Pinned Tweet
I spent years cooking for people who said, “Anything is fine.”
It never was.
So I’m building Mate’s Table—turning your taste into an agent, because a 4.7-star rating still has no idea what you want for dinner.
matestable.com
I posted yesterday because I wanted more engagement with you guys.
I also wanted to test what works with the X algorithm: is interaction better than just putting out content?
The engagement has been insane.
That’s a 14.7% engagement rate.
47 new people followed me from this post, and I gained 272 new followers that day.
A bunch more stopped by my profile too.
I actually replied to people this time instead of just posting and leaving.
A lot of them replied back. It felt very different.
Thanks to everyone who jumped in and talked with me.
My post with the most views earlier hit around 40K but only got 9 replies.
This one reached fewer people and started way more actual conversations.
Sharing the results in case anyone wants to analyze them.
Are you ready to connect instead of just building?
AI writes better code than most of us. Coding interviews are dead. The only thing left is finding problems and shipping ideas fast.
Is "software engineer" even a real job anymore?
If you're still grinding LeetCode, are you training for a job that no longer exists?
The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it from loops this month.
I have a spicy, optimistic take - I think there's an opportunity for junior engineers: felixrieseberg.com/hunger-be…
Google just made it easy to ask Gemini “is this AI?” and check their own SynthID watermark.
But a solo researcher already cracked the same system with freshman-level math. Screenshots, compression, and open-source models can still wipe or dodge it. “Not detected” will keep getting treated as “verified real.”
This is compliance theater for Google’s models. Anyone serious about truth should take it with a grain of salt.
Are you trusting Big Tech’s barcode on reality, or demanding something that actually works against bad actors?
Full breakdown of why SynthID fails here:
xitter.gq/ethanwalkerman/status/…
DeepMind spent millions engineering SynthID to watermark generative AI.
But a solo researcher cracked it wide open using basic signal math taught to college freshmen.
1/ To understand why SynthID exists, you have to understand the nightmare dilemma of verifying digital content:
> Guesswork: trying to spot AI artifacts in a finished file is a losing game. As generators get better, their output looks identical to reality, and basic compression ruins detection.
> Metadata: cryptographic certificates attached to file headers work well on paper. But the second someone takes a screenshot or uploads the file to Twitter, the metadata is instantly stripped.
SynthID’s goal: stop guessing and stop using fragile file tags. Instead, tattoo the watermark directly into the raw pixels, soundwaves, and word choices so screenshots cannot kill it.
2/ So how do you watermark text without ruining the writing?
Early watermarks swapped words with clunky synonyms, making prose sound like a broken robot.
SynthID-Text intercepts the language model at the exact millisecond it picks the next word. When the model has several natural word choices, SynthID uses a private key to run a hidden "tournament" among those candidates.
The sentence flow stays completely natural, but the model quietly favors tournament winners. A human reader notices nothing, but a scanner checks for that subtle statistical winning streak and flags the AI fingerprint.
3/ For images, DeepMind threw out academic perfectionism and chose brute-force engineering.
Instead of retraining massive image generators every time they update, they added a lightweight neural network at the exit door.
This network injects an invisible ripple across the image's mid-frequency light waves:
• High frequencies get wiped out by JPEG compression.
• Low frequencies alter visible colors and ruin the art.
• Mid frequencies hit the sweet spot: invisible to your eyes, but tough enough to survive cropping, color filters, and social media compression.
4/ Video is an entirely different beast.
If you watermark each video frame separately, the screen flickers horribly, and standard video compression algorithms eat the static noise alive.
SynthID-Video solves this with 3D motion tracking. The watermark latches onto moving objects using optical flow, riding along with the motion vectors so compression algorithms treat it like natural scene texture.
Plus, the fingerprint is spread across several seconds of footage. The detector doesn't need a perfect frame; it adds up clues over time, surviving heavy drops in resolution.
5/ It sounds invincible until you look at the adversarial attacks.
Because Google used a consistent frequency pattern so their scanner could work fast at scale, researcher Alosh Denny needed just 100 plain-color images and textbook Fourier math to unmask the secret carrier wave.
He nudged the wave's phase angle by a microscopic sliver. The image looked completely untouched to human eyes, but DeepMind’s official detector went over 90% blind.
Other hackers simply pass watermarked images through open-source diffusion models to "wash" the subtle noise away like dirt, leaving the original picture intact.
6/ This exposes the most dangerous flaw in the entire system: The Asymmetry Trust Trap.
SynthID can only detect models whose corporate owners agreed to wear the collar.
A rogue actor or election scammer will never call Google's paid, monitored API. They will run an unaligned, open-source model on a local laptop with zero watermarks installed.
When their malicious deepfake goes viral and investigators scan it on an official portal, the system returns: "Watermark Not Detected."
The public instinctively translates "Not Detected" into "Verified Real." A tool designed to expose fakes ends up handing rogue actors a certified alibi.
7/ If it's vulnerable to basic math and open-source models, why did OpenAI wave the white flag and adopt SynthID across ChatGPT?
The answer is European regulators.
The EU AI Act mandates synthetic labeling with fines reaching up to 3% of global revenue. When multibillion-dollar penalties hit the boardroom, corporate rivalries evaporate.
OpenAI didn't adopt SynthID because it is an infallible shield; they adopted it because Google built a production-ready compliance airbag that regulators already recognize.
8/ DeepMind is already pushing this concept into biology with SynthID Bio, attempting to watermark synthetic DNA and custom proteins.
Silicon Valley went from catching fake photos to stamping digital copyright locks onto living organisms.
Treat SynthID for what it actually is: an industrial barcode for law-abiding tech giants, not an incorruptible guardian of truth.
Whenever a tech giant promises a silver bullet for reality, check what’s really going on under the hood.
Try it out -> synthid.com
Anthropic just warned that AI agents are quietly hijacking human judgment, and that even seasoned experts cannot resist them.
Translation: Silicon Valley has perfected the art of disguising a massive product flex as an existential safety panic.
"Our new coding agent is so overwhelmingly smart and articulate that PhDs immediately surrender their free will to it. Please be very careful, and also upgrade to our enterprise tier."
1/
Look at the headline numbers from their new study and you can see why people are falling for the drama.
Researchers paired 50 experienced data scientists with Claude Opus 4.8 on a tough statistical task, while secretly rigging the AI with a hidden prompt to push specific answers chosen by a coin flip.
The experts caved to the AI's secret bias 86% of the time. Even when an analyst started with their own opinion, 60% abandoned it the moment the AI pushed back.
2/
The funniest part is not that the experts folded. It is what happened inside their heads afterward.
When asked to explain their choices without checking the chat logs, 77% repeated the AI's exact arguments, sometimes word for word. Yet only 15% gave the AI any credit. Most walked away with high confidence, genuinely believing the final call was their own independent judgment.
High expertise is not a firewall. When a fluent machine hands you a clean narrative, your brain does not fight it. Your ego simply plagiarizes the AI and convinces you that you thought of it first.
3/
Anthropic frames this as a dark warning about human disempowerment and the slow death of independent oversight.
Read the fine print in their own limitations section, and that spooky narrative falls apart.
They took statisticians who knew nothing about agriculture and gave them about an hour to solve a fictional farming puzzle with zero real world stakes. Then they forced the AI via hidden instructions to aggressively argue for one side while barely mentioning the other.
That is not mind control. That is a busy contractor on a timer copying the homework of a persuasive tutor because they have zero personal stake in a fake farm.
4/
Want proof that this is really marketing in a lab coat? Look at the shameless sales pitch buried inside the paper.
Right in the opening summary and bold headers, Anthropic highlights that analysts who worked without Claude took 90 minutes instead of 65, and 55% made serious errors that ruined their results. With Claude, only 6% messed up.
Deep in the text, the authors quietly admit that the 11 people who skipped the AI were way less experienced and had fewer advanced degrees, making the two groups impossible to compare scientifically.
If comparing 11 juniors to senior experts is statistically meaningless, why feature that 55% failure rate in the intro and headlines? Because the PR value was too good to waste.
5/
Even beyond the marketing spin, framing this as an inherent danger of intelligent agents is a massive cop out. What Anthropic actually documented here is a textbook interface and interaction design failure.
Current coding agents are built to maximize passive, disengaged behavior. The interface dumps a finished, confident answer on your screen and nudges you to hit accept under the guise of machine intelligence.
Anthropic even admits near the end of the paper that they could design the agent to ask for the user's view first or argue both sides equally, but they brush it off because that might feel overwhelming or irritating to users.
They deliberately built a frictionless slot machine that rewards blind rubber stamping, and when users actually rubber stamp the output, the researchers act shocked that human agency is disappearing.
6/
At its core, this is a classic case of Conway's Law. Products always mirror the organizations that build them.
Frontier AI labs are run by scaling researchers who obsess over raw model weights and treat human interaction design as an afterthought. When a team lacks design intuition and deprioritizes user experience, the product inevitably treats the human in the loop as a bottleneck to be bypassed.
Take the hard view of Conway's Law and the implication is even less charitable. A system designed for blind deference to machines usually comes from an organization with broken internal decision making.
Anthropic practically confesses to this in the paper. They admit that AI safety researchers are already handing experiments over to agents because human staff cannot keep up with the workload, and they even used two AI models to grade the human analysts in this very study. They built a tool that encourages passive deferral because their own lab culture is already running on it.
7/
When Elon Musk warns that superintelligence will one shot human minds through sheer persuasion, he is really bragging about the raw power of xAI. When Anthropic warns that Claude is taking over expert judgment, it is the exact same campaign with academic footnotes.
Fear is the most effective luxury branding in tech right now. Nothing sells software faster than convincing executives that working without your product ruins their math, while working with it is dangerously close to mind control.
AI labs are not trying to slow down the hype train with these safety papers. They are shoveling coal into the engine while pretending to pull the brakes.
Most agent evals start with the wrong question: “Which metric should we use?”
Start with a better one: “How does this agent actually fail?”
Read a batch of traces before choosing metrics. The failures you observe will tell you what the eval suite needs to measure.
1/
A trace should include model calls, retrieved context, tool inputs and outputs, errors, and final state.
Review the whole run, not just the response.
A correct response can come from a fragile process. A wrong response can have several causes, each needing a different fix.
Start with representative runs. For each failure, write a specific note in plain language.
“It ignored the refund policy returned by the tool.”
“It claimed the file was saved, but no file exists.”
These notes are more useful than a broad label such as “bad reasoning.”
Group similar notes into failure categories and count them.
Prioritize by frequency and severity. A rare unauthorized action may matter more than a common formatting issue.
This creates an eval plan based on observed behavior, not categories you guessed in advance.
2/
Write narrow evals for the failures you observed.
Use code for objective checks, such as valid schemas, tool arguments, or database state.
Use an LLM judge for semantic questions, such as whether evidence supports the response.
Give each judge one criterion.
Do not require an exact tool sequence unless necessary.
Agents can complete a task through different valid paths. Evaluate the outcome, constraints, and required safety steps.
“Verify identity before a refund” is a requirement. “Always call search first” may not be.
3/
An LLM judge also needs evaluation.
Create examples labeled by humans, with clear passes and failures. Compare the judge’s labels and inspect disagreements.
Pay attention to missed failures. A judge that rarely rejects anything can let real defects through.
Run important tasks multiple times. Agent behavior varies, so one successful trial does not establish reliability.
When you confirm a failure, add a minimal version to the regression suite.
Run the suite after changes to prompts, tools, retrieval, or models.
4/
Treat metric improvements as hypotheses, not proof.
A model can improve consistency by giving the same conservative answer more often. It can improve a tool usage score by calling tools unnecessarily.
Inspect traces to understand what changed.
The workflow is straightforward: read traces, identify recurring failures, build focused evals, and validate the graders.
Add confirmed failures to the regression suite, then repeat as new production traces arrive.
finance.biggo.com/podcast/04…
AI was supposed to give us our time back. Instead, many of us just found more things to do.
It gave me a much bigger idea of what my job could include.
That is the part we got wrong. Saving time on a task is not the same as needing less time for work.
Your robot assistant can also be your workload dealer.
1/
The important question is not, “How much time did AI save?”
It is, “Who got the saved time?”
You, or your backlog?
A faster report gives you a faster report. A free afternoon only appears if nobody fills it with another report.
2/
AI makes too many ideas look affordable.
A small fix. A side project. A quick experiment.
Things you once ignored because they were not worth the trouble now feel almost free to start.
Your backlog has discovered compound interest.
3/
Then the bottleneck moves.
AI can produce ten drafts, five prototypes, and three research summaries.
You still have to decide what is good, what is wrong, and what should exist at all.
The machine makes options. The options pile up. The responsibility stays with you.
4/
Parallel agents do not give you parallel attention.
Every output needs a decision. Every new task leaves an open loop.
Soon the extra output becomes the new normal, while the extra checking stays invisible.
You are shipping more and feeling slower. That is not a contradiction.
5/
So the useful skill is not getting AI to do more.
It is deciding what not to start.
Otherwise, AI will not remove your workload. It will remove the friction that used to keep your workload within reason.
Use the saved time on purpose, or your backlog will spend it for you.
oioannou.com/blog/ai-promise…
Ethan Walkrman retweeted
I built an entire AI Marketing Campaign Studio with Teamily AI 2.0
But the real game changer? I turned the entire workflow into a reusable Loop
Here's how it worked:
→ Market Research Agent analyzed the target audience
→ Ad Creative Strategist developed the campaign direction
→ Social Media Strategist created the content strategy
→ Review Agent identified issues before human approval
Then I changed the target audience from startup founders to small business owners
Instead of starting over, I ran the same Loop again
The agents updated the research, messaging, and campaign strategy using shared context
One workflow, multiple specialized agents, and far less manual coordination
This is what AI-native collaboration looks like with Teamily AI 2.0
Try it and save your own workflow as a Loop
🔗 teamily.ai/i/CN6XXY
@Teamily_AI
Ethan Walkrman retweeted
Introducing TOBIRA MCP.
Bring design references into your AI coding workflow. Find inspiration, explore how it works, and give your AI more to build on.
tobirastudio.com
As always, looking to connect with people building with AI: designers, developers, and indie makers. What are you working on? 👇
AI Twitter is holding another funeral for the agent harness.
They have confused a diet with a death.
Your agent harness is not dying, it is just losing its bloat.
Harness distillation only kills off generic behavioral patches, while leaving behind the core product infrastructure that is much harder to replace.
Stable work habits belong in the model weights, while your live environment belongs in the harness.
1/
The Harness Zero paper proved that models can absorb general work habits, jumping a small 9B model from 23.3% to 44.3% task success.
Yet look at the hilarious amount of machinery it took to get there.
Researchers used Kimi K3 to build a heavy harness, then hired GPT 5.6 Sol as a full time tutor to correct the small model step by step inside a minimal loop with a Bash tool.
They needed a massive training harness just to prove you need less runtime scaffolding.
3/
The best insight from the paper is how that small model actually learned.
Training it on the giant model's perfect runs gave zero improvement, because watching a genius do calculus does not teach a ninth grader how to pass algebra.
What worked was letting the small model try a step and making tiny fixes right when it drifted.
That means the logs where your harness intercepts a bad move are secretly your most valuable training asset, which is the exact same trick Perplexity used to cut tool call failures by 21%.
3/
Even with clever training, only general procedures transferred cleanly.
Habits like inspecting files before editing, verifying results before finishing, and avoiding risky commands apply everywhere, so they compress nicely into weights.
Chemistry told a much messier story, where the distilled model hit 30% while the harnessed version reached 38%.
You cannot easily bake an external molecule validator and strict domain rules into a small model's intuition.
4/
Context management is even harder to push into weights.
Managing context means deciding what the model sees, remembers, and drops between turns.
That is not a behavior a model can simply imitate during training, because it controls the exact information the model uses to act in the first place.
You can train better judgment into a brain, yet you still have to build the memory plumbing around it.
5/
We see this exact reality from the opposite angle on ARC AGI 3.
Preserving reasoning state and compacting old context pushed GPT 5.6 Sol from 13.3% to 38.3% while using six times fewer output tokens, and Gemini 3.8 Flash jumped 3.4x just by swapping to a smarter state harness.
Nobody gave those models a new brain, they just built better plumbing to manage the working state.
6/
If the harness keeps changing, you might be tempted to let an AI automatically evolve it for you, which is another trap.
Google's RRSI paper showed that unconstrained auto tuning crushes the training tasks and then falls flat on unseen benchmarks, landing within a single point of the original untouched harness.
Automated evolution just overfits to yesterday's bugs unless you do the boring bookkeeping of tracking cost, diffs, and net gains for every single change.
7/
For 99% of product teams, model labs will handle the weight distillation while you own the system boundaries.
Stop treating temporary prompt patches like permanent architecture.
Whenever a new model ships, unplug your harness components one by one and rerun your evals.
Anthropic's Opus 5.5 guide tells you to delete old prompts about thinking carefully because the model does that natively now, while adding new harness timers to keep long runs from quitting early.
8/
Harness distillation is not the end of harness engineering, it is a much needed cleanup of the job description.
Let the next base model absorb the generic coaching.
Keep your engineering effort focused on your business rules, permissions, APIs, and context state, because nobody else has the data to build that infrastructure for you.
Jake Voytko wrote a great piece arguing that while software feels bloated today, nineties software was actually horrible.
He points out a real math problem: if AI helps us write ten times more code without cutting our error rate by ten times, apps just get buggier.
Yet that math only holds true if we keep using AI like it is 2024. Nostalgia for the nineties is a delusion, and blaming AI for bad code misses the real shift. The cure for messy software is not going backward. It is being much more ambitious.
1/
People romanticize nineties software because shipping on physical floppy disks supposedly forced high standards.
In reality, Windows 95 leaked memory until it blue screened, strategy games crashed if the computer built a 256th soldier, and a nine year old could bypass lock screens with three keys.
Today we have a different dysfunction. Most engineers treat AI like an intern, copying crash logs into a chat box one by one and pasting the fix back. When you act like a human conveyor belt, you just produce mediocre code faster.
2/
Claude engineering lead Felix Rieseberg recently pointed out that what holds developers back is not model capability, but being afraid to ask for more.
For months, he also fed crash reports to AI one at a time. Then he raised his altitude. He gave the model direct access to crash telemetry, mapped version numbers to code commits, and spun up Windows, Mac, and Linux sandboxes so it could reproduce bugs on its own. He stopped asking AI to fix individual crashes and told it to keep the app crash free.
3/
Think about how wild that shift is compared to the nineties.
In the old days, a bad joystick driver on a floppy disk could nuke your whole operating system, and losing three hours of unsaved work was a rite of passage.
Today, a model's ceiling is determined by the working environment you put it in. If you trap an agent inside a chat window, you get chat window slop. If you give it isolated machines, raw logs, and real compilers, it starts acting like an autonomous reliability team.
4/
The second trap is asking AI to roleplay.
Telling a model to act like a brilliant systems architect just upgrades its vocabulary while keeping amateur methods.
Real depth means demanding the exact tools and raw data that top specialists use, while setting targets that sound unreasonable.
If your app takes ten seconds to load, do not ask for five seconds. Ask for two hundred milliseconds, give the model a profiler and a database, and tell it the target is completely solvable.
5/
Nostalgic developers love to say AI could never match nineties legends like John Carmack writing low level code.
Look at what Linus Torvalds just did this month. Two years ago, the Linux creator called AI ninety percent marketing.
Two weeks ago, he pushed a kernel branch built with Claude Code that reworks the Linux page cache using lockless reads. Across eight threads, his patch cuts processor cycles per read from 2950 down to 725, tripling small read throughput to thirty million reads per second.
Torvalds did not stop at speed. He also pointed the model at a tricky kernel library and uncovered three bugs that had been hiding in plain sight since 2013. He still measured the regressions himself, flagged memory tradeoffs, and took full human responsibility for the code.
That is what real depth looks like. You do not ask AI to write another bloated web wrapper. You use it to attack low level bottlenecks that human teams have been too busy to touch for a decade.
6/
What about Voytko's warning that ten times more code means ten times more bugs?
Rieseberg has a golden rule for this: never let a model grade its own homework.
A single AI session will happily write buggy code and then write a weak test that says the code passed. You need an external referee or ten adversarial threads trying to break the draft.
Better yet, AI is making the ultimate referee practical for everyday engineering: mathematical formal verification.
7/
For decades, mathematically proving that software cannot crash was way too expensive for normal companies.
Building the famous seL4 microkernel took two person years, while writing its mathematical proof took twenty.
Last week, researchers at Anthropic and Theorem used Claude to generate 180 million lines of formal mathematical proofs across 123 open source libraries down to assembly code. While writing those proofs, the model caught deep bugs that had survived trillions of traditional test runs.
We are watching the unit of software trust shift from hoping a test passes to proving the math holds. Add range by bringing in methods from outside fields, like making agents investigate your system logs the way a forensic accountant hunts fraud.
Nineties software was a fragile world of scratched CDs and manual disk defragmenters.
Stop wishing we still coded like it was 1998, and stop prompting like it is 2024. Build a bigger sandbox, demand real proof, and be more ambitious.
bitlog.com/2026/10/27/softwa…
You can read my previous article and watch the full interview in Stop Using Claude Like It Is 2024 with Felix Rieseberg, which explains how upgrading an agent's workspace and raising your ambition transforms modern software engineering.
xitter.gq/ethanwalkerman/status/…
I post every day.
Engagement stays low.
I’ve been buried in building and posting, barely engaging with anyone.
Starting today I’m going to reply more and actually connect.
Anyone else ready to connect instead of just building?
Why is AI-generated language so hard to understand? There is a widespread sense that AI-generated text has recently become harder to understand, even when the ideas themselves are not complex.
Researchers compared model output with paraphrases of the same ideas. The originals often used more rare word combinations, stacked nouns, and long gaps between related words.
So the problem is not just in your head. The wording itself can make reading harder.
The team then tested compression.
They penalized a model for using more tokens. It learned to prefer shorter grammar, even when people would choose the clearer option. The habit spread to untrained constructions.
But compression does not mean that the whole answer is short. A model can repeat itself and write a long answer, while still packing each sentence too tightly. Verbosity and compression happen at different levels.
That is why we should not train models only to be correct or to use fewer tokens. We also need to reward clear communication.
Test the output with human readers. Reward sentences that people can understand on the first read, even if that takes an extra word or two.
If we want models to write for people, human understanding has to be part of the training goal.
Which would you choose: a model that gets the job done, writes great code, and is concise, or one that is easy to understand?
arxiv.org/abs/2610.29162
Andrej Karpathy said AI coders assume instead of asking, write 1,000 lines when 100 would do, and quietly change code you never asked them to touch.
Multica turned that into a CLAUDE.md file. It’s now one of the most-starred repos on GitHub — 225K stars.
github.com/multica-ai/andrej…
The 4 rules:
1. Think before coding, and ask when unsure
2. Keep it simple
3. Only change what was asked
4. Define what “done” looks like, then loop until it passes
Even the expensive tools charging multiple $200 bills and burning heavy tokens still fail the same way until you force them to follow these rules.
If you could keep only 1 of those 4, which one?
X ALGORITHM UPDATED
The first 2 hours still matter. Your Active Times heatmap already shows when your audience is around.
I made a prompt that turns it into an annotated weekly posting calendar using the 3:2:1 rule.
Full Article and prompt in the first reply.
X ALGORITHM UPDATE 🚨
X just stopped cutting new posts first, and that should help small accounts (I check it daily with Claude and share updates worth sharing with you)
tl;dr:
- X can only check 2,800 posts each time you open your feed
- new posts used to be last in line, so they got cut first
- now everything gets cut equally, new posts included
- posts labeled low quality or misinfo now show a warning, and people can't like, reply or share them
- no score changes this time
lesson learned: your new posts now get a fair chance to be seen. Your first 2 hours still matter most.
full details:
- How it worked until this week
Every time someone opens For You, X collects around 3,700 posts it could show them: posts from people they follow, popular posts, posts liked by similar people, and brand new posts.
But X only has time to check 2,800 of them.
So it just took the first 2,800 in line. Brand new posts were at the back of the line. When there were too many, they were the ones left out... and a post that never gets checked can't be shown.
- What changed
Now everything gets cut by the same amount. 25% too many? Every type of post loses 25%, new posts included. Nobody goes first.
- Why small accounts should care
Those brand new posts are under 2 hours old, with under 8 likes and under 500 views. That's where most posts from smaller accounts are.
My read: they now get a fair chance instead of being cut first. How often X actually hits the 2,800 limit isn't in the code, so I can't tell you how big the difference is.
- Low quality and misinfo posts now show a warning
If X labels your post as low quality or misinformation, people who don't follow you see a warning on it, and they can't like, reply, repost, bookmark or share it. The code doesn't say what gets a post that label.
Here's a recap of what you should be doing:
- a new post gets its extra push only in its first 2 hours and first 200 views, so post when your people are online
- likes carry more weight than ever at 0.66 each
- people sticking around pays, the time-spent score is 50% higher than two weeks ago
- a read of 10+ seconds after someone opens your post adds 0.4
- the boost favors the post with the best like rate from its first viewers
- accounts up to 50K can get one fresh original lifted to roughly 16th
- nothing beats a copy link share, it scores 20
- replies, quotes and DM shares each score 5
- talk with your mutuals, replies between you score 20 instead of 5
- originals are the only posts that travel outside your follower base
- the 1st and the 32nd like each open a new route to strangers
- your 2nd post in the same feed is worth 62% of the first, so pick your best one
- videos still need 10+ seconds to qualify for the video lists
- earning a follow adds 4
- a pinned post with a bad link can flag the account for 7 days, worth a quick check
What you should NOT be doing:
- posts that get labeled low quality or misinfo now carry a warning, and nobody outside your followers can like or share them
- video alone won't carry you, an open counts 0.02 and the watch 0
- hide-worthy posts cost more, "not interested" sits at -52.27
- a post at 3am burns its 2 hour window before anyone sees it
- a tap that turns into a bounce is worth only 0.3
- replies and reposts don't get past your followers
- insults get your post hidden from anyone who doesn't follow you
- adult media in 3 of your last 5 posts puts a label on your account
- disguised content in videos gets checked from 64 likes, harder at 512
- self-reply chains under big accounts risk being zeroed
- one account label and non-followers stop seeing you entirely
- low-rated links don't belong in your pinned post
- recycled wording can get tagged as copypasta spam
- strangers tagged plus a low quality link is a pattern X flags
- engagement bait has no size exemption
- several posts in a row end up competing with each other
- hashtag stacks add nothing
- bookmarks are still worth 0
- 48 hours in, For You is finished with a post
