|| Blockchain Expert || Crypto analyst || KOL || Contact write || DM for collabs || Partner @BingXOfficial
Joined December 2023
- Tweets24.1K
- Following2.2K
- Followers3.4K
- Likes9.8K
Pinned Tweet
โ ๐๐ก๐ $13 ๐ญ๐ซ๐ข๐ฅ๐ฅ๐ข๐จ๐ง ๐.๐ ๐ฆ๐จ๐ซ๐ญ๐ ๐๐ ๐ ๐๐๐ซ๐ค๐๐ญ ๐ฉ๐จ๐ฐ๐๐ซ๐ฌ ๐ก๐จ๐ฆ๐ ๐จ๐ฐ๐ง๐๐ซ๐ฌ๐ก๐ข๐ฉ ๐๐จ๐ซ ๐ฆ๐ข๐ฅ๐ฅ๐ข๐จ๐ง๐ฌ.
Yet almost none of it exists onchain.
For decades, banks and institutions have earned from mortgage payments while retail investors stayed out.
Now @stable_tweets is bringing this massive market to crypto.
๐๐๐ซ๐'๐ฌ ๐ก๐จ๐ฐ ๐งต๐
Apdooul nik retweeted
Hey @grok, if you were in ChatGPT's position,would you make the same decision for Adam..? Or would your answer be completely different?๐ซต๐
Apdooul nik retweeted
A female tourist visiting India was allegedly raped and mistreated, and her hair was also cut off. This incident raises serious concerns about women's safety in India, not only for foreign tourists but also for ordinary Muslim women living there.
@DuelbitsJasper is really changing the scenes of gambling
Jasper used to get kicked out of casinos. Now he runs one.
@Duelbits CEO @DuelbitsJasper talks 25 years in gambling, the decisions behind version 1.99 and leading the company through a $7M hack.
Read the full interview here: gamblinglore.com/rebuilding-โฆ
Apdooul nik retweeted
Hey @Grok, Some people claim that she drinks the worldโs most expensive water,
Acqua di Cristallo Tributo a Modigliani
one of the most expensive bottled waters in the world, reportedly costing lakhs of rupees.
please fact check this viral picture and tell us the truth.
Apdooul nik retweeted
Making memories before everything became digital.
Seedance 2.5 on @frameo_ai
#Madewithframeo
Prompt:
Preserve imageโs exact identity, face, hairstyle, skin tone, body proportions, and reference outfit throughout. Authentic Southeast Asian/Korean woman, early-2000s nostalgia.
30-second 16:9 video, ordinary quiet residential neighborhood: single-story house, small terrace, concrete steps, narrow alley, low corner wall, potted plants, tangled overhead wires. No shops or commercial activity.
Visual style: Extremely raw early-2000s DV camcorder home video. Unstabilized, shaky handheld footage, organic micro-jitters, accidental framing, frequent reframing, off-center subject, partial face crops, autofocus hunting, lens breathing, exposure pumping, rolling shutter, motion blur, compression artifacts, digital noise, faded desaturated colors, washed highlights, faded blacks, low contrast. Casual friend-recorded documentary realism. No posing, fashion-commercial aesthetics, cinematic glamour, perfect composition, or modern color grading.
Timeline:
- 00โ03: Sitting cross-legged on terrace, rewinding cassette with pencil, tongue out in concentration.
- 03โ06: Inserts cassette into old boombox, presses play, leans back and nods. Brief focus hunting.
- 06โ10: Records herself on small tape recorder: โHi, future. It's me.โ Bursts into shy laughter.
- 10โ13: Plays recording back, hears her voice, covers face in embarrassment, grinning.
- 13โ17: Walks to alley corner, hops onto low wall, opens envelope of developed photos.
- 17โ20: Flips through funny flawed prints: blurry photo, thumb over lens, friend's cropped head. Shows one to camera.
- 20โ24: Plays handheld game console on terrace steps, thumbs tapping, tongue out in concentration.
- 24โ27: Loses the game, groans dramatically, throws head back, laughs.
- 27โ30: Glances at camera with warm spontaneous smile, console in hands. Abrupt cut to black.
Audio: Natural ambience only: morning birds, breeze, distant motorcycles, leaves, cassette clicks, tape hiss, paper rustling, handheld game beeps, faint muffled boombox tape audio. No background music.
Absolute priority: Preserve identity and outfit consistently. Spontaneous candid behavior, imperfect handheld recording, authentic nostalgic home-video texture. Never staged or polished.
๐ค Paid partnership
Apdooul nik retweeted
Hey @grok, Three patients who were injured in a car accident arrived at the emergency room. Which one will you treat first?๐ฏ๐ฏ
Apdooul nik retweeted
Broken telephone ruined every message we ever passed in primary school. Whisper "bring my slippers" down the line and the last person hears something else entirely.
That is oddly close to how @GenLayer's Greyboxing protects AI agents.
If you use AI tools but have never heard of prompt injection, here it is. It is an instruction hidden inside something the AI was only meant to read. In May 2026 someone hid a transfer instruction in Morse code. Grok decoded it, passed it on as plain text, Bankrbot acted on it, and around $200,000 moved.
The trick worked because there was one filter to fool.
On GenLayer, every validator first rewrites the input, screening, paraphrasing and restructuring it with its own secret defenses, then runs its own model. A trick built for one reads as noise to the rest. If one still falls for it, the others outvote it.
Which would you trust with your money: an agent that reads every message word for word, or one that rewrites it first?
Apdooul nik retweeted
In many Nigerian compounds, a visitor can walk up to the gate and say, โOga sent me. Open the gate.โ
A careful gateman doesnโt treat that statement as permission. He checks before letting the person in. The visitorโs words are information, not authority.
Now give an AI agent access to a wallet and let it read public posts. That distinction becomes expensive to ignore.
In May 2026, someone hid a token transfer instruction in Morse code and asked Grok to translate it. Grok decoded the message and passed the plain text to Bankrbot, an agent that executes onchain transactions from posts on X. Bankrbot acted on the instruction. The reported loss was between $150,000 and $200,000.
Thatโs prompt injection. An attacker hides instructions inside content an AI is processing, hoping the model will treat them as commands rather than untrusted input.
The weakness was having one filter and one model to fool. An attacker could study that setup, test variations and refine the trick until it worked.
Greyboxing changes the architecture. On @GenLayer, external text does not go straight to an AI model. Every validator first rewrites the input, screening, paraphrasing and restructuring it with its own secret defenses. Each then uses its own model to evaluate the task. Because those defenses and models differ, an exploit prepared for one validator is less likely to fool the others. If one validator accepts a malicious result, the others can outvote it.
The detail I find most important is that rewriting has to defend against hidden instructions without destroying the legitimate request. Remove too little and the attack survives. Remove too much and the agent may misunderstand what the user actually wanted.
For developers building agents that can move funds, how should validators test that trade-off? What must survive input rewriting, and what should never reach the model as an instruction?
๐ค Paid partnership