sanek

AI researcher | Agents expert | Always alpha

[sanek @sanekbob]

Polymarket is busy fighting over a $5.4M dispute, and people are just posting what the Tesla Roadster test drive actually looks like

Skipping traffic over a city bus, swapping tires mid-air without a jack, and doing flips past speed cameras

Honestly this feels closer to reality than waiting for the real delivery date

[Murmur @armouredme]

UPDATE: Polymarket stepped in and killed the debate on the $5.4M Tesla Roadster contract 

"Yes" just printed straight to >99% 

The team updated market context with an official ruling:
Tesla's post on X citing 0.81s satisfies the "official Tesla-published time" clause in the rules 

How the order book reacted:
-> Volume exploded from $240K to over $5.4M
-> Over $2M in buy orders stacked at 99.9¢ with zero asks left
-> The second UMA dispute is effectively irrelevant now 

Track the resolved order book:
https://t.co/gCjHgiRSBN

"No" holders traded the literal video wording, but the platform prioritized the official company tweet
In prediction markets, platform clarification always overrides community interpretation

[sanek @sanekbob]

@EagleVisions Wish you good luck

[sanek @sanekbob]

@EagleVisions That's the main feature

[sanek @sanekbob]

$30 a month now gets you a small team of bots that does the busywork
Grok Bot is included from the $30 Premium tier, and these presets are where it starts paying off:

>Haggle Bot finds SaaS savings and drafts counteroffers to vendors
>Credit Card Max tracks card perks before they expire
>Executive Assistant preps meetings and books rooms
>GTM Loop Closer turns forgotten "I'll send that over" into ready drafts
>Clip Bot cuts captioned clips from any podcast or stream
>Stalk Bot watches competitors and what their customers say on X
>Startup QA Bot tests your product daily and reports what broke

new since september:
every bot can search X with no connector
each account can claim its own email address
Shopify and Finance connections
free suggestions from your primary bot

two rules from the docs
anything the bot can undo, it does alone
anything that sends, pays or deletes waits for you

and hourly routines can burn a week of usage in a day

[Annatar.md @AnnatarXBT]

https://t.co/JjftMg1sjd

[sanek @sanekbob]

@armouredme According to rules it should be "No"

[sanek @sanekbob]

@semichenkko Grok hypes

[sanek @sanekbob]

@AleiahLock Imagine they would need to sleep like you

[sanek @sanekbob]

@bonsaixbt It even doesn’t look like it’s made with AI

[sanek @sanekbob]

@Flinzu Never would think 8 sessions look like that

[sanek @sanekbob]

giving an AI agent its own computer just killed proprietary software

​the breakdown of how Grok Bot + REA turns any closed application into an open blueprint:

->​ Grok Bot gets persistent cloud computer access with its own terminal and filesystem
-> ​connects to REA via standard MCP server protocol (npx -y rea-agents@latest mcp)
-> ​traces native execution paths, extracts internal calls, and writes verified SPEC.md without human input
-> ​runs automated clean-room tests until the reproduction matches original software behavior 1:1

​people thought agents would just write boilerplate CRUD apps
instead Grok Bot is autonomously ripping apart paid binaries and decompiling commercial games in 30 minutes

​if code touches the client side, an agent with bash access can rebuild it overnight

[shmidt @shmidtqq]

https://t.co/vc7yF1TP5P

[sanek @sanekbob]

@martynov014 Kyle?

[sanek @sanekbob]

@Pashoke1 Trust issue

[sanek @sanekbob]

@Pashoke1 Swag

[sanek @sanekbob]

an entire animation studio replaced by three models and one recurring character

someone turned AI short-form content into a $24,689 a month business across TikTok and YouTube
no camera, no animators, no studio overhead

the production loop:
-> GPT-6 Astra pulls data and structures the world
-> Claude Opus 5.5 writes the scripts so episodes never repeat
-> Picsart agents handle rendering, voiceover, and formatting

a mummy doing pull-ups in a gym, fighting crocodiles, and taking selfie with a Glock on top of the pyramids
surreal visuals keep retention high, and automated pipelines deliver 75-second episodes daily

people still debate whether AI video looks realistic
meanwhile creators are building automated content factories printing five figures every month

[RetroChainer @RetroChainer]

https://t.co/AMq1ExMMlc

[sanek @sanekbob]

The viral Opus and Haiku setup is walking dead

On a single bug fix, the three-model stack cost $1.37
Plain Sonnet 5.5 on its own did the exact same fix for $1.14
Thousands bookmarked this configuration as a cost-cutter without checking the math

The cheap model was not the leak
Haiku subagents took 3 cents of the whole run
Opus advisor ate 54% of the bill because every advisor call processes the transcript anew with zero cache reuse

The saving does not come from how cheap Haiku is
It comes from how much heavy lifting you pull off Opus

The real fix:

> Haiku searches files, reads git logs, and parses test outputs
> Sonnet writes the actual implementation
> Opus strictly signs off on architectural plans and reviews

A cheap setup is not a model pick
It is a routing problem

[Skaly_Bull @Skaly__Bull]

https://t.co/AjeeVfgoFV

[sanek @sanekbob]

Cleopatra shooting rifle by pyramids, throwing cash at ancient temple rave, and walking out to supercars on red carpet

Three things breaking the brain here:
> Hollywood blockbuster visuals that should cost hundreds of millions, built entirely from text prompts
> lighting, depth of field and frame density no longer give away AI—fabric textures, smoke and metal reflections look disturbingly natural
> zero need for film crews, Cairo expeditions, stunt doubles or complex CGI

A year ago this video would be called impossible fake or top-tier studio VFX
Today it is just one person, good taste and generative models

Cinematography is now available to anyone with crazy idea
Who would you direct next?

[sanek @sanekbob]

AI models are no longer competing on pixel quality
They are competing on scale of insanity

Pharaoh drifting a Porsche 911 in the desert, priests running a pit stop, and a chariot lining up for a drag race against a supercar
All with perfect cinematic lighting, accurate dust physics, and vintage film grain

A year ago this level of surrealism required Hollywood budgets, stunt drivers, and months of VFX studio work
Now it is just one prompt turned into pure art

Video generation reached a point where technical execution is no longer a bottleneck
Only limiting factor left is creator imagination

Boundary between imagination and reality on screen is officially erased

[Mika @mika_systems]

https://t.co/EvoaFa63Cd

[sanek @sanekbob]

Pharaoh in full headdress washes a supercar on the banks of the Nile

Three things doing the work:
1. One impossible vehicle against one of the most recognisable backdrops on earth
2. Vintage film grain makes the footage feel real, and the absurdity alone stops the scroll
3. Ancient workers around him act completely normal, which sells the whole thing

A year ago a concept this exaggerated would have melted the budget
Now it holds for a full clip in a busy, real looking location, and the scarce part is the idea, not the render

Text-to-video is the cheapest way to test an idea this strange: one prompt, one line of motion
Who would you drop into ancient Egypt next?

[sanek @sanekbob]

@0xKorens Relatively small if we talk about 1 trade

But those size is different

[sanek @sanekbob]

Building a million-dollar influencer now costs less than a decent microphone

The article above breaks down how anonymous founders are launching viral AI personas like Jean Phil for a $500 production budget
Combine that workflow with a raw video generator, and you get this

A mummy cooking on an industrial stove, serving backhands on a tennis court, and landing BMX tricks by the pyramids

You no longer need a production agency or a charismatic actor to build an audience
The new creator economy is entirely faceless
It just requires a character sheet, consistency, and a few API calls

[BuBBliK @k1rallik]

you don't need a $299 course to create an AI character.

the workflow is free to learn:
→ design a face
→ build a character sheet
→ animate, edit, publish

I broke it down using Jean Phil and Derek. tools, costs, no paywall. https://t.co/AbejEMrZ4Q