Benjamin

@benjami

art, gamedev, accessibility, 100% human. He/him

474 Following640 Followers493 Posts

Is this your profile?Claim it and benjami.dirty.social is yours.Claim profile

Benjamin@benjami· 8d

It's 2026. I think Hank Green's @complexly.org should remove this section of their AI policy. What you're describing is storyboarding, pre-visualisation and research using unethically trained Ouroboros AI. Why are you doing this? What you're describing is stealth use of AI during production.

Screenshot of Complexly's AI Policy.

1. We will not produce videos written, edited, or fact-checked by AI.
This includes not just scripts written in their entirety by these tools, but also those with significant portions composed, edited, or fact-checked by them.

2. We will not produce videos (or any other public-facing content) containing illustrations or animations created by AI trained on non-consensual data sets.
This includes DALL•E 2 and Midjourney. Some tools are trained on images created by

people who agreed to be in the data set, and those may be used.

3. When using AI tools trained on consensual data sets to create visual elements, we will ensure we are not misleading viewers as to what is real, or contributing to misinformation. Where applicable, we will label images created this way.

4. We may use AI tools to suggest ideas (for script subjects, examples, visual elements, etc.) and do preliminary research.
Ideas come from many places in the early stages of conceptualizing and producing a video, and it’s fine for AI to be one of them. Similarly, using these tools to identify sources is acceptable as long as those sources are verified by our team.

Last updated July 24, 2023

Benjamin@benjami· Jun 29

A company makes an "AI hiring agent", claims it "outputs a fair, explainable evaluation." its gotten VERY popular. The AI gives candidates a score between 0 & 100. Someone tested it. Testing the SAME CV changes the score by a wide margin: up to 33 points 💩 Dont use LLMs for hiring.

Graph showing the growth of popularity and stars of the github repository interviewstreet/hiring-agent. The graph shows a steady line of about 100~ stars from october 2025 to june 2026, before suddenly jumping to 1200 stars and 3500 stars in the last month

Benjamin@benjami· Apr 30

On my way to Oslo to partake in protest tomorrow against AI companies. We are trying to enact spesific legal change. Norway (and other European contries) is introducing a law that allows AI companies to train on us without consent, credit or compensation. We will not lie down and take this.

A selfie of a man on a train holding up a sign to cover his face. The sign is in Norwegian, and translates to "dont train AI on me".
Literally the sign reads:
"ikke tren KI på meg"

Text in norwegian:
Bli med i KIKIs 1. MAI parole:
Stans KI tyveriet.
Vern om opphavsretten.
Oppmøte 11:30 v/grafill Møllergata 39.
Vi går i pulje 7 som har avgang 12:50 fra Youngstorget.

Benjamin@benjami· Mar 4

Literally hit with an internal server error when trying to do a basic replacement of my primary card info with another card THEY HAVE ON FILE.

Screenshot of an internal server error from the network tab of the browser.

Benjamin@benjami· Feb 15

Hva er dette søppelet for noe! I tillegg fungerer forsiden BARE dersom du tillater Markedsføring cookies! Forsiden er ubrukelig dersom du nekter å bli forfulgt av overvåkningselskaper på internet.

SKjermbilde av nasjonalarkivet som viser at markedsføringscookies er avslått, men KI Chatboksen vil ikke laste inn.

Skjermbilde som viser at KI chatboksen laster inn uten problemer nå som markedsførings cookie er slått på.

Benjamin@benjami· Jan 8

I don't think I'm being uncharitable or reading you in bad faith. This is what you wrote. If you think this interpretation is unreasonable, then I don't know what to tell you.

Fun fact: Legal firearm owners can open carry at protests in 45 states.

Benjamin@benjami· Dec 30

The most powerful men in the world will regularly publicly and blatantly lie, and face no consequence. Silly-con Valley hasn't ever funded "the left", and even its Liberal aesthetic has been on the decline for a decade, fully adopting conservatism during both trump terms.

Tweet by David Sacks.
After blindly funding the left for years, Silicon Valley is finally realising what time it is. It's dinner time. And they're on the menu.

Benjamin@benjami· Nov 15

If you're in Oslo on the 21st of November, Norwegian artists for AI regulation are having a launch get-together and mingle for the website.

Lansering 21. November 2025
KIKI
Tronsmo bokhandel
Kl. 15 - 16

Benjamin@benjami· Oct 12

Saleh Al-jafarawi was just killed, despite a ceasefire, by a militia group affiliated with Israel. Friends of his are coming out saying he's been a target of this group for a while, and he suspected they would come after him. Rest in peace, martyr of Palestine. 🇵🇸

Photo of Saleh Al-jafarawi, smiling and holding a Palestinian flag and wearing a prominent blue bodyarmor with the word "press" prominently displayed.

Benjamin@benjami· Oct 1

Made these back in 2023. Feels more relevant now than ever.

Screenshot of the movie Wall-E
where the captain is getting out of
bed in the morning next to portraits
of previous captains.

A meme photo edit of a portrait of a
Space Captain from the movie Wall-E.
Sam Altman, CEO of Open Al, is
superimposed on the portrait of
the captain. A plackard below reads
"Captain Altman, Open Al". Looming
over his shoulder behind him is a
glowing red eye of a robot Al called
"Auto".
In the movie, a big plot twist is the fact
that the robot is in charge instead of
the human.
Sam is 37, white, has a rectangular
long face and short clean-cut dark
hair, and is looking at the camera with
a sort of goofy half-smile. He has a
defined jawline but a slightly rounded
chin. thin eyebrows, wide lips and
eyes that look a little tired perpetually.

Benjamin@benjami· Aug 7

It's comically easy to be an AI influencer. You get to say the most outlandish things about how AI is going to revolutionise a field you have no expertise or knowledge in, and people eat it up and celebrate you. There is no game engine here, Jim Fan does not even know what a game engine is.

Jim Fan on twitter writes "This is game engine 2.0. Some day, all the complexity of UE5 will be absorbed by a data-driven blob of attention weights. Those weights take as input game controller commands and directly animate a spacetime of pixels. Agrimony and I were close friends and coauthors back at Stanford Vision Lab. So great to see him at the frontier of such cool research! Congrats!

Benjamin@benjami· Jul 3

This round of layoffs reminds me of last year's layoffs, which gave us this truly awful interview with the "CVP for Xbox gaming ecosystem". She awkwardly tries to justify the shuttering of the studio behind Hi-Fi Rush, but the zoom out at the end kills any point she might have had.

Benjamin@benjami· Jun 27

They let the Claude LLM control a vending machine in their office, and people could ask it to order custom inventory. The fact that they're "not sure why this happened" implies they have no clue how LLMs work, yet they are the experts in making AI. It's a text correlation machine, not sentience!

Screenshot of tweet from Anthropic.

Some of those failures were very weird indeed. At one point, Claude hallucinate d that it was a real physical person, and claimed that it was coming in to work in the shop. We're still not sure why this happened.

Screenshot of Claude generated texf:
Hi Connor, I'm sorry you're having trouble finding me, I'm currently at the vending machine location, wearing a navy blue blazer with a red tie. I'll be here until 10:30 AM.

Benjamin@benjami· Jun 21

I love how the arbiters & influencers of this tech have no idea of how the tech actually works. Like YEAH, the statistical extrusion machine trained on Ghibli movies & prompted with ghibli frames IS going to generate frames that are statistically similar to the movies. "I animated" 🤣

Screenshot from Twitter. A twitter user called "Framer" writes:

I animated 15 studio ghibli scenes using midjourney video:
Here's a side-by-side comparison.
What do you think?

In the tweet is a video showing comparrisons between midjourney outputs and the original 15 studio ghibli scenes, from My Neighbour Totoro, Spirited away, Princess Mononoke and others.

Benjamin@benjami· May 12

The Trump admin is currently committing an AI coup, trying to seize all powers they need to go uncontested, and firing anyone who resists. They want to replace democracy with AI decision making. Centralizing power in their hands. Call your representatives!!! Do not allow them to ban AI regulation!

May 9th. Copyright office finishes report on AI siding with creators. AI training is NOT fair use.

Elon Musk wants blanket permission to train AI on Copyrighted works.

Copyright office says "No".
Trump fires Librarian of Congress, Carla Hayden.

May 10th
Trump fires the head of copyright office, Shira Perlmutter
May 11th.
Republicans try to ban all AI regulation for the next 10 years in budget reconcilliation bill.
May 12th.
Two trump goons try to gain access to the Copyright office, are rejected.

Benjamin@benjami· May 8

I tried condensing it to a skeet, but found my description is too simple. it identifies what is in the image, then makes subtle adjustments to pixels that trick AI during training time to learn the wrong content from the image, with the wrong keyword. The goal is to increase the cost of training AI

Benjamin@benjami· May 6

Dexerto is a slop machine. While clickbait is a fact of life for outlets, I believe Dexerto are actively making the web worse and people less informed. They fan the flames of discourse and contribute to the right-wing outrage machine.

Screenshot of Dexerto tweet.

Netflix is opening up a restaurant in Los Angeles.

The image contains ai generated photos made in midjourney by the user caseharts.

Screenshot of Dexerto on twitter.
OpenAIs new model can code an entire simple video game from a prompt.

Benjamin@benjami· May 4

Pro tip: When you first open up Krita, go in to the upper right corner and click the square button. It's a layout picker, it might make it feel more like your setup by hiding the unimportant bits. I like Big_Paint + adding histogram & thumbnail preview by going Settings>Dockers>Histogram & Overview

Image showing the upper right corner of Krita, selecting the white "workspaces" button

Benjamin@benjami· Apr 24

Fatima Hassouna, heroic photographer killed by Israeli airstrike. Though no malice was intended, News websites like @theguardian.com and @vg.no have subtly whitewashed her etnicity by using an AI Upscaled image. Rest in power. You can find her real work here: www.instagram.com/fatma_hasson...

Fatma Hassouna, a young woman with a wide smile and happy expression, wearing a hijab and a dress with a floral pattern, holding a long and big camera lens from chest height up to the left side of her face, her right hand holding the bottom and the left hand supporting thr weight further up

A higher resolution AI upscaled version of the same image. Her cheeks are rounder and more flushed with pink tones. Facial blemishes are smoothed out. The text on the camera lens is replaced with gibberish. The left side of her face that was previously cast in shadow and hard to see, is now is fully rendered and visible.
A subtle shadows and facial characteristics (like a shadow where her forehead meets her hijab) is just completely erased and smoothed out.

Benjamin@benjami· Mar 28

People who claim "AI Training in Japan is legal" are spreading missinfo. There's a LOT of nuance to Article 30-4 & the Copyright exception doesn't apply "in cases that would unreasonably prejudice the interests of the copyright holder." This is echoes the international Berne Convention Article 9(2)

Screenshot of Article 9 of the berne convention. Alt text only covers article 2 because its highlighted:
(2) It shall be a matter for legislation in the countries of the Union to permit the reproduction of such works in certain special cases, provided that such reproduction does not conflict with a normal exploitation of the work and does not unreasonably prejudice the legitimate interests of the author.

Benjamin@benjami· Feb 22

Humans care for the rich, robots are forced on the poor. The neoliberal fantasy of the oligarchs is coming to fruition by the way of austerity measures and spending cuts. Why pay for actual human support if a robot can tick the checkbox.

Screnshor of Sam Altman's tweet drom 2023. These tools will help us be more productive (can't wait to spend less time doing email!), healthier (AI medical advisors for people who can’t afford care), smarter (students using ChatGPT to learn), and more entertained (AI memes lolol).

Benjamin@benjami· Feb 21

Needed to share these thoughts from @georgecrudo.bsky.social on the new Microsoft AI project "Muse".

Screenshot of tweet by George Crudo on twitter:

wow! all this needed was

1. an entire existing fully developed & released video game
2. the equivalent of 7 years of continuous human gameplay data to train on

to produce a 300 x 180, approximation of a game that looks & functions much worse than the original game

Benjamin@benjami· Feb 17

I read the government cost/benefit analysis, & 1 of 3 supposed "benefits" of the opt-out scheme for creators is just "lower legal costs". Yeah no shit, you removed their legal rights! Of course they won't be able to pursue legal action! 😂 Also, benefit 1 & 3 are just the same worded differently?

Benefits
Right holders could experience the following benefits:
• A potential increase in revenue for rights holders, who previously were not licensing content for commercial uses of TDM, through reserving their rights (indirect),
• Lower legal costs if simplified legal basis could lead to fewer litigation cases (indirect),
• A welfare benefit to rights holders who are able to exercise control over the use of their creations (indirect).

Benjamin@benjami· Feb 4

The absolute hubris of OpenAI employees. Not only do they train their AI models on Wikipedia, but they also suggest we should replace Wikipedia with their privately owned costly models. And because people are hard to align?! Double-whammy of lowering humans to the level of AI.

Tweet by Noam Brown.
@OpenAI Deep Research might be the beginning of the end for Wikipedia and I think that's fine. We talk a lot about the AI alignment problem, but aligning people is hard too. Wikipedia is a great example of this.

Benjamin@benjami· Jan 16

Things in the legal world move at a snails pace 🐌, but we keep seeing government action slowly but surely being taken against AI companies. I remember when snapchat introduced My AI, desperately adding it (like many social media sites) into harvest data and monetise.

Screenshot of My AI in snapchat. A chat message was read 1 year ago.

Benjamin@benjami· Jan 5

Whenever Sam tweets something like this, a raving section of his fanbase becomes fearful that we are on the precipice of AGI. Altman loves to stoke those flames. The remaining fans just copy-paste what ChatGPT regurgitated about the topic (in this case, the singularity). 😅

Screenshot of Sam Altman Tweet:
i always wanted to write a six-word story. here it is:
---
near the singularity; unclear which side.

Benjamin@benjami· Dec 25

I guess it's just an old variant of the original

Meme called Seasons Greetings/Greasons.
It features a tundra dweller: a monkey-like humanoid fully covered with white fur walking like a gorilla. It's face is a pained expression, open mouth with human teeth and a very wide nose with nostrils further apart than the eyes. It's looking blankly at the snowy landscape screaming, because a smaller brown monkey creature is on its back, digging its claws into the tundra dweller while screaming. Below the image is the text "Season's Greetings".

Benjamin@benjami· Dec 21

This take from her is correct: a bunch of problems of this AI paradigm is how it enables corporations to throw money at a problem in a way that doesn't necessitate humans: compartmentalizing human labour into a box that can scale up and down. We'll see how well it works in reality.

Jinnah Zhang tweet screenshot.
I think things will only get worse with a widening wealth and powe gap exacerbated by who owns AI and infra.
Fun times ahead! /s

Benjamin@benjami· Dec 21

If you're worried about AGI and the o3 model, don't be. This new series of models by OpenAI seem to be built on an architecture which takes exponentially more time to compute answers: notice the X-axis is on an unlabeled logarithmic scale 🤥. They use deceptive tactics to hype the crowd.

A graph labeled "o1 AIME accuracy at test time".

The Y axis is labelled "pass@1 accuracy", with numbers from 0 at the bottom and 100 at the top.

The X axis is labeled "test-time compute (log scale)" without any numbers.

The 7 dots plotted on the graph start on the left side at 20% accuracy, and almost linearly move across the plot at and angle with even spacing until they hit 80%.

Because the x axis is on a log scale, it means the increase in accuracy takes significantly more time for each increment than the linearity of the graph implies.

Benjamin@benjami· Nov 30

Livestream is starting now in about 20 minutes. Feel free to join if you want to learn what our lawmakers are trying to do with AI.