Max Spero

I also checked with CEO @max.pangram.com after reading this @emilymbender.bsky.social post, and he confirmed that Pangram never publishes benchmarks that overlap with its training data. ("That would be a pretty major faux pas.") buttondown.com/maiht3k/arch...

I consider games I played in my teens and early 20s as more formative than my childhood games (Zoo Tycoon, RuneScape, Neopets etc.). And somehow I have probably sunk as many lifetime hours into League of Legends as the rest of this list combined.

I profiled the startup behind Pangram, the AI detector everyone’s talking about. It works really, really well. I wanted to understand how they did it. and also: why people still don’t trust them. (also, news: Penguin Random House is using it.) gift link! ✨

On Monday, @bcantrill.bsky.social and I were joined by @max.pangram.com to discuss the social imperative of separating human- and AI-written content. There’s a big difference between being served home cooking and fast food!

Stoked to announce that @max.pangram.com, co-founder and CEO of @pangram.com, will be joining @ahl.bsky.social and me on the next episode of Oxide and Friends to pick up the hot (and important!) topic of AI detection. Join us, Monday, 5p Pacific! discord.gg/QrcKGTTPrF?e...

Pangram is surprisingly good. Teachers and graders should check it out. (And journal editors .) www.pangram.com

The sudden release of an accurate AI-writing detector is going to have hilarious repercussions in politics and journalism, let me tell you. (It’ll also literally save higher education.)

People have real qualms with AI detectors, and so do I, but the evidence is mounting that Pangram in particular very rarely gets it wrong (especially on longer chunks of text, but increasingly even on short ones). @brianx.bsky.social tested it and it didn't miss: www.nytimes.com/2026/08/13/t...

if you're going to post on bluesky complaining about how an AI detector flagged your writing as AI, then you probably should make sure that you're not on record talking about how you use AI for writing www.techdirt.com/2024/04/29/h...

I talked to the head of the Commonwealth Foundation and to one of the accused writers in the Granta AI scandal — & also had the past 15 years of the prize run through Pangram, which flagged only four stories as containing AI: all from this year and last. [gift link!]

We just published a guest blog post from Joe Stech @thinkweirder.com: The Information Theory Behind Why AI Writing Sucks. Check it out! pangram.com/blog/joe-ste...

Pretty cool analysis of AI use on Substack by @taylorlorenz.bsky.social

How Much of Substack Is Actually AI? I used @pangram.com (AI detection tool) to analyze thousands of posts from the top Substack newsletters across every category to find out. Here's what I discovered 👇🏻

Lately, Pope Leo has been sounding the alarm about the affect of AI on the human spirit. But several of his X posts on that subject—and others—appear to have been generated by AI.

yiiiikes

If you do this, or your company does this, the rest of society should treat you like someone who walked into Notre Dame and pooped on the altar.

You have got to be kidding me

Pangram is releasing a new model today! Pangram 3.2 is a significant improvement over 3.1 in several aspects - Improved performance on 'humanized' text - Improved performance on adversarial prompts - Minimum word reduced from 75 to 50 - Overall improvement on Claude 4.6 recall

Wow this is so cool

ai dot com purchased for $70m. METR evals going exponential. former crypto grifters signalling we're in a bubble big labs signalling recursive self improvement I wonder, if the bubble pops, will it even matter?

I was going to write a whole effortpost about model evaluations but instead here, just take this chart

There are only two hard problems in computer science: building artificial superintelligence, and naming things

🚨 New Study 🚨 @arxiv.bsky.social has recently decided to prohibit any 'position' paper from being submitted to its CS servers. Why? Because of the "AI slop", and allegedly higher ratios of LLM-generated content in review papers, compared to non-review papers.

so far I have found three separate tpot starter packs with nearly zero overlap. completely distinct ingroups. interesting

Hello bluesky.... its @max_spero_ on x dot com formerly twitter