everyone is busy removing em dashes from their AI posts. a benchmark published this week says that's not what makes them bad
they compared AI-written articles with human ones. the AI text was unclear 54% of the time. the human text: 10%
the 3 things AI does, shown with their own example, a duck asking for grapes at a lemonade stand:
1. it hides who does what
AI: "a request for grapes was made upon the duck's arrival at the lemonade stand"
human: "a duck walked up to a lemonade stand and asked, 'got any grapes?'"
2. it glues nouns together into things nobody says
AI: "the duck made a grape availability request as he arrived at the beverage retailer"
3. it says things that mean nothing, and the point gets lost
AI: "nothing else depends on anything but the result, so i'll report as soon as the poll returns"
human: "i'll report the result when the poll finishes"
so before you post an AI draft, check 3 things:
- is the subject of each sentence the one doing the action?
- would you say these words out loud?
- can you find the point in one read?
a draft with no em dash can still fail all three
teapawt.org/zhizdev/status/2119874…
Why is AI slop so frustrating to read?
AI writing often obscures main ideas and invents awkward phrases, making sentences more complex than necessary.
Many of these slop patterns violate writing principles established in English writing books.
In our benchmark, AI writing breaks these principles far more often than human writing, with Claude Code replies by far the worst offenders.
We show examples of the most common slop patterns in this thread
30
this list comes from a free claude skill on github called "declauding"
you give it a draft, it looks for these patterns (it knows more than 50) and rewrites them as plain sentences
its rule for fixing a sentence: say who does what, then stop
teapawt.org/MoritzSchauer/status/2…
40
in 1911 hiram bingham had to travel to peru and climb a mountain to find machu picchu
last month a guy on holiday in peru asked his AI agents what bingham would do today. 48 of them went through NASA laser scans of 1,200 km² of mexican jungle, one image at a time
they found a hilltop with a plaza and a pyramid at least 8 meters tall, hidden under the trees. mexico's archaeology institute replied that no site is registered there
from the idea to that reply: 3 days
teapawt.org/PietroMiotti/status/21…
My AI agents found an unmapped Maya site in Mexico, using LiDAR data from a NASA dataset.
72 hours from a random thought on vacation to a reply from Mexico's national archaeology institute.
What a time to be alive…
24
every shop owner had to learn to claim their google listing. now there's a second one to claim
since yesterday a business can claim its page in chatgpt, update its info and prices, and see how many times chatgpt mentioned it and how many people clicked
1.2 billion people use chatgpt every week. most businesses have no idea what it says about them
teapawt.org/OpenAI/status/21202165…
Now in beta: Businesses in ChatGPT.
Claim your business, manage your information, and get insights on how people engage with you.
Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.
openai.com/index/introducing…
1
13
in 2024 a mathematician got 5 years of funding to do this by hand. AI just did it in 30 hours teapawt.org/claudeai/status/211986…
Replying to @claudeai
A team of 1,000 Fable 5.2 agents fully formalized Fermat’s Last Theorem in 30 hours, translating the proof into a language a computer can check step by step.
With the same setup, Opus 5.5 completed only 26% in 18 hours, and barely made any more progress after 30 hours.
13
a year ago AI couldn't lay out a circuit board. this week it designed three working gadgets: a speakerphone, a voice assistant hub and a pocket virtual pet
it wrote the specs, picked the parts, drew the boards, designed the cases for a 3D printer and wrote the software
the factory built all three from its files without changing anything. they work teapawt.org/claudeai/status/211989…
During pre-release testing, Claude Fable 5.2 and Claude Mythos 5.2 worked on ambitious projects in many different fields. Here are a few.
From a one-paragraph prompt, Fable 5.2 designed three working electronic devices, including this home voice assistant hub.
18
imports of AI hardware into america this year: up 69%
imports of everything else: down 10% teapawt.org/TheEconomist/status/21…
American imports of AI-related goods are on course to exceed an eyewatering $1trn this year. We explain how the protectionist-in-chief’s embrace of the technology has complicated his ambition to shrink his country’s trade deficit economist.com/finance-and-ec…
10
people don't realize an AI is now looking at about 92 million fish, one by one, all day
14,000 cameras in the pens. when the AI sees the same louse in two frames, it fires a laser at it. the fish's scales reflect the beam, the louse dies
nobody checks each shot. it just runs, 24/7 teapawt.org/itsolelehmann/status/2…
Norway now has more than 2,000 underwater AI laser robots shooting lice off its salmon
they're made by a company called Stingray, and they already guard more than 50 million salmon and trout 24/7
the reason is sea lice, tiny parasites that latch onto salmon and feed on them
as few as 11 can kill a small fish, and one pen can hold 150,000 salmon
the usual fix is pumping the fish out and blasting them with warm water or water jets to knock the lice off
but the handling is so rough that Norway's fish health report ranks it the top cause of death in full-grown farmed salmon
so instead, Stingray puts a robot inside the pen that hunts lice one at a time, all day:
> two cameras watch every salmon that swims past
> AI checks two frames to make sure it's really a louse
> then a green laser fires from up to 2 meters away
> the salmon's scales reflect the beam, so the fish swims off fine and the louse is dead
we now live in a world where AI lasers do pest control on your dinner lmao
12
3 different accounts posted this exact text in 4 days, word for word
45,405 views and 478 comments between them
a price people hate, a cheap fix, then "comment workflow". it's a template, and it works teapawt.org/Alexneyef/status/21187…
paying influencers $1k for 200 views? look at this.
$2 for a 15-sec video.
a yapping AI influencer I made with Seedance 2.5 + GPT-2.5, using just 2 reference images.
comment "workflow" and I'll send it over.
7
an AI monk on instagram got 131 million views in 10 months
at youtube shorts rates, that many views pays about $4,000 to $9,000
the page isn't built for that. one link in the bio, one product: beetroot capsules, $49.96 for 2 bottles
1 sale per 10,000 views would be about $650,000
the views get the attention. the capsules get the money teapawt.org/ErnestoSOFTWARE/status…
Holy sh*t.. this monk is an everlasting goldmine 😭
its been a year since the first AI monk appeared on instagram
and this guy is still averaging 799K views per video
131M views in 10 months
1.3M followers
heres how you can make your own
AI influencer with @elevenlabs:🧵
6
a faceless AI channel now has to show a face to get paid
one creator had 3 avatar channels refused monetization this past week. every channel before that got approved
the fix so far: an appeal video, on camera, showing how you write the scripts and the prompts yourself
the channel can be AI. someone still has to show up once teapawt.org/SappyYT/status/2119413…
YouTube is starting to reject monetization for some avatar channels‼️
Over the past week, I've had 3 channels not be approved for monetization. Prior to this, all of my avatar channels were accepted.
With that being said, if you create an appeal video with your face, explaining step by step how the videos are created, there is a good chance it gets approved. They want to know it's not fully automated, so explain how you create the scripts, review the scripts, and how the voiceover is AI generated using the voice of a friend, family member, etc., that you have permission to use.
Explain how you write the prompts for each image generation/video generation scene.
If you do this and show proof, you will get monetized. At least as of now.
8
britain's biggest banks just locked AI agents out of your account. one of the reasons: AI could move up to £700bn out of low-paying savings accounts teapawt.org/IntCyberDigest/status/…
‼️ Five of Britain's biggest banks, from HSBC to Barclays, fear that AI agents will go rogue, so they are locking AI agents out of customer accounts.
Some have banned agents from making payments without the customer directly involved. It is unclear who pays if a bot mismanages someone's money.
The other fear is profit. McKinsey says AI advice could move up to £700bn, about $ 900bn, out of low-paying savings accounts.
Banks fear agents acting in unison could pull deposits at once, a bank run "at machine speed."
11
a free AI tool is now going after banks
microsoft says a crime group in cambodia started using it in october. the AI looks for holes in a bank's systems and moves through them faster than any human could
bank robbers used to need a getaway car. now they need a free download teapawt.org/IntCyberDigest/status/…
‼️ The AI hacking agent China-linked hackers attacked 2,250 Microsoft 365 tenants with last month is now aimed at banks.
Microsoft says a Cambodia-based crime group has used ARTEX since October against financial firms in Southeast Asia and some big global banks.
Its agents exploit web-app flaws such as open file uploads at machine speed, plant web shells and drop remote-access malware.
18
in 2023 south koreans bought more strollers for dogs than for babies teapawt.org/TheEconomist/status/21…
Declining birth rates could cause the world’s population to shrink later this century for the first time since the Black Death. Why aren’t people having children? A new book offers some answers bit.ly/3UJX0gH
Illustration: Anna Parini
2
86
imagine explaining to someone in 1926 that you can apply to be an astronaut on a government website. starting pay $164,395, open to the public, deadline december 3 teapawt.org/NASA/status/2119136607…
Do you want to be an astronaut?
Applications are now open to join NASA's astronaut corps. Join us as we expand the frontiers of discovery in low Earth orbit, on the Moon, and beyond — apply by Dec. 3: go.nasa.gov/4iGqLqu
18
the most watched thing on american TVs is youtube. not a network, not netflix
15.8% of all TV time in august. disney, with abc, espn, hulu and everything else it owns: 9.3%. netflix: 7.7%
disney paid $71 billion for fox. most of youtube is someone with a camera teapawt.org/unusual_whales/status/…
YouTube is now the most-watched TV distributor in America, and it’s pulling away, per WSJ.
YouTube’s share of all US TV viewing time has climbed to about 14%, up from roughly 9% at the end of 2023, per Nielsen data charted by The Wall Street Journal.
That puts it ahead of every traditional media giant, including Disney, NBCUniversal and Fox, and well ahead of Netflix at about 8%.
26
> AI comes up with the idea
> robots run the experiment
> AI reads the result and picks the next one
that's a drug lab in shenzhen. hundreds of robot arms, day and night, almost no human in the room teapawt.org/TheEconomist/status/21…
Robotised labs have existed for decades, but they require humans to analyse data and refine experiments. AI models can take over the drudgery. China is watching America closely for a strategy economist.com/china/2026/11/…
14
every "AI travel agent" startup should watch this clip
airbnb gets 500,000 bookings a day, and its ceo says the traffic coming from AI agents is "very non-material"
AI can write code and prove math theorems now. booking someone's holiday is still unsolved teapawt.org/chamath/status/2118883…
“— most traffic comes between 9 to 5 when people are at work (using Airbnb as escapism)”
I believe this and other forms of escapism (X, YouTube, games, shopping) are where most of the current productivity gains in AI have gone.
1
36
a new anthropic study says AI can already do about 38% of the work behind a scientific study: going through data, analysing it, writing it up
if scientists used it for everything it's good at, research in statistics and computer science would go about twice as fast
chemistry and biology gain less, because someone still has to run the experiments in a lab teapawt.org/emollick/status/211885…
Some new data from Anthropic looking at the impact of AI on the sciences by field: "With current capabilities, if adopted widely, AI could take on 27–52% of the researcher time needed to produce a new study, with a median of 38% across fields..."
Here: anthropic.com/research/how-r…
1
3
169




































