought the Indeed AI Work Report was pretty fascinating. In particular this chart that goes skill by skill and breaks down how much each will be transformed by AI. It reinforces that few jobs will go away completely due to AI, but the vast majority will be transformed by it to some degree.
What do we think is the best way to go? Agent-to-Agent (A2A)? Agent Communication Protocol (ACP)? Agent Network Protocol (ANP)? Model Context Protocol (MCP)? Something else?
The launch of ChatGPT Atlas makes me think about how inefficient it is for AIs to use websites made for humans. It seems clear that we'll end up with a lot of traffic happening directly between AI agents - I hope that as we move in that direction we can keep a focus on standards and interoperability
Check out our new pet bear cub!*
*Of course, this is not real. But it demonstrates another step in the inexorable march toward AI content becoming impossible to detect. Watermarks are easy to remove. We need to pivot to data provenance to verify what's real instead of trying to detect what's fake.
A bright side about Nano Banana is it's giving people both "wow, that's incredible" moments, immediately followed by the "oh, so deepfakes are now easily available to anyone" realization. The conversation around content provenance just went from "important" to "hair-on-fire" #AIgovernance #AIsafety
UPGRADING IS COMPULSORY (but good governance is essential.) As we build powerful new technologies, we must ensure they serve humanity, not the other way around. On that note, I’m excited to share that I completed Georgetown University's certificate in AI Governance and Compliance.
#7 Digital equity, which is maybe a good one to close on. I actually think this will be very a very powerful tool...that is currently only available to those with the money and knowledge to use it, potentially worsening the "productivity divide".
#6 Enabling malicious use...flipping it around, ChatGPT agent could be a powerful tools for bad actors themselves to use, phishing, disinformation, etc. 7/8
#5 Prompt injection...again kudos to them for calling this one out directly, but I think it's only a matter of time before someone figures out how to do this, and commandeer your agent with a rogue link or similar. 6/8
#4 Data exposure and handling! Now I'm connecting ChatGPT to my GMail and logging it into other websites with my usernames and passwords. OpenAI has been making strides regarding privacy and security but this ups the ante. Also more catastrophic now if my ChatGPT account is compromised. 5/8
#3 Accountability is unclear to me...if my ChatGPT buys that non refundable ticket, am I on the hook? OpenAI? The airline? I'll give you three guesses but the first two don't count. 4/8
#2 They put the "Biological and Chemical Risks" one right out there. It's great they are being proactive about this! But if past jailbreaks are any indicator, it's only a matter of time before people get past this 3/8
#1 Taking hallucinations directly to the real world! Now when my AI gets confused it can seal the deal by ordering me a $99,999.99 Michael Jordan Autographed Rookie Card, or booking non refundable tickets to Florence. Yeah it's supposed to get approval from the user, but I know how that can go 2/8
I'm excited to try the new ChatGPT agent! That said, I have...*concerns*. A thread! #AI #AIgovernance #AIrisk (1/8)
Ooo we made it until July with this prediction. "Taking over" might be putting it strongly, but I'd say that the ChatGPT agent launch confirms this one.
It was back in March that we met #davidtennant at Cleveland Fan Expo, but too fun not to share. He was lovely to meet in person even if he looks a bit nonplussed in the photo 😆 #doctorwho
So I took my mom to one of those paint your own pottery places for Mother’s Day. There was really only one choice of what to put on my mug. #DungeonCrawlerCarl #LitRPG
Key finding from the new 2025 Stanford AI Index: While enterprise AI adoption continues to speed up, implementation of formal governance frameworks remains low. The gap between deployment and oversight is a significant operational risk for organizations.
#AIGovernance #AIRisk
Colin I already loved you but now I extra love you.
