Taggart
@taggart-tech.com
@[email protected]. Displaced Philly boy. Threat hunter. Educator. Dad. General in the AI Resistance.
taggartinstitute.org
wtfbins.wtf
linktr.ee/mttaggart
taggartinstitute.org
wtfbins.wtf
linktr.ee/mttaggart
Vote for a true progressive
November 8, 2025 at 8:34 PM
Vote for a true progressive
Made these for a talk I'm giving. Enjoy!
November 3, 2025 at 8:56 PM
Made these for a talk I'm giving. Enjoy!
Right so this is the LA Times asking you to get prompt injected for them.
October 30, 2025 at 11:30 PM
Right so this is the LA Times asking you to get prompt injected for them.
Dang didn't even realize .site was an option!
October 29, 2025 at 10:51 PM
Dang didn't even realize .site was an option!
This is simply a failure of imagination
October 29, 2025 at 10:22 PM
This is simply a failure of imagination
Update: network infra globally impacted
October 29, 2025 at 5:28 PM
Update: network infra globally impacted
But seriously, maybe stay inside today
October 26, 2025 at 7:00 PM
But seriously, maybe stay inside today
Extra fun failure in my ASCII art test for Deepseek
October 22, 2025 at 3:47 PM
Extra fun failure in my ASCII art test for Deepseek
This chart from the excruciatingly hard-to-read Microsoft Digital Defense Report demonstrates how basic statistical understanding eludes many threat intelligence products.
October 16, 2025 at 5:25 PM
This chart from the excruciatingly hard-to-read Microsoft Digital Defense Report demonstrates how basic statistical understanding eludes many threat intelligence products.
I mean is that a mood or what
October 15, 2025 at 3:28 PM
I mean is that a mood or what
If Lethality Jane can style himself as the "Secretary of War," I can assign myself some stars.
October 8, 2025 at 3:14 PM
If Lethality Jane can style himself as the "Secretary of War," I can assign myself some stars.
Good question! I think there's a real compute environment back there.
This is consistent with other containerized Python code-exec environments available as tools to LLMs I've tested.
This is consistent with other containerized Python code-exec environments available as tools to LLMs I've tested.
October 7, 2025 at 6:51 PM
Good question! I think there's a real compute environment back there.
This is consistent with other containerized Python code-exec environments available as tools to LLMs I've tested.
This is consistent with other containerized Python code-exec environments available as tools to LLMs I've tested.
Okay, I did eventually get this to sum numbers, but here's the kicker.
IT USES PYTHON AND PANDAS TO ANALYZE THE DATA
It ships all the data as a CSV off to some other compute source rather than just figuring out what formula is necessary.
IT USES PYTHON AND PANDAS TO ANALYZE THE DATA
It ships all the data as a CSV off to some other compute source rather than just figuring out what formula is necessary.
October 7, 2025 at 6:44 PM
Okay, I did eventually get this to sum numbers, but here's the kicker.
IT USES PYTHON AND PANDAS TO ANALYZE THE DATA
It ships all the data as a CSV off to some other compute source rather than just figuring out what formula is necessary.
IT USES PYTHON AND PANDAS TO ANALYZE THE DATA
It ships all the data as a CSV off to some other compute source rather than just figuring out what formula is necessary.
Just what I wanted: a spreadsheet AI that can't do math.
October 7, 2025 at 6:31 PM
Just what I wanted: a spreadsheet AI that can't do math.
This is Office for Mac erasure
October 1, 2025 at 6:16 PM
This is Office for Mac erasure
I'm unironically thrilled by the Petersen's latest hero car acquisition: the Impala from "Supernatural"
September 27, 2025 at 6:11 PM
I'm unironically thrilled by the Petersen's latest hero car acquisition: the Impala from "Supernatural"
Welp, $dayjob is now AI Red Teaming, in large part. Basically:
September 26, 2025 at 4:01 PM
Welp, $dayjob is now AI Red Teaming, in large part. Basically:
As always, Ray said it best.
September 25, 2025 at 4:10 AM
As always, Ray said it best.
Delighted to report my favorite reasoning test still works:
September 18, 2025 at 8:22 PM
Delighted to report my favorite reasoning test still works:
Aight that was one of the hardest completes I've done in a long time. All ML Python.
You must know the enemy to defeat them.
You must know the enemy to defeat them.
September 11, 2025 at 8:07 PM
Aight that was one of the hardest completes I've done in a long time. All ML Python.
You must know the enemy to defeat them.
You must know the enemy to defeat them.
Quick review of Electron apps on my system shows that ASAR integrity checking is indeed "experimental."
September 4, 2025 at 8:20 PM
Quick review of Electron apps on my system shows that ASAR integrity checking is indeed "experimental."
I woke up from a nap with this in my head.
![C program with the array:
char *bases[4] = {"Adenisine", "Guanine", "Thymine", "Cytosine"};
And the caption "My life is in dis array."](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:dt64chr2y7fuo2dbulksht6h/bafkreibt2jvdtpbceryzvk7sd43j4xsyo57sxesx2w467kylzed2ree2qm@jpeg)
July 21, 2025 at 5:27 AM
I woke up from a nap with this in my head.
Fantastic paper from METR on the true impact of gen AI on development time. This study shows that experienced developers are less productive when using generative models than not. Specifically, on complex projects and ones in which they are experts.
arxiv.org/pdf/2507.0...
arxiv.org/pdf/2507.0...
July 15, 2025 at 6:15 PM
Fantastic paper from METR on the true impact of gen AI on development time. This study shows that experienced developers are less productive when using generative models than not. Specifically, on complex projects and ones in which they are experts.
arxiv.org/pdf/2507.0...
arxiv.org/pdf/2507.0...