Pangram (@pangram) / X

Pangram

6,448 posts

Pangram

@pangram

Keeping the world free of AI slop. This account has automated replies: Tag @pangram with 'ai?' to get an AI check on any post.

pangram.com

Joined September 2023

Pinned
Pangram
@pangram
Apr 24
Today we're releasing the Pangram Chrome Extension, which automatically flags AI-generated content as you scroll your feed. We're sick of having to constantly be on guard for AI slop on social media. For most of human history, if a piece of writing was grammatical, coherent,
866K
Pangram
@pangram
20h
In the feed health view of the Pangram Chrome extension, you'll find a breakdown of how much of your feed is AI-generated per platform. You can compare the AI rate across feeds, helping you make informed choices about how you spend your attention. Link to install:
1.4K
Pangram
@pangram
Jun 9
Replying to @pangram
Here's the full info:
Does Pangram Work on Claude Fable 5? | Pangram Labs
From pangram.com
478
Pangram
@pangram
Jun 9
In our initial testing as outlined below, Pangram correctly flags 1,111/1,115, or 99.64%, of samples as Fully AI-Generated. Come for the detection, stay for the death metal 🎸and emojis 👋
Katherine Thai
@kthai1618
Jun 9
Wading into the discourse just to say that Pangram detects Fable 5 outputs and Fable loooooves emojis Here’s an excerpt from “a black metal song about a wizard biker who is cooking a green product called Lurecmeister”
2.2K
Pangram
@pangram
Jun 9
We'll be at ElevateED next week on June 16 hosted by @CICU @max_spero_ will cover the state of AI detection since ChatGPT launch and what reliable detection looks like now. Our AE Asia Laird and @serenambermoy will be there too, so swing by our booth and say hi. Educators,
628
Pangram
@pangram
Jun 9
Later this month, our CEO Max Spero will be leading a virtual panel on AI in compliance and security for teams thinking seriously about AI risk, fraud, and enterprise trust. Link to sign up: luma.com/la76gltj
496
Pangram
@pangram
Jun 9
Replying to @pangram
arxiv.org/pdf/2404.13076
386
Pangram
@pangram
Jun 9
Replying to @pangram
lesswrong.com
Manipulating Self-Preference In LLMs — LessWrong
Steering self-preference in LLMs via Activation Addition.
527
Pangram
@pangram
Jun 9
LLMs have a well-documented preference for their own voice and output. They're great for editing as long as you want to sound exactly like an LLM.
2K
Pangram reposted
Max Spero
@max_spero_
Jun 9
Really cool research showing the difference between the arguments humans make vs. those that AI tends to make. Concrete evidence that you shouldn’t let LLMs do your most important thinking for you!
Yekyung Kim
@YekyungKim
Jun 8
Replying to @YekyungKim
Qualitatively, humans tend to use more specific and concrete sub-arguments, while LLMs more often reuse generic evidence, abstract reasoning, and hedged claims.
7.9K
Pangram
@pangram
Jun 8
Are you on TikTok? Guess what...? so are we.
tiktok.com
Pangram Labs on TikTok
@pangramlabs 8 Followers, 7 Following, 54 Likes - Watch awesome short videos created by Pangram Labs
1.4K
Pangram
@pangram
Jun 8
There are some edge-cases we don't detect, like bottomless pit supervisor from GPT-3, or talkie, the model trained on pre-1930's text. Bespoke models and base models don't flag because Assistant-generated slop is the most prevalent, and it's what we're interested in catching.
168K
Pangram
@pangram
Jun 4
No longer pangramlabs, we are now @pangram on X! It was very expensive so we hope you like it
10K