petek

Big fan of volunteering. Living and learning.

[petek @Petek98]

I creep towards the white glow hoping for some sort of answer to this existential pain. I meet the beaten down doorway and grip the cold metal of the knob, twisting it and pulling the creaking door towards my figure. With the glint growing vastly I peek my eyes i

[petek @Petek98]

@semichenkko

[petek @Petek98]

ADDVOCATED BY SOME ELITES, THIS BELIEF CAN LEAD TO INCREASED CRIME RATES AFFECTING POORER COMMUNITIES.EEFUNDING THE POLICE
CXAMPLES OF “LUXURY BELIEFS”
HARACTERISTICS OF “LUXURY BELIEFS”

[petek @Petek98]

@xvoon0 there's a reason this went viral and it's not the obvious one...

[petek @Petek98]

@BlumeDotCodes

[kerazcity @0xkerazcity]

your agent is not a loop

a loop answers one question:

should execution continue?

production agents need two more:

where does execution go next?

what is execution allowed to change?

LOOP vs GRAPH vs HARNESS ENGINEERING

LOOP ENGINEERING

controls time.

- decides whether another attempt is justified
- owns retries, budgets, progress, and exits

if execution repeats without new evidence, the loop is broken.

GRAPH ENGINEERING

controls state.

- decides the next valid transition
- owns routing, branches, checkpoints, and recovery

if the right action reaches the wrong state, the graph is broken.

HARNESS ENGINEERING

controls execution.

- decides which tools and side effects are allowed
- owns permissions, isolation, evals, and approvals

if the workflow performs the wrong action, the harness is broken.

LOOP answers:

should we continue?

GRAPH answers:

where do we go?

HARNESS answers:

what may we change?

the model generates one step.

the architecture decides everything after that.

save this before adding another retry.

[Avid @Av1dlive]

holy shit this is a f**king gold mine

I replaced opus 5 with deepseek v4-flash, it's basically free

and someone figured out a way to run it's subagents as a self-improving loop inside codex

[the setup takes 2 mins here is how]

1. clone the repo 'codex-self-imrpoving-loop'
2.give the '3 loop workflow' image to your agent
3. voila you are done

save this no matter what. it'll be the most productive thing you do this week

[Terminator @TerminatorCV]

ONLY ONE OF THE FOUR WAYS TO SPEND MONEY IS ACTUALLY EFFICIENT.

Milton Friedman explained why in a single interview clip that's outlasted entire economics departments.

Box one: your money, on yourself. You watch the cost and the quality, because you're the one paying and the one using it — this is the only box where both instincts fire at once.

Box two: your money, on someone else. You still watch the cost, since it's coming out of your pocket, but the quality matters less to you, because you won't be the one experiencing it.

Box three: someone else's money, on yourself. Cost stops mattering — it's not your money — but quality matters intensely, because you're the one who has to live with the result.

Box four is where almost every dollar of government spending lives: someone else's money, spent on someone else. Nobody in the chain has a personal stake in either the cost or the quality. The taxpayer funding it isn't deciding how it's spent. The official deciding isn't paying for it. The recipient isn't paying either.

Friedman's argument wasn't about corruption or bad actors. It's that incentives, not intentions, decide outcomes — and box four removes every incentive that makes boxes one through three work.

A grid simple enough to draw on a napkin, and it still explains budget waste better than most policy debates manage in an hour.

[Defileo🔮 @defileo]

CLAUDE TREATS YOU LIKE A F*CKN STRANGER EVERY TIME.

mem0 fixes that and it just hit 62,900 stars / https://t.co/iJWyU4Bdb4

> Purely technical, zero token, the biggest player in this category
> Has its own benchmark, LoCoMo, other memory tools get measured against it
> Works with any AI agent, not locked to Claude only
> MCP server exists (coleam00/mcp-mem0) to wire straight into Claude Code

Here's how to use it ⬇️

> Install the SDK, pip install mem0ai
> Or grab the MCP server and connect it directly to Claude Code
> Your agent starts remembering facts, preferences, and past decisions on its own

This is the memory layer most other memory tools get benchmarked against.

[KalqiX @kalqix]

QuickSwap added trustless CLOB execution without becoming an exchange.

QuickSwap stays QuickSwap. KalqiX powers the execution.

[Yuki❄️ @YukiDailyVN]

X cập nhật ph���n 500k Tick

Hôm trước 140K

Sáng nay 184K

Hiện tại 181K

Đang cập nhật nên mọi thứ sai sai

[petek @Petek98]

@defileo rate the transition from 1-100. honestly, i'm giving it a solid 11

[TCL CSOT @TCL_CSOT]

ChinaJoy2026 Opening Day Highlights ✨

TCL CSOT showcases the latest innovations in gaming displays and AI-powered visual experiences, including the 27" IJP OLED Professional Monitor and the FHD 1000Hz Gaming DisplayChinaJoy2026 Opening Day Highlights ✨

TCL CSOT showcases the latest innovations in gaming displays and AI-powered visual experiences, including the 27" IJP OLED Professional Monitor and the FHD 1000Hz Gaming DisplayChinaJoy2026 Opening Day Highlights ✨

TCL CSOT showcases the latest innovations in gaming displays and AI-powered visual experiences, including the 27" IJP OLED Professional Monitor and the FHD 1000Hz Gaming DisplayChinaJoy2026 Opening Day Highlights ✨

TCL CSOT showcases the latest innovations in gaming displays and AI-powered visual experiences, including the 27" IJP OLED Professional Monitor and the FHD 1000Hz Gaming Display.

[rari @0xwhrrari]

OpenAI just dropped a free 3-hour course

On building complete agentic systems from scratch:

00:00 - Build your first workspace agent
37:42 - Master OpenAI AgentKit
1:24:01 - Add memory, context, loops, and evals
2:21:44 - Train agents with OpenAI's playbook

Most people are still building agents that forget everything between runs

OpenAI is already teaching the complete stack:

Agents → Tools → Memory → Loops → Evals → Training

Stateless agents are the old workflow

Systems that learn from every run are the new one

This free 3-hour course is worth more than most paid agent engineering programs

Bookmark and watch it today

Then learn how to build a complete team of AI agents below ↓

[petek @Petek98]

Card readers for kids #cardreader #learningtoys #educationtoys #createtowin #sarahverhoeven

[petek @Petek98]

@vovudebosh

[Jimin @Jiminiesbaby_3]

https://t.co/YUTtHd6S94
https://t.co/Vycrrt0wmv

[Macarena Moyano @Maca_Moya4]

https://t.co/U2X8ujMrcw
https://t.co/0mIMLfy5XS

[Medine yılmaz @Mediney29661923]

https://t.co/Qie4IqMDBJ
https://t.co/H44IrDLl0y

[Yyshdjdjdb @yyshdjdjdb]

https://t.co/zuYVRlG6dz
https://t.co/uLuE93vZGj