Patrick Gerard

Built my first website 20 years ago, now running multiple projects with 100k + users.

building and other cool things.

[Patrick Gerard @grumpypcodes]

State-of-the-art AI image detection even on public benchmarks.

[WasItAIGenerated @wiagenerated]

We tested our AI image detector against Pangram, TruthScan and nine open models.

On five public benchmarks, WasItAIGenerated V3 matches or beats Pangram Image 1.0 on every one:

Mirage-Test (49k images): 99.6% vs 99.5%
X-AIGD: 100% vs 99.96%

[Patrick Gerard @grumpypcodes]

@ChezluiCJar how do you even not think this is AI

[Patrick Gerard @grumpypcodes]

@ChezluiCJar please guys

https://t.co/hF1opjomuG

[Patrick Gerard @grumpypcodes]

can recommend

[Patrick Gerard @grumpypcodes]

Cooked

[WasItAIGenerated @wiagenerated]

We just released tropa-voice: an AI voice detector built for today's cloned voices.

Open detectors were trained on 2019–2022 TTS. We tested them on ElevenLabs v4, MiniMax, OpenAI, Gemini & xAI. Most fall apart. Ours doesn't. 🧵

[Patrick Gerard @grumpypcodes]

@levelsio Hi @levelsio is the stuff somewhere saved? like the slow videos? would love to rerun / download

[WasItAIGenerated @wiagenerated]

ChatGPT Images 2.6 launched yesterday.
We ran 79 images from both new models through our detector, which was trained before 2.6 existed.

78/79 flagged as AI at our production threshold. ~95% still caught after WhatsApp-style compression.

#AIDetection #DeepFake #FraudDetection

[Patrick Gerard @grumpypcodes]

@_lhermann it already caught lots of attention of you, so I was wondering ;)

[Patrick Gerard @grumpypcodes]

@loktar00 same thing happened to me, then I saw the guy was literally sitting in a call center - or let's say "deepfake scam center",..

[Patrick Gerard @grumpypcodes]

@Cbb0fe @Airbnb that made my day. So do I have to check the pictures for AI now before I book something? https://t.co/O7uSXdXNEM

[Patrick Gerard @grumpypcodes]

👀

[Rikiya Takehi @rikiyatakehi]

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.

We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://t.co/bOWCcFhCVq

[Patrick Gerard @grumpypcodes]

@levelsio Rio

[Patrick Gerard @grumpypcodes]

@evgeniymikholap sick, congrats. what's your margin?

[Patrick Gerard @grumpypcodes]

@levelsio @Brazili95543959 🤣🤣🤣🤣🤣 Brasil expert

[Patrick Gerard @grumpypcodes]

@larsjaner @AmauryGrowthPR @levelsio @peterthiel amo tambem. mas ainda prefiro sem balanca kkkk

[Patrick Gerard @grumpypcodes]

@levelsio @peterthiel I thought you'd live in Lisboa?

[Patrick Gerard @grumpypcodes]

the logic is no longer yours to dissapear into

[Patrick Gerard @grumpypcodes]

@richiemcilroy how?