Sabitlenmiş Tweet
whydeso
955 posts
whydeso
@whydeso
Polymarket bets × AI tools I find edges so you don't have to
onchain Katılım Aralık 2019
168 Takip Edilen229 Takipçiler
China just shipped the largest open-weight ai model ever.
Kimi k3, 2.8 trillion parameters, full weights public july 27
On independent tests it ranks 4th in the world, ahead of claude opus 4.8, behind only Fable 5 and GPT-5.6.
And its a third the price $15 per million tokens vs fable's $50
This isnt "china catching up". its the floor moving. When a frontier model is open AND cheap AND self-hostable, the expensive part of building stops being the model
everyone gets to start
Moonshot's raising at $31.5b because investors see exactly what that unlocks
open, massive, cheap
you trying it or staying locked in your api
English
Spain vs Argentina
the wall vs the comeback kings
Spain 59%, Argentina 41%
but spain havent lost in 37 games and conceded ONE goal all tournament
theyve never even trailed
Argentina were five minutes from elimination vs England and won anyway
Thats just what they do.Messi gets Yamal, the kid he held as a baby in 2007, in his last ever World Cup game
$4.28B traded this tournament, only 6.5% touched the final
the market saved its biggest question for last
Who picks up the World Cap?
English
@greednvirtue Exactly, the delays are whatever, but the constant missed dates killed all trust
English
Gemini 3.5 Pro missed its release target for the third time
Look at the odds. The near dates are all in single digits
The money's been pricing "delayed again" for weeks and its been right every time
I flagged this exact model on July 9, the empty chair while GPT-5.6 and Grok 4.5 shipped
Google now scrambling for a stopgap
When the board fades your roadmap harder than your rivals, thats not a calendar issue
Whats going on at google?
English
yesterday i said Kimi K3 built an RDR2-style scene in 10 hours and gaming twitter has been beating me up in the replies ever since
they're right about one thing: a 10 hour AI scene is not Red Dead Redemption 2. obviously. RDR2 is 500,000 lines of dialogue, years of mocap, a decade of craft. no prompt replaces that
but the pitchforks are aimed at the wrong sentence:
> that scene cost about $10 in tokens. ten dollars, against a $500M pipeline. the floor moved, not the ceiling
> regular coding sessions on K3 run $0.08-0.10. the "expensive" part of building just stopped being the building
> nobody said AAA is dead. the wall around "playable and beautiful" got a door, and doors change who walks in
the angriest replies think this is about replacing artists. it's about who gets to start
in 2013 one guy with a GTA calculator site got 2M visitors in a month. the K3 version of that guy is reading this thread right now, with $10 of credits
keep the pitchforks coming though. the algorithm loves you
Dep@0xDepressionn
Rockstar spent 7 years, 2,000+ developers, and $500 million to build Red Dead Redemption 2. Kimi K3 rebuilt the same scene in 10 hours. One prompt. 4 million tokens. It's open source. 2.8 trillion parameters. A 1 million token context window. It beats Claude Opus 4.8 and GPT-5.6. It sits on par with Fable 5. Fable 5 results, at Sonnet pricing. The wall between you and a AAA game used to be a studio. Now it's a prompt.
English
@RoundtableSpace This tool turns public data into real intelligence, saving it
English
THIS OPEN-SOURCE TOOL BRINGS PALANTIR-STYLE DATA ANALYSIS TO EVERYONE
* Connects public records into searchable knowledge graphs with AI agents
* Runs on Windows, macOS, and Linux with OpenAI, Anthropic, OpenRouter, and Ollama support
Repo: github.com/ShinMegamiBoso…
English
NVIDIA CEO, Jensen Huang:
"Nobody writes prompts anymore. The new job is to write and handle loops."
This is the shift that's going to define the rest of 2026.
53 minutes of pure insight from one of the richest men on earth.
Watch it, then read the full guide on how to actually use loops below.
Anatoli Kopadze@AnatoliKopadze
English
@PolymarketTrade Bonnie Blue just cooked the daily Combo Cup, absolute monster play
English
Polymarket user "Bonnie Blue" hit an 8 leg Combo, turning $10 into $1300 — & got a $20k bonus for being #1 in the daily Combo Cup challenge.
English
IF YOU USE CLAUDE CODE, YOU CAN'T SCROLL PAST THIS ONE.
Moonshot published an official config: 9 variables, and your exact setup runs a 2.8 trillion parameter model at a third of the price.
And everything survives the swap:
→ your CLAUDE.md, hooks, and subagents keep working untouched
→ auto-compact stretches to the full 1M context
→ input drops from $10 to $3 per million tokens
→ repeated context bills at $0.30 on cache hits
→ thinking on by default, no toggles needed
→ /status confirms the swap (the /model menu won't even show it)
The config costs $0. The model behind it runs at a third of Fable's price.
Copy it free 👇
darkzodchi@zodchiii
English
I spent a year renting intelligence that was sitting on my desk the whole time.
Ollama crossed 150,000 stars on GitHub because developers keep discovering the same thing: a $799 Mac mini runs most of what they pay $200 a month for.
The trick is unified memory. Apple shares one pool between CPU and GPU, so the box loads models a gaming PC at the same price cannot. Silent, 30 watts, running around the clock.
And it compounds: new open models drop every month, and the same box gets smarter for free. The subscription price only moves up.
here's the whole thing:
> Ollama: one curl command, any model becomes a local API: 5 min
> Qwen2.5: 25 tokens a second, instant chat: loaded
> Open WebUI: a private ChatGPT in the browser: 10 min
> one line: base_url points home, every tool follows: done
15 minutes to set up, and the meter stops running every time you ask a question.
The exact config to buy is in the article above.
Grimmer@0xGrimmer_
English
Player to Miss Penalty in Finals: This is a new market at @Polymarket
In this market when you see the odds, it's low on YES
Because people thinking there will be No penalty comes.
But they need to remember it's a final match so all players will show there best
We will see players' aggression and some cards too (maybe) so in that moment we will see penalties.
So these are the top 4 who has higher chance to hit a penalty.
-> Lionel Messi
-> Mikel Oyarzabal
-> Ferran Torres
-> Lautaro Martínez
Now it's also not sure that they are going to miss, but pressures can do anything so maybe they will miss.
If you invest $1 on these 4, the cost will be $4, but if won then the minimum profit is $7 and the maximum $18
English
A QUANT SPENT YEARS AND A PHD BUILDING A STRATEGY THAT LOSES MORE THAN HALF ITS TRADES AND STILL PRINTS
The secret is one line.
w(t) = Rmax/(J·σ_risk) · (Σ sⱼ) · [1 + η·Σ 𝟙(c ≥ nₖ)]
It sizes every bet by current risk, and that last bracket presses harder every time the signal confirms.
In plain words. Keep the losers tiny, and pyramid hard into the winners.
So it wins under half the time and still beats the market, because one winner erases ten small losses. That positive skew is the edge quants chase for years.
Minara lets you build that exact behavior by describing it. 500+ factors, risk-scaled sizing, real backtests. The PhD math is real, the wall is gone.
I took one plain sentence to a live, backtested strategy & posted every number in the article. Check it ↓
slash1s@slash1sol
English
@akshay_pachaar Love the diagram, this setup seems really well thought out
English
I just built my own coding harness, just like Claude Code. (100% open-source)
A harness is the code wrapped around an LLM. The model only decides the next step, and the harness handles everything else, planning, tools, memory, and safety.
In simple words model acts as the brain and the harness provides the hands.
A coding harness is that same wrapper pointed at a codebase. It turns a plain text generator into something that reads files, edits code, runs tests, and fixes real bugs.
Here's everything that went into mine:
- the core agent loop
- file tools that double as external memory
- planning for long-running tasks
- subagents that work in their own context
- sandboxed execution in a throwaway VM
- human-in-the-loop approval
- persistent memory and checkpointing
The whole thing is built on CrewAI, a 100% open-source framework.
I also wrote an article on this. It covers everything that goes into building a coding harness, the agent loop, planning, subagents, sandboxing, memory, and checkpointing, built step by step.
The article is quoted below.
GIF
Akshay 🚀@akshay_pachaar
English
A TEAM IS PLANNING TO RUN A 2.8-TRILLION-PARAMETER MODEL WITHOUT SENDING A SINGLE TOKEN TO A CLOUD
Most open-weight models this capable simply don't exist yet, so teams handling sensitive data default to closed APIs.
Kimi K3's full weights are scheduled for release on July 27 under a Modified MIT license, right after launching as the largest open model ever shipped.
Once those weights land, fine-tuning it on internal codebases and documentation becomes possible for the first time.
For client systems, financial models, or healthcare data, running it locally means nothing ever leaves the infrastructure it's already on.
A 2.8T MoE model still needs serious GPU infrastructure, so this favors teams with existing clusters before it favors individual developers.
See what changes on the 27th below👇
wast3@0xWast3
English
STOP PAYING FOR CAPCUT A PROJECT CALLED OPENCUT JUST SHOWED UP ON GITHUB, SAME EDITOR, FULLY FREE.
Why is no one talking about this
75,300 stars and 7,600 forks on GitHub and the count keeps climbing.
CapCut slaps a watermark on your exports, hides features behind a paywall, and charges a subscription for things that used to be in the base version.
A group of developers got tired of waiting for that to change and built an alternative from scratch.
It’s called OpenCut. MIT license you can do whatever you want with it, including commercial use.
What they’ve shipped so far:
→ Full timeline with multi-track editing
→ Rust core with a GPU compositor, effects and masks (compiles to WASM for the web)
→ Your video stays on your device nothing gets uploaded to a server
→ Web version is ready, desktop is in active development on GPUI
→ No watermarks, no paywalls, no account required
How to run the web version locally:
→ Fork and clone the repo
→ Spin up the database and Redis with docker compose up -d db redis
→ bun install and bun dev:web - editor opens on localhost:3000
This is what CapCut should have been from day one.
Save this post you’ll want it the next time CapCut asks you to renew your subscription just to export without a watermark.
GitHub below 👇
AdiiX@adiix_official
THIS DEVELOPER JUST KILLED THE ENTIRE VOICE CLONING INDUSTRY WITH ONE GITHUB REPO Claude now speaks in your own voice across 23 languages through a single MCP call. Free. Local. Offline. ElevenLabs charged $22-$330 a month for what Jamie Pine just gave away under an MIT license. Claude Code, Cursor, and Cline hook in with one command and start replying in your cloned voice straight from the terminal. → developer Jamie Pine dropped the repo jamiepine/voicebox on GitHub 41.6k stars in a matter of weeks → clones any voice from 10 seconds of audio, seven TTS engines to choose from → runs on your own machine via Tauri (Rust, not Electron) voice and samples never leave your computer → Claude Code, Cursor, Cline, and any MCP-aware agent speak back in your chosen clone through a single voicebox.speak call → plus global dictation with one hotkey inside any app → generate content in Arabic, Japanese, or Polish without opening your mouth Replaces ElevenLabs and WisprFlow in one MIT-licensed app. Bookmark this for the moment you want to give your agent a voice, repo below 👇🏻
English
THIS IS WHAT CLAUDE FABLE 5 NIGHT SHIFT LOOKS LIKE. FIVE STAGES, 48 TESTS, $0.06 SPENT, AND ZERO PERMISSION TO APPROVE ITS OWN WORK.
the system starts by reading the latest logs, then Fable chooses one high-value task instead of wandering through the entire codebase.
a cheaper worker edits the code inside an isolated branch, while a fresh Fable instance receives only the work order and final diff for review.
the screen shows the full separation. 46 lines added, 17 removed, review passed, then typecheck, tests, lint, and health checks run automatically.
all 48 tests pass in 1.7 seconds, but the task remains on probation at 16 out of 20 successful runs before it can ship without approval.
one complete shift costs $0.0648 against a $5 daily budget. This is what happens when you stop chatting with Claude and start managing it like an employee.
Gipp 🦅@gippp69
English
Andrej Karpathy in Stanford classroom said:
"GPT is not a language model - It's a general-purpose computer"
90% of AI Twitter still thinks it's just autocomplete.
It's not.
>90% of the tools built on top of LLMs treat them like a smarter search engine.
In this 40-minute lecture, Karpathy builds nanoGPT live
300 lines, reproduces GPT-2, explains every single decision.
Worth more than any $500 deep learning bootcamp.
Ricker@0xRicker
Andrej Karpathy just said what no one wants to admit: "Your agent loop is generating garbage at scale and you can't see it." 1 output looks fine. 1,000 outputs = dead entropy, dead diversity. 90% of AI companies are building demos, not products. In a 20-minute conversation, Karpathy reveals the full reasoning stack
English
Claude Code turned 4,700 Obsidian notes into a working neural network in one evening.
Not a metaphor. An actual net. 17 inputs. 26 neurons in the hidden layer. ReLU activation. Every input node is a folder from his vault.
The setup took 3 prompts.
Prompt 1: Claude Code reads the vault. 4,700 markdown files, 6 years of notes, 2.1M words. It builds embeddings for every note and maps the link graph.
Prompt 2: it writes the network from scratch. No PyTorch. No TensorFlow. Raw JavaScript, 900 lines, running in the browser. Orange wires for strong weights. Yellow for weak.
Prompt 3: train it on his own behavior. Which notes he opens. Which links he clicks. Which ideas he drops after 2 days.
By 1 AM the screen looked like a brain scan.
Now the net predicts what he wants to read before he searches. He opens Obsidian at 7 AM and 5 notes are already waiting. Last week it flagged a link between a 2021 note on attention and a 2026 note on prediction markets. He turned it into a post. 214,000 views.
The vault stopped being storage. It started being a coworker.
6 years of notes for a future self who never showed up.
He built the future self in one night.
Spike 1%@SpikeCalls
English