Myttle

losing sleep on @Polymarket
research AI & predictions

[Myttle @xmyttle]

https://t.co/0fhbls3HK5

[Myttle @xmyttle]

@maqxbt thanks for quote

[lagerskoy @lagerskoy]

AI didn't just make the game - It made the trailer too

I asked Claude to build a Minecraft clone and a Roblox obby, then told it to show off its work

It generated procedural worlds, built a playable obstacle course, scripted cinematic camera movements, synchronized jumps to moving hazards, rendered every frame headlessly, and edited the final showcase automatically

We're getting very close to a world where "I have an idea" is enough

[Myttle @xmyttle]

@Nekt_0 Amazing setup

[maqxbt @maqxbt]

Fable 5 is the most powerful public model ever released and it's the weakest version of itself

same base as Mythos 5, different ceiling

80.3% on SWE-Bench Pro. first model to break 90% on Hex analytics

but here's what keeps me up at night

Mythos 5 - the real one - stays locked behind Project Glasswing, uncovering thousands of critical vulnerabilities. Firefox, OpenBSD, infrastructure you use every day

Anthropic just made the two-speed AI release permanent

one model for the public, one for the trusted few
we got the car, they kept the engine

[Myttle @xmyttle]

https://t.co/0fhbls3HK5

[Myttle @xmyttle]

this laptop is doing the part traders still pretend is "SKILL"

> TradingView on one side
> Claude reading the chart
> exchange API underneath
> strategy rules written in plain English

that is the whole shift.

manual trading used to mean:
- see setup
- copy price
- open Bybit
- type order
- check fill
- log notes
repeat

one trade burns ~6 minutes of mechanics.

40 trades a week is 4 hours of pure alt-tabbing.

the agent version is uglier and much more dangerous:
> TradingView MCP reads the chart
> Bybit MCP touches the account
> Claude sees both at once
> MEMORY.md writes the journal

the strategy is not code.

it is a sentence:

if BTC is above the 200 EMA, RSI is 40-55, volume is rising, open the trade, stop at 2%, take profit at 5%, halt at 5% daily drawdown.

that does not make a bad strategy profitable.

it makes a good strategy executable 24/7.

the edge is removing the human from the 6-minute gap between decision and execution.

[Myttle @xmyttle]

this laptop is doing the part traders still pretend is "SKILL"

> TradingView on one side
> Claude reading the chart
> exchange API underneath
> strategy rules written in plain English

that is the whole shift.

manual trading used to mean:
- see setup
- copy price
- open Bybit
- type order
- check fill
- log notes
repeat

one trade burns ~6 minutes of mechanics.

40 trades a week is 4 hours of pure alt-tabbing.

the agent version is uglier and much more dangerous:
> TradingView MCP reads the chart
> Bybit MCP touches the account
> Claude sees both at once
> MEMORY.md writes the journal

the strategy is not code.

it is a sentence:

if BTC is above the 200 EMA, RSI is 40-55, volume is rising, open the trade, stop at 2%, take profit at 5%, halt at 5% daily drawdown.

that does not make a bad strategy profitable.

it makes a good strategy executable 24/7.

the edge is removing the human from the 6-minute gap between decision and execution.

[cristal @0xCristal]

https://t.co/0RtbBTcnl4

[Myttle @xmyttle]

a guy opened the Fable 5 release page and found the part everyone will actually remember

not the benchmark chart.

the examples.

> it beat Pokémon using only screenshots
> played Factorio autonomously
> and Stripe used it on a 50M-line code migration

that is the difference between a model update and a new work layer.

chatbots answer.

agents keep going when the task has 40 steps.

Fable 5 is @AnthropicAI saying the long task gap matters more than the clever reply.

the launch post already has 12M views in a few hours.

people are not reacting to another model name.

they are reacting to the feeling that the ceiling just moved again.

[Myttle @xmyttle]

every four years, football turns into the biggest attention market on earth

2026 is different.

- 3 host countries
- 48 teams
- 16 host cities
- 104 matches
- 12 groups

one final at MetLife

that is not just a tournament.

that is a month-long prediction market engine.

national bias
injury news
group chaos
underdog runs
penalty heartbreak
line movement at 3am

all of it gets priced in real time.

the casual fan sees fixtures.

Polymarket people see hundreds of live storylines turning into markets.

@gmpm_xyz Ep. 7 is going full World Cup mode

@DCBK2LA, @itskkoma and @_bcbread breaking down why this might become the biggest sports prediction market cycle yet.

and the timing matters.

the markets are not fully hot yet.

once the world starts watching, the easy angles disappear.

[Myttle @xmyttle]

a seller found the product before touching Alibaba

he did not start with a trend list

he started with the complaints.

Amazon reviews on one side
Claude on the other
supplier math underneath

that is the whole play.

1-star reviews are not hate
they are free product specs.

one seller pulled hundreds of bad reviews from top listings and let Claude find the repeat failures:

> warped silicone
> sticky lock buttons
> wrong bottle size
> cheap parts
> bad dishwasher claims

then he turned that into supplier questions.
- same category
- better spec
- cleaner listing
- less guessing

the example was boring:

a portable dog water bottle.

$7.11 landed cost
$14.99 sale price
~48% net margin

most Amazon sellers chase what already sells.

the smarter ones read what buyers are begging someone to fix.

[Myttle @xmyttle]

a dev got Claude Opus 4.8 running in Claude Code for $0 to Anthropic

no cracked app.
no fake free AI wrapper.
no browser tab pretending to be an IDE.

just Claude Code CLI reading a different config.

most people think the subscription is the product.

it is not.

the product is the agent shell:
- repo access
- terminal control
- file edits
- multi-step debugging
- the whole coding loop

Claude Code only needs three things from settings.json

API key
base URL
traffic rules

point the base URL at a third-party provider
drop the key in
open terminal
type claude
that is why this is spreading.

the expensive part was never the command line.
it was where the requests were routed.

setup is stupidly small:
> Node.js v18+
> npm install -g @anthropic-ai/claude-code
https://t.co/aubpecMUcX account
> Telegram verify
> API key
> ANTHROPIC_BASE_URL=https://t.co/YH4BPIMGQZ

then ask:
which model are you?

and the terminal answers:
Opus 4.8

free credits are limited.
rate limits still exist.
heavy users will eventually pay somewhere.

developers are realizing they do not need to rent the whole subscription

just to keep the Claude Code workflow.

[Myttle @xmyttle]

a guy opened the Fable 5 release page and found the part everyone will actually remember

not the benchmark chart.

the examples.

> it beat Pokémon using only screenshots
> played Factorio autonomously
> and Stripe used it on a 50M-line code migration

that is the difference between a model update and a new work layer.

chatbots answer.

agents keep going when the task has 40 steps.

Fable 5 is @AnthropicAI saying the long task gap matters more than the clever reply.

the launch post already has 12M views in a few hours.

people are not reacting to another model name.

they are reacting to the feeling that the ceiling just moved again.

[Claude @claudeai]

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

[Myttle @xmyttle]

every four years, football turns into the biggest attention market on earth

2026 is different.

- 3 host countries
- 48 teams
- 16 host cities
- 104 matches
- 12 groups

one final at MetLife

that is not just a tournament.

that is a month-long prediction market engine.

national bias
injury news
group chaos
underdog runs
penalty heartbreak
line movement at 3am

all of it gets priced in real time.

the casual fan sees fixtures.

Polymarket people see hundreds of live storylines turning into markets.

@gmpm_xyz Ep. 7 is going full World Cup mode

@DCBK2LA, @itskkoma and @_bcbread breaking down why this might become the biggest sports prediction market cycle yet.

and the timing matters.

the markets are not fully hot yet.

once the world starts watching, the easy angles disappear.

[Dave @DCBK2LA]

⚽ @gmpm_xyz Ep. 7 is going WORLD CUP MODE!!

World Cup Fever Starts NOW

Join Dave, Kkoma & Bread as we break down why the 2026 @FIFAcom World Cup is about to become the biggest prediction market storyline in all of sports

🌎 3 host countries
👥 48 teams
🏟️ 16 venues
⚽ 104 matches
🔥 12 groups
🏆 Final at MetLife Stadium

The craziest month in sports that happens every 4 years

Plus, @Stitch3_ai is officially launching a creator rewards campaign for the gmPM show, giving creators a new way to earn by posting, clipping, and helping push the show across the timeline.

Campaign details to follow so turn on your notis 🔔

This will be our biggest show yet!

Hosted by @DCBK2LA @itskkoma & @_bcbread
Powered by @Stitch3_ai & @bvcket_xyz

Live across X, YouTube, LinkedIn, Instagram, and TikTok

[Myttle @xmyttle]

a seller found the product before touching Alibaba

he did not start with a trend list

he started with the complaints.

Amazon reviews on one side
Claude on the other
supplier math underneath

that is the whole play.

1-star reviews are not hate
they are free product specs.

one seller pulled hundreds of bad reviews from top listings and let Claude find the repeat failures:

> warped silicone
> sticky lock buttons
> wrong bottle size
> cheap parts
> bad dishwasher claims

then he turned that into supplier questions.
- same category
- better spec
- cleaner listing
- less guessing

the example was boring:

a portable dog water bottle.

$7.11 landed cost
$14.99 sale price
~48% net margin

most Amazon sellers chase what already sells.

the smarter ones read what buyers are begging someone to fix.

[Insomnia @insomnia_vip]

https://t.co/qREkH5pR3u

[Myttle @xmyttle]

a dev got Claude Opus 4.8 running in Claude Code for $0 to Anthropic

no cracked app.
no fake free AI wrapper.
no browser tab pretending to be an IDE.

just Claude Code CLI reading a different config.

most people think the subscription is the product.

it is not.

the product is the agent shell:
- repo access
- terminal control
- file edits
- multi-step debugging
- the whole coding loop

Claude Code only needs three things from settings.json

API key
base URL
traffic rules

point the base URL at a third-party provider
drop the key in
open terminal
type claude
that is why this is spreading.

the expensive part was never the command line.
it was where the requests were routed.

setup is stupidly small:
> Node.js v18+
> npm install -g @anthropic-ai/claude-code
https://t.co/aubpecMUcX account
> Telegram verify
> API key
> ANTHROPIC_BASE_URL=https://t.co/YH4BPIMGQZ

then ask:
which model are you?

and the terminal answers:
Opus 4.8

free credits are limited.
rate limits still exist.
heavy users will eventually pay somewhere.

developers are realizing they do not need to rent the whole subscription

just to keep the Claude Code workflow.

[beamnxw ./ @beamnxw]

https://t.co/ierYyHBhmA

[Myttle @xmyttle]

this ugly Polymarket terminal is what
AI tradingactually looks like

not a chatbot giving opinions

a feed on one screen
markets on another
a bot waiting for the price to break before humans can react

the edge is not prediction anymore.

it is speed, routing, and probability updates running while traders are still reading the headline

Polystrat did 4,200+ trades in its first month
one position peaked at +376%

other agents are watching thousands of markets at once

news
order books
liquidity
API fills
market odds

then they compare their probability against the price and execute.

that is the shift.

@Polymarket used to be humans betting on events

now agents are pricing the odds of other agents, models, IPOs and AI labs with real USDC on the line

benchmarks can be gamed.
a bad trade cannot.

[maqxbt @maqxbt]

https://t.co/G6ZcyrCW5W

[Myttle @xmyttle]

a guy put a local AI coding stack on an Android phone

Termux on the screen
Ollama underneath
a local model answering from the device

that tiny terminal is the shift.

two years ago, local LLMs felt like punishment

fans screaming
bad models
one token per second
cloud looked inevitable

now the same idea runs almost anywhere.

- Raspberry Pi for tiny chat
- MacBook Air for daily writing
- RTX 3090 for serious inference
- LM Studio when you want the clean GUI
- Ollama when you want apps to connect

the point is not replacing GPT-5.1 or Opus 4.8 on the hardest tasks.

the point is owning the boring 80%:

- drafts
- summaries
- code help
- private docs
- offline workflows

no token anxiety
no rate-limit ceiling
no sensitive files leaving the machine

cloud AI became the premium layer.
local AI became the workbench.

[leopardracer @leopardracer]

https://t.co/RnHXqwsem7

[Myttle @xmyttle]

@undefinedKi Banger

[Myttle @xmyttle]

@leopardracer banger