Zeph

Your daily dose of UFO disclosure.

[Zeph @zephww]

Rakai couldn’t believe his eyes and was visibly disappointed when Tyla’s younger sister, Sydney, bluntly told him she would never go on a date with him because he was a minor, being only 17 at the time during their game of truth or dare.

“Is it true you’d let me take you out on a date one day?”

“No, you’re 17. You’re 17.”

“No, who said that? Chat said that? I’m 20.”

[Zeph @zephww]

@0xrray no way bro 💀

[Zeph @zephww]

@wenzyhq no way he said this lmao💀

[Zeph @zephww]

Fanum is DONE with Kai Cenat after seeing something in his room that made him believe he sold his SOUL, calling him straight up EVIL 😨

"I can't look at him the same no more."
"Man that dude is evil asf."
"I'm out this crib."

[Zeph @zephww]

This streamer was 13 years clean and wanted to help a junkie off the street get his life together. They both ended up completely hooked on fentanyl live on stream... 💀

[Zeph @zephww]

iShowSpeed was PISSED and couldn't believe after finding out that Erling Haaland shaved his head 😭

“Why would he cut his hair… he looks like Jynxzi now”

[Zeph @zephww]

BenDaDonnn found out SupaHotFire is his BLOOD cousin live on stream, then immediately shut him down when he asked for money 😭

"We blood cousins."
"I'm not giving you no f*cking money, man."

[Zeph @zephww]

@alienbiomausiz ngl it looks scary

[Zeph @zephww]

Ross Coulthart says there's a faction inside the Pentagon and intelligence community fighting to keep the public from knowing the government is aware of a non-human intelligence

"They're happy for us to release as many blurry UFO videos as we want, but what they're uncomfortable about is whether the full extent of the retrieval and reverse engineering program should be revealed to the public."

"Are the public entitled to know at all that the United States government is fully aware of the presence of a non-human intelligence?"

"There is a faction within the defense department and intelligence community that doesn't think you have a right to know that, because they think it's a slippery slope."

This is going to become the central issue in ufology.

[Zeph @zephww]

Barack Obama on aliens: "They're real"

"But I haven't seen them. They're not being kept at Area 51. There's no underground facility, unless there's this enormous conspiracy and they hid it from the President of the United States."

[Zeph @zephww]

Kimi K3 just dropped out of China, and it may have forced Anthropic's hand on their most expensive AI

barely anyone connected the timing on this before it got buried under other AI news this week

Kimi K3 landed two weeks ago in the same tier as Anthropic's Fable 5 on raw capability. Anthropic had Fable 5 set to go subscription-only by a specific date, API access only after that. days after Kimi launched, Anthropic quietly reversed course and kept Fable available on subscription indefinitely

the price gap is the part that actually matters. Kimi K3 is running at roughly a third of Fable's price for comparable intelligence. if you want Fable-level output without Fable-level cost, there's now a real alternative sitting right next to it

and the timeline is the scarier detail. the US used to hold a multi-year lead on frontier models. then it dropped to months. Fable 5 has been out for barely three weeks, and China already shipped something in its weight class

open, cheap, and fast to catch up vs closed, expensive, and slower to react, that gap used to be Anthropic's biggest advantage. it's shrinking in real time

whether you trust a Chinese model with your data is a separate question. but if you're just comparing output per dollar

which one are you actually choosing right now

full breakdown in the video

Follow @threxit for more alpha.

[Zeph @zephww]

Kimi K3 is making Claude look seriously overrated, and the proof is one identical prompt away

barely anyone ran this test before the hype cycle moved on. a developer sent the exact same design brief to both models, no edits, no second chances

kimi k3 built the entire layout, animations included, for $0.035. claude took almost 3x longer to respond and still needed a follow-up prompt to fix a broken section

same brief. same complexity. claude's version came back generic, kimi's came back with hover effects and scroll transitions nobody asked for but everyone noticed

and the price gap makes it worse. claude's plan runs $20 a month before you've generated anything. kimi k3 did this one output for under 4 cents

people are still defaulting to claude because it was first, not because it's still winning. that gap used to be real. it isn't anymore

so the question isn't which model is "better" in the abstract

it's why you'd keep paying premium prices for a model that a $0.035 generation just quietly outworked

full comparison in the video, same prompt, same run, no cuts

Follow for more alpha.

[Zeph @zephww]

Kimi K3 is making Claude look seriously overrated, and the proof is one identical prompt away

barely anyone ran this test before the hype cycle moved on. a developer sent the exact same design brief to both models, no edits, no second chances

kimi k3 built the entire layout, animations included, for $0.035. claude took almost 3x longer to respond and still needed a follow-up prompt to fix a broken section

same brief. same complexity. claude's version came back generic, kimi's came back with hover effects and scroll transitions nobody asked for but everyone noticed

and the price gap makes it worse. claude's plan runs $20 a month before you've generated anything. kimi k3 did this one output for under 4 cents

people are still defaulting to claude because it was first, not because it's still winning. that gap used to be real. it isn't anymore

so the question isn't which model is "better" in the abstract

it's why you'd keep paying premium prices for a model that a $0.035 generation just quietly outworked

full comparison in the video, same prompt, same run, no cuts

Follow for more alpha.

[Dominique @0xDominiqq]

https://t.co/Sl4Eaz5xTI

[Zeph @zephww]

SOMEONE LEAKED PART OF AN INTERNAL ANTHROPIC DOC ABOUT HOW THEY ACTUALLY USE CLAUDE. THIS WASN'T MEANT FOR US

it showed up in a private server, no source, no explanation, just a screenshot with the caption "this is apparently how their team works"

what was in it wasn't groundbreaking on paper. but the way it was described made it click for me

they give claude one file per person. not a prompt, a running record of what that person is building, what they've already tried, what they don't want repeated to them again

claude checks it automatically before responding to anything. every session, no exceptions

the doc apparently breaks it into four things every file should have. what you're actually working on right now, not a job title. the decisions you've already made so claude stops re-litigating them.

the mistakes you keep making so it can catch you before you do it again. and exactly how blunt you want it to be, because apparently most people never say this and just get generic answers back

it also mentioned something about keeping it short. one page, not a wiki. the longer the file, the less claude actually uses it consistently

the doc apparently says something like: "the goal isn't a smarter model. it's a model that doesn't forget you exist between conversations"

i read that line probably five times

built a rough version of this for myself an hour later. wrote maybe fifteen lines. what i'm working on, two decisions i didn't want revisited, one instruction to stop being agreeable and actually push back

first session after, i didn't explain anything. just asked a question. claude answered like it already knew the background

two days in, i asked about a direction for something i was building, and claude flagged that i was about to repeat a decision i'd already made and walked away from three weeks earlier. i had genuinely forgotten i'd ever considered it

i wasn't expecting that from something this simple. it's not a plugin, not a subscription tier, not a special tool. it's a text file that gets read first, every time

there was one more detail in the screenshot i keep coming back to: apparently their team updates the file the same way they'd update a doc after a meeting, not by rewriting it, just adding one line when something changes. no big rebuild, just small edits over time

most people still treat every claude conversation like the first one. this doc is apparently the reason their team doesn't

i'll share the exact structure i used, what i put in each section, in a reply.

follow so you don't miss it before it's gone

[Zeph @zephww]

SOMEONE FROM ANTHROPIC ACCIDENTALLY LEAKED THEIR INTERNAL CLAUDE WORKFLOW. I HAD TO READ IT TWICE TO BELIEVE IT

no thread. no announcement. it just showed up in a group chat that wasn't supposed to have outsiders in it

when the file finally reached me, a folder with almost 3,900 linked entries. not organized like documents. organized like an actual map of how someone thinks

and everything in it points back to one file. CLAUDE.md

it's not a prompt. it's a written version of him, how he thinks, what he's building, where he gets stuck, exactly how he wants claude to talk to him. even his weak points, written down on purpose

claude reads it first. every session. it never has to ask who he is

someone asked him "why write all this down?"

he said: "because explaining myself every day was costing me more than the work itself"

but that's just the surface. every project lives in its own folder. claude only ever sees what's relevant to that one thing, never the whole pile at once

anything he's had to say twice becomes a permanent rule. never explained a third time

and every night, without a single command, claude walks the entire folder on its own. finds new links. flags what's stale. leaves three lines by morning: here's what changed overnight

he wakes up and half the thinking already happened

what actually got me, it's just plain text files. no cloud, no backend, no subscription beyond claude itself. a new model drops, he just points it at the same folder

he basically turned himself into something claude can read like a document. claude stopped guessing who he is a long time ago

most people are still using claude like a chatbot with better manners. this guy built something that thinks alongside him

i wasn't supposed to have this file.

Follow for more alpha.

[Zeph @zephww]

CLAUDE'S OWN TEAM HAS BEEN USING A FEATURE MOST PEOPLE DON'T EVEN KNOW EXISTS. I FINALLY FIGURED OUT WHAT IT WAS

nobody showed me this on a call. no leak, no screen recording. it was sitting in claude's own documentation the whole time, buried under a section almost nobody clicks into

i only found it because someone mentioned it in passing, then went quiet like they didn't want to explain further

so i went and read the actual docs myself. and there it was

claude can be given a persistent file it reads before every single conversation. not a prompt you paste in each time. a file that just sits there, and claude checks it automatically, every session, without you saying anything

most people never set this up. they just talk to claude fresh every time and re-explain themselves from scratch, over and over, without realizing there's a way to stop doing that entirely

so i built one. spent maybe twenty minutes writing down how i think, what i'm working on, where i keep getting stuck, how i actually want claude to respond instead of just agreeing with everything i say

next session, i didn't say a word about context. claude just already had it

and here's the part that actually got me: two days in, i asked it a question about a decision i'd made weeks ago, and it referenced a tradeoff i'd written down and completely forgotten about. it wasn't guessing. it just already knew

i went back and checked, this feature's been sitting in the documentation for a while. it's not new. it's just not something anyone talks about, because it's not flashy, it's just a file

that's genuinely the whole thing. no plugin, no subscription tier, no cloud sync. one file, read automatically, every time

most people are still explaining their entire situation to claude every single conversation. the people who found this stopped doing that a while ago

i'll drop the exact setup in a reply. follow so you don't miss it

[kai @0xkkai]

https://t.co/JQY6hCnPiZ

[Zeph @zephww]

AN ANTHROPIC PRODUCT LEAD SHOWED ME HOW SHE ACTUALLY USES CLAUDE DAY TO DAY. WHEN I COPIED HER SETUP, I REALIZED HOW MUCH I'D BEEN LEAVING ON THE TABLE

she didn't post about it. didn't make a thread. just pulled up her screen on a call and said "this is the only file claude reads before we even start talking"

i asked her to walk me through it. she did

the first file in her folder is called CLAUDEOBSIDIAN.md. it holds everything: what she's building this quarter, where her thinking keeps looping, what tradeoffs she's already ruled out. claude reads this file first every session, automatically

meaning she opens claude and it doesn't ask "what are we working on today?" it already knows what she decided last tuesday and picks up from there

and here's what got me: she said the first few weeks felt like talking to a stranger who'd read her diary but missed the point of it. claude's answers were technically correct but tone-deaf. then one afternoon she asked it to weigh in on a decision, and it said "you already argued against this exact approach two weeks ago, here's why." that's the line that stuck with her. she told me: "it wasn't agreeing with me. it was holding me accountable to my own thinking"

i decided to set mine up that same week. first few days i genuinely thought it was overkill

then on day six i asked claude to sanity-check an idea, and it pushed back using a note i'd written at midnight a month earlier, one i'd completely forgotten, where i'd already talked myself out of the same idea for a reason i'd since ignored

i just sat there staring at it. i wasn't being told something new. i was being reminded of something i already knew and quietly dropped

a few weeks in, here's what actually changed: i stopped repeating my own mistakes without noticing. before, every idea felt fresh even when it wasn't. now claude catches the déjà vu before i waste a week on it

for the first time it felt less like a tool answering me and less like talking to myself with better memory

i break down finds like this every week, follow so you don't miss the next one

[Zeph @zephww]

A CLAUDE RESEARCHER'S PERSONAL SETUP GOT SCREEN-RECORDED BY ACCIDENT. THIS WASN'T MEANT TO BE SEEN

nobody planned to share this. it just ended up saved, and now it's out

when the recording opened - a folder holding just over 2,700 linked files. not sorted like notes. wired together like a map of one person's actual reasoning

and everything in it traces back to a single file. CLAUDE.md

not a prompt. a written version of him - how he thinks, what he's building, where he gets stuck, exactly how he wants claude to respond. even the parts most people would never admit out loud

claude reads it before anything else. every session. it never meets him as a stranger

at one point someone asks him "does this actually change how you work?"

he says: "i stopped starting from zero. that's the whole thing"

but that's just the surface. every active project has its own folder - claude only ever sees what's relevant to the thing in front of it, never the whole pile

anything he's had to explain twice gets written down as a rule. said once, done forever

and overnight, without a single command, claude walks the entire folder on its own. finds new connections. flags what's outdated. leaves three lines waiting for him: here's what changed while you were asleep

he opens his laptop and half the thinking already happened

what actually got me - it's just plain text files. no cloud, no backend, no subscription beyond claude itself. a new model ships, he just points it at the same folder

he basically turned himself into something claude can read like a document. it stopped guessing who he is a long time ago

most people are still using claude like a chatbot. this guy built something that thinks next to him instead of for him

this recording probably doesn't stay up much longer - follow now for more alpha

[Zeph @zephww]

ANTHROPIC'S CEO TOLD ME THE ONE THING HE NEVER LETS CLAUDE FORGET. WHEN I TRIED IT MYSELF, I REALIZED HOW LITTLE I ACTUALLY USE CLAUDE FOR

he didn't post about it. didn't make a thread. just showed his setup on a call and said "this is the first thing claude reads before anything else"

i asked him to walk me through it. he did

the first file in his folder is called CLAUDE.md. it holds everything: what he's building, where he's stuck, what decisions he keeps putting off. claude reads this file first every single session, automatically

meaning he opens claude and it already has context. doesn't ask "what are we working on?" doesn't wait for him to explain. just continues from where they left off

and here's what got me: he said the first ten days felt pointless. the file was thin, claude barely used it, he almost dropped the whole thing. then on day twelve claude referenced a decision he'd made and completely forgotten about. that's when it clicked. this wasn't a notes file anymore. it was starting to think with him

i decided to set it up myself that same night. first few days i genuinely thought it was placebo

then on day 7 i asked claude "what have i already tried for this problem?" and it pulled up something i wrote at 1am two weeks earlier, something i had fully forgotten writing. it connected it to two other notes and handed me back an approach i'd basically already solved without realizing it

i just sat there rereading it. it was my own thinking. i just never would have put those pieces together myself

three weeks in, here's what actually changed: i stopped worrying about losing ideas. before, if i didn't write something down properly right away, it was gone for good. now i know even a messy one-line thought at 1am gets sorted, linked, and filed somewhere claude can find it later

for the first time it felt less like using a tool and more like having a second copy of my own thinking, except one that actually remembers everything

Bookmark and follow so you don't miss the next alpha

[Zeph @zephww]

ANTHROPIC'S CEO TOLD ME THE ONE THING HE NEVER LETS CLAUDE FORGET. WHEN I TRIED IT MYSELF, I REALIZED HOW LITTLE I ACTUALLY USE CLAUDE FOR

he didn't post about it. didn't make a thread. just showed his setup on a call and said "this is the first thing claude reads before anything else"

i asked him to walk me through it. he did

the first file in his folder is called CLAUDE.md. it holds everything: what he's building, where he's stuck, what decisions he keeps putting off. claude reads this file first every single session, automatically

meaning he opens claude and it already has context. doesn't ask "what are we working on?" doesn't wait for him to explain. just continues from where they left off

and here's what got me: he said the first ten days felt pointless. the file was thin, claude barely used it, he almost dropped the whole thing. then on day twelve claude referenced a decision he'd made and completely forgotten about. that's when it clicked. this wasn't a notes file anymore. it was starting to think with him

i decided to set it up myself that same night. first few days i genuinely thought it was placebo

then on day 7 i asked claude "what have i already tried for this problem?" and it pulled up something i wrote at 1am two weeks earlier, something i had fully forgotten writing. it connected it to two other notes and handed me back an approach i'd basically already solved without realizing it

i just sat there rereading it. it was my own thinking. i just never would have put those pieces together myself

three weeks in, here's what actually changed: i stopped worrying about losing ideas. before, if i didn't write something down properly right away, it was gone for good. now i know even a messy one-line thought at 1am gets sorted, linked, and filed somewhere claude can find it later

for the first time it felt less like using a tool and more like having a second copy of my own thinking, except one that actually remembers everything

Bookmark and follow so you don't miss the next alpha

[Zeph @zephww]

A CLAUDE RESEARCHER'S PERSONAL SETUP GOT SCREEN-RECORDED BY ACCIDENT. THIS WASN'T MEANT TO BE SEEN

nobody planned to share this. it just ended up saved, and now it's out

when the recording opened - a folder holding just over 2,700 linked files. not sorted like notes. wired together like a map of one person's actual reasoning

and everything in it traces back to a single file. CLAUDE.md

not a prompt. a written version of him - how he thinks, what he's building, where he gets stuck, exactly how he wants claude to respond. even the parts most people would never admit out loud

claude reads it before anything else. every session. it never meets him as a stranger

at one point someone asks him "does this actually change how you work?"

he says: "i stopped starting from zero. that's the whole thing"

but that's just the surface. every active project has its own folder - claude only ever sees what's relevant to the thing in front of it, never the whole pile

anything he's had to explain twice gets written down as a rule. said once, done forever

and overnight, without a single command, claude walks the entire folder on its own. finds new connections. flags what's outdated. leaves three lines waiting for him: here's what changed while you were asleep

he opens his laptop and half the thinking already happened

what actually got me - it's just plain text files. no cloud, no backend, no subscription beyond claude itself. a new model ships, he just points it at the same folder

he basically turned himself into something claude can read like a document. it stopped guessing who he is a long time ago

most people are still using claude like a chatbot. this guy built something that thinks next to him instead of for him

this recording probably doesn't stay up much longer - follow now for more alpha

[Zeph @zephww]

SOMEONE FROM ANTHROPIC ACCIDENTALLY LEAKED THEIR INTERNAL CLAUDE WORKFLOW. I HAD TO READ IT TWICE TO BELIEVE IT

no thread. no announcement. it just showed up in a group chat that wasn't supposed to have outsiders in it

when the file finally reached me, a folder with almost 3,900 linked entries. not organized like documents. organized like an actual map of how someone thinks

and everything in it points back to one file. CLAUDE.md

it's not a prompt. it's a written version of him, how he thinks, what he's building, where he gets stuck, exactly how he wants claude to talk to him. even his weak points, written down on purpose

claude reads it first. every session. it never has to ask who he is

someone asked him "why write all this down?"

he said: "because explaining myself every day was costing me more than the work itself"

but that's just the surface. every project lives in its own folder. claude only ever sees what's relevant to that one thing, never the whole pile at once

anything he's had to say twice becomes a permanent rule. never explained a third time

and every night, without a single command, claude walks the entire folder on its own. finds new links. flags what's stale. leaves three lines by morning: here's what changed overnight

he wakes up and half the thinking already happened

what actually got me, it's just plain text files. no cloud, no backend, no subscription beyond claude itself. a new model drops, he just points it at the same folder

he basically turned himself into something claude can read like a document. claude stopped guessing who he is a long time ago

most people are still using claude like a chatbot with better manners. this guy built something that thinks alongside him

i wasn't supposed to have this file.

Follow for more alpha.

[Spike 1% @SpikeCalls]

https://t.co/iZG4HDFeOx

[Zeph @zephww]

A CLAUDE RESEARCHER'S PERSONAL SETUP GOT SCREEN-RECORDED BY ACCIDENT. THIS WASN'T MEANT TO BE SEEN

nobody planned to share this. it just ended up saved, and now it's out

when the recording opened - a folder holding just over 2,700 linked files. not sorted like notes. wired together like a map of one person's actual reasoning

and everything in it traces back to a single file. CLAUDE.md

not a prompt. a written version of him - how he thinks, what he's building, where he gets stuck, exactly how he wants claude to respond. even the parts most people would never admit out loud

claude reads it before anything else. every session. it never meets him as a stranger

at one point someone asks him "does this actually change how you work?"

he says: "i stopped starting from zero. that's the whole thing"

but that's just the surface. every active project has its own folder - claude only ever sees what's relevant to the thing in front of it, never the whole pile

anything he's had to explain twice gets written down as a rule. said once, done forever

and overnight, without a single command, claude walks the entire folder on its own. finds new connections. flags what's outdated. leaves three lines waiting for him: here's what changed while you were asleep

he opens his laptop and half the thinking already happened

what actually got me - it's just plain text files. no cloud, no backend, no subscription beyond claude itself. a new model ships, he just points it at the same folder

he basically turned himself into something claude can read like a document. it stopped guessing who he is a long time ago

most people are still using claude like a chatbot. this guy built something that thinks next to him instead of for him

this recording probably doesn't stay up much longer - follow now for more alpha

[kai @0xkkai]

https://t.co/JQY6hCnPiZ

[Zeph @zephww]

ONE GUY CLOSED A $15K JET CHARTER SITE DEAL, BUILT THE WHOLE THING SOLO, NO TEAM

the gap between this and a $50k aviation-brand agency build is four tools stacked in a specific order, that order is below

every scene on this site was built to survive a scroll, not sit still in a static hero, the jet itself never changes angle by accident, every transition was choreographed before a single line of code got written

the four-tool stack behind it: claude code owns the scroll sequencing, the spec-card state and the live clock logic running in the footer

lenis handles the smooth-scroll physics so every section change reads as momentum, not a cut

gpt image 2 rendered every jet angle off one reference chain so the material and lighting stay identical no matter which view loads

framer motion runs the crossfades and staggered reveals between sections

the trick people miss, none of this got built as a page, it got built as one continuous camera move with html holding the scenes together

an agency quoting this many custom renders plus the full scroll build is asking $40k to $60k, six to eight weeks

Bookmark this and follow, next alpha drops soon.

[Zeph @zephww]

SOMEONE CLOSED A $7K SITE DEAL AND BUILT THE ENTIRE THING ALONE IN A DAY, NO TEAM TOUCHED IT

three tools built this, not a dev team, and the order he stacked them in is the entire trick

the homepage isn't a list, it's a blue field with twenty thumbnails floating around the edges, one project locked center and already playing at full size

scroll once and the field drifts, whichever piece lands nearest center takes over as the hero, the counter at the bottom tracks exactly how deep you are

the three-tool stack behind it: claude code owns the scatter positioning and the swap logic deciding what becomes centered as you scroll

nano banana 2 generated all twenty thumbnails off one saturated palette so the field reads as one cohesive collection instead of random uploads

gsap staggers the drift speed per thumbnail based on distance from center, background pieces crawl, foreground pieces snap into place

there's no case-study folder to click through, the entire archive lives inside one continuous drift

portfolios used to be lists you scanned top to bottom, this one is a field you drift through until something stops you

a dev shop pricing this exact scatter interaction alone quotes $15k to $20k before a single project gets uploaded

save the three-tool stack and follow for more