AI & Tech Developments - Feb 17

If you're hosting your @openclaw on a Hetzner VPS and you're non technical.. you should "harden" your VPS.. basically secure your server so that only YOU can access it from your computer.. you can prompt your openclaw to do a security audit of the server its hosted on. It

— @ayushtweetshere

Big moment. Pine AI just released Pine Voice on OpenClaw: PineClaw Your OpenClaw agent can now make real phone calls. Not browser audio. Not a chatbot pretending to talk. Actual calls over real telephony infrastructure. Low-latency. Human-like. Smart enough to crush phone

— @PineAiCallAgent

On va en profiter pour vous donner le bon plan du moment (Nym est "Boites Noires Proof" : contrairement aux VPN classiques, il préserve également vos métadonnées, seules choses que les "boites noires" savent intercepter en clair). https://nym.com/fr Et bonne nouvelle il y a

— @bluetouff

For all my lifters: computer vision app to measure back curvature during deadlift! main technical highlights: — RF-DETR (Roboflow) to segment the person (great performance out-the-box with no additional training!) — YOLO11n (Ultralytics) for bounding box prediction around the

— @IlirAliu_

The bull case for orchestrator.

— @rareZuhair

wow this new claude code feature is pretty cool

— @banteg

five ways your openclaw agent will shoot you in the foot. 1) drift your agent will write to agents-md, memory-md, local disk, remote databases. wherever seems convenient in the moment. if you're not watching, it scatters state everywhere and then can't find it again. you have to

— @chrysb

claude now builds itself a second brain arscontexta is a claude code plugin that generates a complete knowledge system from conversation

— @arscontexta

OpenAI and Anthropic are earning $11-14B a year while burning $100B's and being unprofitable, while Apple just licensed Gemini for $1B/year and stands to increase $20-80B on hardware sales if they can get their AI products right. Then, they'll release M5 chips and customers will

— @jbrukh

How can I put @openclaw in a loop so it is never idle? like proactively continuously working towards a goal. Mine just stops and waits for me to tell it to go to work again. As if it was lazy.

— @QuantRob

Meet Huihui-gpt-oss-20b: an uncensored, conversational AI that's making waves. This open-source text generation model is built for raw, unrestricted dialogue. Perfect for developers who want authentic AI interactions without guardrails.

— @HuggingModels

Supermemory makes OpenClaw's memory 2x better on avg, by: - Building a knowledge graph and remembering details using hybrid mode - Using hooks, instead of tools, to inject context - Forgetfulness and temporal knowledge built in using the supermemory engine setup in one line.

— @supermemory

18 Nano Banana Pro prompts that mass-produce stunning 3D visuals Glass FX, X-rays, Holograms, 3D Metallic, Submerged effects and more... each one is a different style you can run on any object or product Grab the full collection below

— @lloydcreates

"Replacing a $10,000 degree" is the usual marketing fluff. Learning to call the Claude API is like learning to use a library—it's the bare minimum, not the ceiling. The real gap in 2026 isn't knowing how to use MCP; it's understanding how to manage the KV cache and prefix

— @GenAI_is_real

Lots of great updates in QMD. Big performance upgrade and node + bun support both.

— @tobi

Seedance 2.0 is finally FREE. The China Model that just cooked Hollywood is now open. We are talking about the tool that is creating 1080p, CGI-level video so realistic. You don't need a credit card or even a sign up!

— @MustafyOf

Rizz is legit the only model we use for AI UGC. Has the best lip syncing and takes minutes compared to others u have to train for ages

— @Camicees

Setting up OpenClaw with a multi-agent "team" has been... a process. Question after question to figure out. Some technical, some strategic. I detailed it all in today's video.

— @CasJam

"How much are you paying for AI with OpenClaw?" This is the most common question I get from people, and the answer is "not much" despite running multiple instances and being a power user of it. This guide is how I keep my costs down

— @KSimback

Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people.

— @riomadeit

Your AI agents can now master entire tech stacks, not just single tools. Introducing batch mode for HyperSkill. Auto-generate dozens of SKILL.md files at once from live docs.

— @hyperbrowser

so you boot claude code and you use the heygen api for UGC ads. 11 labs API for voices. then you have claude code scrape reddit for pain points and desired outcomes your app can solve. then you have it write scripts based on this. then you create UGC ads using heygen API + 11

— @codyschneiderxx

Study finds that skills written by the LLM itself do not increase task completion rate. Definitely matches my own experience. LLMs are noisy amplifiers: when you ask them to amplify themselves, you just get more noise.

— @nateberkopec

I've been struggling with the OpenClaw setup. Burned over $1,000 on tokens in just 3 days. Then this guy saved me weeks of time and serious money.

— @gabrelyanov

JUST IN: Meta patents AI that continues posting and messaging from deceased users' accounts by replicating past online behavior.

— @WatcherGuru

OpenClaw for ___ Is about to be the new Cursor for ___

— @NickADobos

Ils sont tellement dans le mensonge qu'ils paniquent pour une vidéo qu'ils arrivent à faire supprimer en qqes heures.

— @revelateur_de_C

codex consistently finds 7 of the top 5 bugs in claude code's design docs.

— @corbtt

Made a new robot for my site Hoodmaps.com that uses its data to find good and bad areas and mark them. It collects: crowdsourced drawing, crowdsourced tags data, positive places data like lots of trees,

— @levelsio

You don't need a "Longterm Memory MCP", just create a LEARNING md file in your project and put this in CLAUDE md: The LEARNING.md file contains insights, tips, and best practices discovered during previous tasks. Read this file before running new tasks to

— @tdinh_me

When I first switched to M1, back when native application support for Apple Silicon was mixed, I discovered my terminal had been running in x64 mode for 4 months. Switched to the arm64 build and Node.js became 2x faster

— @DanielLockyer

Tell me everything you know about how you force AI to design beautiful designs

— @Austen

Security tip: if you're running OpenClaw on a Hetzner VPS, you can prompt it to audit its own server security.

— @ayushtweetshere

GitHub was built for humans who write 50 lines of code a day, not agents that commit 600 times in a weekend.

— @GenAI_is_real

kinda ironic how the average work day is getting harder, as we progressively automate the easier parts of our daily jobs

— @giansegato

Figma meets Claude Code! Pencil is an infinite design canvas that runs parallel design agents locally.

— @dr_cintas

Y'a des gens ils vont voir ça et être dans le déni en mode "pfff y'a pas de saveur y'a pas d'âme notre métier est safe"

— @fakenine_

I built a wrapper around Claude Code and Codex that uses my subscription!

— @_StanGirard

honestly crazy that openclaw sold for $1B... vibecode an open-source project and make $40B in a couple months now

— @willccbb

anthropic just released an update to claude code, and it hide the names of files the tool was reading, writing, or editing.

— @forloopcodes

This is such an insane change in just 12 months. The future is going to be absolutely insane.

— @Pentosh1

This aired tonight to 1 billion people in China. A year ago these robots could barely wave a handkerchief, now they can do backflips and kung fu.

— @Tristan0x

ALIBABA JUST OPEN-SOURCED THE MOST POWERFUL MODEL IN THE WORLD - Qwen3.5 plus

— @RAVIKUMARSAHU78

Finally trying 4x Claude Code's on server in one @TermiusHQ, really nice and you don't have to fiddle with shortcuts

— @levelsio

what's the best agent browser skill that can use my actual chrome with logged in creds?

— @iannuttall

You can run production-grade LLM inference on a phone or laptop CPU

— @paulabartabajo_

I lost 3 hours yesterday. Started fixing a bug. Ended up refactoring auth, adding dark mode

— @_StanGirard

Kanban board for Claude Code tasks

— @tom_doerr

You can also make an UNCENSORED AI AGENT by simply accessing Venice's API and choosing the GLM 4.7 Flash Heretic model

— @ErikVoorhees

Study OpenClaw diagrams to learn some things you probably didn't know

— @kloss_xyz

I cannot tell you how much browser and auth hell Matt has helped me avoid in OpenClaw

— @Austen

So glad to see that the woman who is in charge of making AI moral has sane and rational beliefs like we have to kill all predator animals

— @0x49fa98

I'm a 42 year old former filmmaker. I made this in 3 days with $39 in credits

— @ryanlightbourn

anet -- a CLI that combines onchain primitives into one interface

— @MurrLincoln

They are now 3D printing a 12 meter boat in one piece with robots

— @TansuYegen

IL A ALIMENTÉ SA MAISON PENDANT 8 ANS AVEC 1 000 VIEILLES BATTERIES D'ORDINATEUR PORTABLE RECYCLÉES

— @Cultination1