Bercus (@BercusNA) header
Bercus (@BercusNA) profile picture

Real stacks, real prompts, real invoices.

Joined October 2026
Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)Photo by Bercus (@BercusNA)
Former p*rn star Lana Rhoades says s*x work feels like "pulling teeth." So she's building an AI version of herself to do it for her. Lana Rhoades told Julia Fox in a 2024 Interview Magazine sit-down that she's been back on OnlyFans for only a year and is already making millions annually (she specified: per year, not per month). Her actual exit strategy is simple and cold: "I'm trying to make an AI Lana." The money isn't the problem. The problem is that the work feels like "pulling teeth" if you're not genuinely sexual, and almost nothing else pays for the lifestyle she's built. She was one of the first adult performers to land real brand deals and build a serious Instagram following. She's walked for Alexander Wang and Dilara Findikoglu. Traditional modeling agencies still won't touch her because of the porn, so the jobs she does get come from people sliding into her DMs. That's why the AI version makes sense to her. It keeps the income flowing without her having to keep performing. She went further and said the industry should do this for every performer. She's racing to get her fashion work and other projects solid by 30, so that when the real Lana steps away, the digital one can keep earning in her place. If an AI version of you could do your job, would you let it?
10
2
29
2,768
The first bullet passed six inches above her seven-year-old son’s head. The second one was supposed to kill him. OnlyFans model Ansley Pacheco was still half-naked in the bathroom when the screaming started. Two men had followed a Mercedes through the gate of her northwest Miami house on October 4, 2020, right in the middle of the Heat’s Finals game. They put every man in the living room face-down on the floor and started ripping gold off necks like they were stripping a carcass. One hundred and two thousand dollars. A $50,000 chain. A $33,000 watch. All of it gone in seconds. She grabbed her son, slammed the bedroom door, and yanked the pistol from the nightstand. Her hands were shaking so hard the gun almost slipped. For one sick second she saw her own father’s face — the cop who taught her that hesitation gets people killed — and then the bedroom exploded. Seven shots. Glass raining. Bullets punching through the closet door, the bathroom door, the headboard, the TV. Every single one sailing right over the kid’s head while he screamed into her chest. “DON’T SHOOT, MY SON IS IN HERE!” They kept shooting. She opened the door and put rounds back at them until the magazine was light and the hallway was empty. Her husband Daniel tore the gun from her hands and sprinted barefoot down the driveway, firing at the taillights as the car vanished. No one inside the house was hit. The robbers were never caught. The money never came back. Two weeks later she stood next to the bullet holes with her son’s blood still under her nails from where she’d clutched him too tight and said the only thing that mattered: “They were shooting right over his head. They didn’t care if he lived or died. So neither did I.” A room full of grown men on their stomachs. The only person who shot back was the mother in the bedroom. Would you have opened that door knowing the next round might kill your child?
10
1
20
2,502
P*rn star Madison Ivy says she took mushrooms at dessert, then walked straight into a private Lakers dinner. It was BOA Steakhouse in LA. She says she and her best friend had a habit: every once in a while, at dessert, they'd sprinkle a little "magical fungi" on top. That night she noticed a big table of model girls. Security kept coming out, taking two of them at a time to a back room. A big door. A guard on each side. Right as the mushrooms kicked in, she turned to her friend: "I bet they'll open the door for us." Her friend asked why on earth they'd do that. Madison said she was curious. Her friend said she'd follow her lead. So Madison walked up to the door and looked at both guards. They opened it. Inside was the entire Lakers team, from the days when Kobe Bryant was still playing. The plan was to stand in the background so nobody would notice them. Then one of the players walked right up to her: "Madison Ivy, you look just like Jessica Rabbit tonight." She won't say which player, "out of respect." But she says she almost hit the floor, because she was such a fan. "This might be the greatest moment of my whole life." Would you have walked through that door?
GTA 6 is out, so I built an adblock for it. Spoilers or the whole f*cking game, your call. Waiting for the PC version? Haven't gotten to it on console yet? Or just tired of seeing GTA 6 literally everywhere? It's called HypeBlock. Free, works on X, YouTube, Reddit and Google. Pick your mode: Spoilers → blurs posts about endings, twists and leaks. News and memes stay. Mute → hides every mention of GTA 6. It runs in Chrome and Firefox, and you can switch it on for any other site too. The code is open and everything runs locally. github.com/Coolerwave/hypebl… Unofficial fan project. Not affiliated with Rockstar Games and not an ad for GTA 6. So which one are you?
11
21
1,265
Former p*rn star Luxury Girl says a fan paid her $800 to eat a steak on camera. Then he asked her to bite a live fish. Then a mini pig. She says she still regrets that first yes every single day. It happened at the peak of her career. She was Pornhub's Amateur Model of the Year, and the request sounded like easy money. Ten minutes. No s*x. Black leather, red wine and a steak. She filmed it, smiled and got paid. Then she spent months replaying those ten minutes in her head. She says she still doesn't know what he really wanted. A few weeks later he wrote again: "Can you bite a living creature? A fish." He even sent her a screenshot of another girl doing it. She didn't answer. Then came the next message. Bite a mini pig. She says the fear hit so hard her hands went numb, and she blocked him everywhere. That night she couldn't sleep. For weeks the smell of meat made her sick. She pulled away from people, and when her partner asked what was wrong, she couldn't explain it without feeling dirty. She was making more money than ever and felt more ashamed than she had in years. So she started saying no to anything that made her skin crawl. "If I did something against my will just for the money, who was it actually good for?" She says she left at the peak. Now the only living things she touches on camera are fish she catches herself. Where would you have drawn the line?
You open your feed and it's already GTA 6. Spoilers included. Waiting for the PC version? Haven't gotten to it on console yet? Or just f*cking tired of seeing GTA 6 literally everywhere? I built HypeBlock: an adblock for GTA 6. Pick your mode: Spoilers → blurs posts about endings, twists and leaks. News and memes stay. Mute → hides every mention of GTA 6. It works on X, YouTube, Reddit, Google and any other site you turn it on for, in Chrome and Firefox. Free and open source, and everything runs locally. github.com/Coolerwave/hypebl… Unofficial fan project. Not affiliated with Rockstar Games and not an ad for GTA 6. So which one are you?
11
18
1,254
P*rn star Ava Amira was born in Baghdad and escaped the war. She says her first collab was with Johnny Sins. Before p*rn she grew up across four countries and worked as an international flight attendant. She messaged Johnny herself. The scene ran two and a half hours.
4
22
4,227
P*rn star Cory Chase served six years in the Army National Guard. Then she built a career playing stepmoms. She says many of her fans aren't after s*x. They want the approval they maybe didn't get from mom as kids. "It's just fantasy," she says.
8
33
5,429
P*rn star Little Caprice says she isn't jealous when her husband f*cks other women on camera: "It's just a role, like an actor." Married since 2015, they run a p*rn studio together. Their marriage is "completely open," she says. The one question she dodged: do they have kids?
10
37
8,065
P*rn star Vanna Bardot: "How are girls doing this for 40 minutes on camera? What the f*ck?" 5 years in, she'd never done an*l on camera. Then her 2023 movie Influence: first an*l, first DP. Both scenes won AVN Awards. She says she'd never done a DP even off camera.
3
18
2,777
1,000 CLAUDE AGENTS FINISHED FERMAT IN 30 HOURS. 99.4% OF REPOS DON'T RUN A SINGLE AGENT LOOP. Anthropic says Fable 5.2 formalized Fermat's Last Theorem in Lean with 1,000 agents. Opus 5.5 hit 26% in 18 hours on the same setup and stalled a recent paper scanned 36,487 repos and confirmed autonomous loops in 217. that's 0.59%. most of the gap sits in the setup around the model the setup, layer by layer (numbers from a Fable 5.2 harness guide): → give the agent a check it can't skip: tests, a linter or a second model that says done or not done → write down when to call the advisor. without that line it was consulted 0 times out of 198 → keep tables and logs as files. 25/25 right vs 7/25, at about 1/12 the cost → run low effort first, high effort only on failures. about 40% cheaper → let Fable plan and Sonnet do the hands-on work. a bug hunt cost $563 with Fable alone, $165 split, at a lower score → add a "finishing work" block to the prompt. it halved unfinished sessions → stop leaving stale "never retry" notes. one cut retries to about 4 in 10 before: you watch the agent and keep typing "continue" after: it checks its own work and stops when the tests pass save this before your next long agent run ↓
same 1,000 agents. Opus 5.5 stalled at 26%. Fable 5.2 finished in 30 hours. I went through every guide Anthropic shipped with it plus 4 research papers, and turned it all into one guide. 6 layers, 8 visuals, a prompt for every layer.
22
1
31
414
APPLE WANTS $129 FOR A TRACKPAD. THE PHONE IN YOUR DRAWER CAN BE ONE FOR $0. it has a touchscreen, a gyroscope and Bluetooth. since Android 9 it can pair with your computer as a normal Bluetooth mouse. no root, no server, no app on the computer I turned my old Android into a trackpad with Claude Code. here’s the $0 version: → Claude Code writes a small Android app → it uses BluetoothHidDevice, the API Google added in Android 9 → phone flat on the desk: the screen becomes a trackpad → phone in your hand: the gyroscope moves the cursor where you point → a small dead zone keeps the cursor still while your hand rests → tap = click, hold = drag → Mac or Windows pairs with it like any Bluetooth mouse paste this into Claude Code: “Build an Android app (min SDK 28) that registers as a Bluetooth HID mouse with BluetoothHidDevice. Flat on a table it works as a trackpad, held up it maps the rotation vector sensor to the pointer with a small dead zone. Tap = left click, long press = drag. Use Google’s wearmouse sample as a reference. Show me the plan first.” the catch: some phone makers block this Bluetooth mode, so test yours before you sell the trackpad $129 trackpad or $0 drawer phone. save this ↓
28
34
719
Claude just f*cked the Magic Mouse... again This is f*cking crazy. My Android phone (Solana Seeker) is now a Wii remote and a trackpad for Mac and Windows. > 2 apps: desktop (Mac menu bar / Windows tray) + Android > 0 servers > Bluetooth only, nothing goes online > Cursor updates 240 times a second > Built-in aim trainer Point the phone at the screen and the cursor lands where you aim. Tap to click, hold to drag. Lay it flat on the desk and it's a trackpad. One aim trainer mode is tennis against AI versions of Elon Musk, Sam Altman, Jensen Huang and a few other CEOs (stylized bots, not the real guys). Train your aim, roast imaginary billionaires. AirPods case, Apple Watch, iPhone, now Android. Coffee maker or toaster next?
4
98
YOUR AI HAS AMNESIA AND A CONFIDENT VOICE. THAT’S THE WORST COMBO. Stop trusting Claude’s memory. Give it a second brain that shows its sources. Most people treat Claude Code like a chat window. The ones who get it treat a folder like an atom: plain-text files orbiting one map. Claude loads the map first, then answers only from what’s written. It remembers what you decided and why, and it refuses to invent a reason you never wrote down. No apps. No plugins. No context rot. What circles the map: → notes/: one fact per file, named after the content → people/: one card per person (role, responsibilities, update cadence) → projects/: current status + next hard deadline → root: long-term memory, lessons learned, and a numbered decision log that always includes the reason → CLAUDE.md: the single source of truth that sits in the middle and is read before anything else Every answer comes back in exactly one of three ways: → Cited: it names the file it used → Not written down: it says so instead of guessing → New: it creates the file and updates the map in the same reply facts · people · projects · decisions · map The prompt to paste into Claude Code is in the reply. Save this and set it up tonight ↓
11
20
589
Make this folder my second brain: 1.Create notes/, people/, and projects/. 2. Add root files for memory.md, lessons.md, and decisions.md (numbered log; every entry must include the reason). 3. Write CLAUDE.md as the map that explains where each type of information lives. Pin memory and decisions at the top. 4. One topic per file. Filenames are lowercase with hyphens, named after the content. 5. Answer only from the file the map points to and name the file under the answer. If nothing is written down, say so. Never invent. Show me the full layout first.
2
70
YOUR CLAUDE.MD IS MAKING CLAUDE DUMBER. Anthropic’s own docs say it: keep it under 200 lines because longer files eat context and Claude follows your rules worse Claude reads the entire CLAUDE.md at the start of every session, before you type a word the fix is to turn CLAUDE.md into a short map and move everything else into files Claude opens only when it needs them: → CLAUDE.md: what lives where, plus your rules. that’s all → notes/: one topic per file, named after what’s inside → people/: one file per person → projects/: status and where you left off → decisions.md and lessons.md: what you never want to explain twice two traps people fall into: → @-importing everything doesn’t help. imported files load together with CLAUDE.md → a CLAUDE.md inside a subfolder loads only when Claude works in that folder, so folder rules go there and the rule that changes every answer: if it’s not in a file, Claude says “not written down” instead of guessing paste this into Claude Code in an empty folder: “Set up my notes here. Keep CLAUDE.md short: a map of what lives where and my rules. Create notes, people and projects folders plus decisions.md and lessons.md. One topic per file, lowercase with hyphens. Answer only from the file the map points to and name it. If it isn’t written down, say so. Show me the plan first.” save this before your CLAUDE.md hits line 201 ↓
20
26
730
DON’T BUILD YOUR STARTUP ON THE MODEL EVERYONE IS HYPING THIS WEEK. TwIL-8B just dropped. 8.79B parameters, 4.98 GB, runs on a laptop, 93% of the best frontier model on LiveCodeBench. 577K people saw the launch post. the catch sits in the model card Read this before you ship a single line on it: → the license is webAI Non-Commercial. test it all you want, but a paying customer can’t touch it → it burns tokens. on AIME it uses about 19.5k per problem. GPT-5.5 uses about 5k → the 93% coding score comes from the “verbose” variant, not the balanced one → every score is webAI’s own run, against competitor numbers from different test setups → hard science is still far off: GPQA Diamond 70.2 vs 93.6 → tool use, agent tasks and SWE-bench weren’t tested at all → no extra safety tuning beyond the base model, the card says so itself imagine building 3 months of product on it and finding the license on launch day How to use it the smart way: → prototype locally for $0 in API calls → run it on 10 of your real tasks before you trust any benchmark → need a commercial 8B: it’s built on IBM’s Granite 4.2 8B, which is Apache 2.0 small models are closing the gap fast. the license decides what you can sell model card in the reply ↓
Modest intelligence is arriving faster than most people realize. Today, we're open-sourcing TwIL-8B. Just 8.8 billion parameters, yet approaching frontier-model scores on several difficult benchmarks: • 93% of frontier performance on coding • 92% on instruction following • 90% on contest mathematics All from a model small enough to run on your own hardware. But the benchmarks aren't the most important part of this announcement. For years, the AI industry has focused on building the most capable model, regardless of the cost required to run it. I believe that equation is changing. The next era of AI will be defined by the cost of completing work. Not simply the intelligence of the model, but how efficiently that intelligence can accomplish the task. If a model with 8 billion parameters can reliably complete the same work as one with hundreds of billions, why pay for the larger model? As smaller models become more capable, intelligence becomes something we can distribute, specialize, and run across the hardware we already own. That's the future we're building at @thewebAI TwIL-8B is available today. Open weights. And this is just the first of several announcements we're making today. HF Link: huggingface.co/webAI-Officia…
17
1
21
1,713
BREAKING: Binance says the $6.34B hack started with a fake job interview and a poisoned AI coding tool their official report dropped yesterday. the attack took 5 months to set up: → June: an engineer searches for a popular AI coding tool and installs a tampered Homebrew package. security swaps his laptop, but his role on the wallet team is now exposed → September: another engineer gets a job offer on a real job platform. fake HR video call, fake CEO, then an “assignment” repo to run on his work laptop. the malware inside was a 0-day and nothing flagged it → Oct 20: from that laptop the attacker goes through an internal AI tool, reaches two wallet engineers and finds a deployment token → Oct 29, 17:15 to 19:56 UTC: a container gets deployed and a script hits the wallet signing service → 20:05: XRP starts leaving. then SOL, OP, 18,270 BTC, and ETH through an EIP-7702 delegation → 23:56: the last transfer about $6.34B was hit and $2.85B is frozen or intercepted. Binance says it covers the loss itself, and 98% of the tokens were replenished by Oct 31 the FBI attributes it to North Korea’s TraderTraitor group. Binance says it stopped 57 North Korea-linked attempts this year before this one got through if you have a work laptop: install dev tools only from the official site, and never run an interview “assignment” on the machine you work on full report linked in the reply ↓
13
4
23
755
STOP PAYING A FRONTIER MODEL TO READ YOUR EMAIL. Superlinked’s tests: a small open model does the same job in 13 ms instead of 1.48 seconds, at 0.7% of the cost most of what an agent does all day is sorting, extracting and searching. over and over what they measured against GPT-6.1 Sol at list prices: → classify text: 90% of the quality at 0.7% of the cost → detect objects: 104% of the quality at 0.4% of the cost → redact personal data: 95% at 2.9% of the cost → OCR: 91% at 3.8% of the cost → PDF to Markdown: 87% at 9.4% of the cost → extract names and dates: 80% at 0.06% of the cost → chat: 99% at 27% of the cost → median across 14 tasks: 98% of the quality at 3% of the cost the speed gap hurts more than the price. an agent pays it on every call, so a 10-call task pays it 10 times they just launched SIE Serverless, a hosted version of their open-source engine (Apache 2.0, 3.5k GitHub stars). new accounts get $20 in credits, and teams can apply for $500 more the move: send the repetitive calls to small models and keep the frontier model for the work that needs real reasoning these are the vendor’s own benchmarks, so run a few tasks on your own data before you switch anything links to the benchmark and the engine in the reply ↓
🚨 Huge news 🚨 today we are launching SIE Serverless - a managed inference product focused on small models. It’s useful for spiky workloads on 20+ task specific workloads across Search, OCR and Agentic execution. Sign up for a $500 credits launch grant 👇 It’s powered by our open source inference engine that has received thousands of github stars in the last few months, battle-tested across finance, legal & tech. We believe that self-hosting is going to be critical given all the trust & security issues in AI, but practically there are always workloads where you’d like somebody else to deal with the GPU quota like synthetic data generation, evals, public research and beyond. The team at Superlinked has been burning the midnight oil for months and I’m so proud of every single one of you - what a privilege to call you my team mates y’all! We built it, now help us make some noise 🙌 A like, a comment or a repost would mean a lot! ❤️
20
1
31
565
Stop letting Claude invent your life. Most people keep feeding it the same context over and over and still get hallucinations. This is the opposite: one empty folder, four plain-text files, and a rule against guessing. No apps. No plugins. No memory tools. Just folders and one map that loads first. It works like this: → notes: one file per fact or topic → people: one card per person (who they are, what you agreed, last contact) → projects: exact status and where you left off → root: long-term memory, lessons, and every decision you’ve made → CLAUDE.md: the map Claude reads before anything else. It tells it exactly where every piece of information lives Claude is only allowed three answers: → It opens the right file and cites it → It says “not found” instead of making it up → It creates a new file and updates the map Paste this into Claude Code inside an empty folder: Build me a second brain here: 1.Create three folders: notes, people, projects. 2. Create root files for long-term memory, lessons learned and decisions. 3. Write CLAUDE.md as the map: where each kind of info lives, with memory and decisions pinned so they are always in focus. 4. Rules: one topic per file, name each file after what’s inside, lowercase with hyphens. 5. When I ask something, open only the file the map points to and cite it. If it isn’t there, say not found. When I give you something new, create a file and update the map. Show me the structure before you create anything. Set it up once. Use it every day. You stop repeating yourself.
21
26
589
Apple and Google are holding your photos hostage for $9.99 a month. That’s nearly $1,200 over 10 years. Stop paying — and new photos stop backing up. Your memories are now on a subscription. Two 4TB drives cost about $380 once. that’s 3.2 years of rent, and then the monthly bill is gone. One prompt in Claude Code does the boring parts: → checks your disk and tells you if your library fits → installs Immich, an open-source Google Photos on your own computer, plus Tailscale so your phone reaches it privately → pulls your iCloud originals with icloudpd. you type the password yourself, never in the chat → imports your Google library from Takeout with immich-go, so dates and albums survive → compares file counts until it can show N of N on your disk → turns on auto backup in the Immich phone app → copies everything to the second drive → stops there. cancel the plan yourself, and only after those counts match the catch: your computer has to be on for new photos to back up, and every drive dies eventually. that’s why you buy 2 before: $9.99 a month for as long as you keep taking photos after: about $380 once, on 2 drives you actually own prices change by country and by month, so run the math on your own plan first stop renting your own memories. the full prompt is in the reply ↓
10
1
19
600
full prompt is in the reply ↓ Help me move my photo library off iCloud and Google Photos onto this computer. Work one step at a time and ask before you install anything. 1. Check my free disk space and tell me if my library fits. Ask me for my library size if you can’t see it. 2. Install Immich with Docker, then Tailscale, and show me how to connect the Immich phone app. 3. For iCloud, set up icloudpd to download originals. I will type my password and 2FA code into the terminal myself. Never ask me to paste them in this chat. 4. For Google Photos, walk me through Google Takeout, then import the archives with immich-go. 5. Compare file counts between the source and Immich and show me any missing files. 6. Copy everything to my second drive and verify that copy too. Do not delete anything from any cloud, and do not cancel any subscription.
1
2
97