Ben Denny

[Ben Denny @BenDenny]

I feel like I sent you a message about this last time I saw you saying this in your timeline and it sat unread (not rejected, unread). Since you don't have a career's page on your company website, this is sort of a game of "try to figure out how to contact Tommy" at the moment.

Since you are also not on wellfound/workatastartup/indeed, etc., the clearest thing talking about hiring does for you right now is drive engagement to your tweets.

[Ben Denny @BenDenny]

@theisaacmed I'm honestly curious, as a contract writer who knows at least some out-of-work or underemployed writers: How are you looking and what are you looking for? It's not like there's *none* of us on the market. Is it just the slop rendering them invisible?

[Ben Denny @BenDenny]

People should, broadly speaking, apply to this: Pangram is and is likely to remain a very important part of the new world we are living in, and any resume entry saying you worked with them is going to be worth a lot.

[keysmashbandit @keysmashbandit]

If you are a smart and talented writer who would like to get paid some amount of money in exchange for your part-time labor: DM me with a short writing sample you are very proud of

Number or data enjoyers preferred but not required

[Ben Denny @BenDenny]

@keysmashbandit I sent something! Like others, I suspect my DM is caught in the X shadowrealm unless you go looking for it.

[Ben Denny @BenDenny]

@Marc__Watkins I mean, people keep saying it's an arms race, but when every single product that's supposed to defeat Pangram is all of "Does that by degrading text quality" and "still doesn't work", it's hard to see why. Usually in an arms race there's at least SOME actual competition.

[Ben Denny @BenDenny]

@agapekeleta Like don't get me wrong I'll check it out tomorrow but text-degrading rephrasers have always worked against AI detection to some extent. People don't use them; the non-detectability doesn't matter when the text isn't useable. You have to have both.

[Ben Denny @BenDenny]

@agapekeleta That's the thing, though. Yeah, the "humanized" passes pangram, but it does it by degrading the text to unusuable levels. So on one hand you have it making it a little worse than normal claude output (and just as detectable) and on the other it's unusably bad/non-natural.

[Ben Denny @BenDenny]

Deft yesterday or the day before, now this, both claiming to beat Pangram, both out-of-the-box detectable. It's like, want a bunch of views? Just slightly repackage an existing model and say it beats detection. Nobody checks!

[Agape Keleta @agapekeleta]

I trained an LLM to remove AI slop and to read more natural by matching how humans actually write.

It can also remove the newly introduced statistical watermark embedded in Claude-generated text.

Across my evaluation set, Pangram v4 classified every output as human-written while keeping high quality.

If you'd like to try a demo, you can find a link below. 1/n

[Ben Denny @BenDenny]

I know I say this a lot, but there are so many people obviously pulling "We know Pangram gets tripped up if you say too many big words" stuff out of their ass on the internet at all times that it's hard to believe they aren't conspiring to do it as a meme thing.

[Ben Denny @BenDenny]

@profgoose @jon_stewartmill Do "we know" that? Point me at some studies or examples of us knowing that, please.

[Ben Denny @BenDenny]

@damintoell I was screaming this from the rooftops on the very first day they implemented this. @Substack and @chrisbest implemented pangram, buried in the UI, and then let the slop-folk just choose to be immune to it. It was a PR thing they never intended to work.

[Ben Denny @BenDenny]

Oopsie!

[doubleunplussed @doubleunplussed]

@MaxRovensky Bro that is not even slightly the whole diff, you obviously had AI edit the whole thing. 58% looks about right!

[Ben Denny @BenDenny]

Imagine going "This program doesn't work it says my work is AI assisted" then someone asks for receipts and you run back and take out the assisted part, then send the link, and it shows up as 100% human (accurately) and you say you are only taking a partial L knowing all that

[Ben Denny @BenDenny]

@kthai1618 My question remains, it's just a matter of who I'm pointing the finger at. What gives, 2010 social scientists?

[Ben Denny @BenDenny]

@kthai1618 what you got against seventh graders?

[Ben Denny @BenDenny]

@VBallTuan That's why you use the phone, or labor-saving devices normal people can afford, etc. when you make this argument. Because you can show ways the average person is benefiting. "Cost of diamonds has fallen" isn't it, and 200k over ten years for cars is "cost of diamonds" thinking.

[Ben Denny @BenDenny]

I'm a conservative who LOVES capitalism, but

"My wealthy wife buys a brand new Tesla at least every ten years, cash, sometimes spending a middle-class year's salary to do so. They are much bettery recently"

Is a *hell* of a tone-deaf statement. Like wow, that's bad.

[Paul Graham @paulg]

The Tesla that Jessica just bought cost $48k and has a range of about 350 miles. The one she bought in 2015 cost $135k, adjusted for inflation, and had a range of 270 miles. People who think they dislike capitalism don't know how good they have it.

[Ben Denny @BenDenny]

@VBallTuan Again, I think they def deserve the cars; I have no ill-will to Paul, he seems like a good guy! He's a good writer! But "This thing most of you still can't afford is cheaper, for people like me in particular" is a *weak, bad* version of the argument.

[Ben Denny @BenDenny]

@hooji2 I mean, yes, we both agree Paul isn't a slave who has to somehow bend his will to socialists. The part where you think this makes "My wife bought two expensive cars, socialists don't know what they are talking about" an actual argument against socialist views is depressing.