Recent searches

No recent searches

Search options

Only available when logged in.
eupolicy.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
This Mastodon server is a friendly and respectful discussion space for people working in areas related to EU policy. When you request to create an account, please tell us something about you.

Administered by:

Server stats:

195
active users

eupolicy.social: AboutProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.4.1

#model

10 posts8 participants0 posts today
@gurupanguji

Prompt injections are still a problem – August 2025 edition

Independent AI researcher Johann Rehberger (previously) has had an absurdly busy August. Under the heading The Month of AI Bugs he has been publishing one report per day across an array of different tools, all of which are vulnerable to various classic prompt injection problems. This is a fantastic and horrifying demonstration of how widespread and dangerous these vulnerabilities still are, almost three years after we first started talking about them.

Johann’s published research in August so far covers ChatGPT, Codex, Anthropic MCPs, Cursor, Amp, Devin, OpenHands, Claude Code, GitHub Copilot and Google Jules. There’s still half the month left!

The Summer of Johann: prompt injections as far as the eye can see

Previously:

https://gurupanguji.com/2025/06/15/prompt-injection-continues-to-be-a-major-vector-of-attack-for-llms/?utm_source=rss&utm_medium=rss&utm_campaign=prompt-injection-continues-to-be-a-major-vector-of-attack-for-llms

https://gurupanguji.com/2025/04/23/notes-on-llms/

https://gurupanguji.com/2025/08/06/trust-in-the-world-of-ai/

Simon Willison’s WeblogThe Summer of Johann: prompt injections as far as the eye can seeIndependent AI researcher Johann Rehberger (previously) has had an absurdly busy August. Under the heading The Month of AI Bugs he has been publishing one report per day across an …
#ai#engineering#llms
💧🌏 Greg Cocks

Landsliding Follows Signatures Of Wildfire History And Vegetation Regrowth In A Steep Coastal Shrubland
--
linkedin.com/posts/tmatthew_th <-- shared LinkedIn post
--
pubs.geoscienceworld.org/gsa/g <-- shared paper
---
#GIS #spatial #mapping #massmovement #debrisflow #debrisslide #California #USA #climatechange #extremeweather #fire #engineeringgeology #risk #hazard #geoscience #water #hydrology #rainfall #precipitation #wildfire #postwildfire #landslide #vegetation #cover #type #causality #coast #coastal #shrubland #model #modeling #atmosphericriver #storm #geomorphology #geomorphometry #SantaYnezMountains #imagery #remotesensing #delineation #downslope #sediment #damage #cost #economics #infrastructure #geology #regrowth #postfire
U.S. Geological Survey (USGS) USDA Forest Service California Geological Survey

💧🌏 Greg Cocks

Video Shows Pulsing and Curving Fault Behavior
--
doi.org/10.1029/2025EO250307 <-- shared technical article
--
doi.org/10.1785/0320250024 <-- shared paper
--
youtu.be/77ubC4bcgRM?si=9Rbmj3 <-- shared video
--
"A chance video captured a fault rupture during March’s devastating [Burma] earthquake, delivering real-time evidence of how major seismic tremors propagate..."
#earthquake #geology #engineeringgeology #risk #hazard #Burma #fault #pulsing #curving #realtime #video #slipzone #velocity #crust #seismic #offset #groundmotion #pulse #earthquakeengineering #engineering #slickenlines #faultslip #curved #strikeslip #infrastructure #risk #hazard #trajectory #model #modeling #coseismic #rupture #observation

C.

On "AI"...

I finally had an opportunity to personally evaluate the output of an #LLM that was asked to do a technical evaluation / summary of the state of an industry. A friend asked a modern model about a particular #technical field and asked it to summarize various aspects of it. It happened to involve supercomputing, which I have some knowledge of, and he asked me to look at the answer it gave.

It was a detailed writeup, a little over 1300 words. It had the structure of the kind of document he'd asked it to create, with appropriate sections, headers, with the information divided up sensibly, etc. It very much sounded like the type of thing you would expect a computer science #expert to respond with if asked the same question.

But some of the specific points in the answers were obviously wrong to me. Some I suspected were wrong. Some I didn't know. So I checked. I didn't check every fact and figure, but a decent number of them - and every single one I checked was flat-out wrong. It gave specific numbers or attributes for many aspects of the state of the #tech, and they were all wrong. Sometimes badly #wrong, and some stupendously wrong.

There were only a couple of things in the text that were obviously inconsistent, so unless you happened to already know something about the subject, there was almost no hint of how incorrectly it stated the facts. And of course it sounded supremely confident.

1/2

#AI#fact#incorrect
TrendingLive feeds
About