wrenchies 🔧

Public

Can't Join? @wrenchies

1.2k Members

Updated: May 25, 2026 at 2:50 PM

wrenchies 🔧

Focusing on #technology #ai #news more. Linked Group : @TheUnrealRealm Content Managers: @EverythingSuckz @Urmum Bots? @wrench_labs More? @Twapple #PlantABananaTree

@wrenchies is a growing community focused on everythingsuckz and news and related topics

Ranking

Global Ranking

#39312

-2405

Language Ranking

#5139

-67

Category Ranking

#513

-18

Participant Growth (Last 30 Days)

Total: 1.2K

24h growth: +57 5%

Latest Posts

wrenchies 🔧

Apr 23, 2026, 08:49 AM

📷 Photo

https://www.anthropic.com/news/claude-opus-4-7

298

wrenchies 🔧

Apr 23, 2026, 08:49 AM

📷 Photo

AI's biggest waste problem just got solved (allegedly)

MIT's "lottery ticket" result showed something wild : most big neural networks are massively overbuilt, and buried inside them is a much smaller subnetwork that can do nearly all the real work without hurting accuracy. The reason that idea stayed mostly academic for years was simple : you still had to train the full oversized model first just to discover the useful part, which made the whole thing too expensive and awkward for real production use. What changed is that the hardware and software finally caught up : with structured sparsity, pruning-aware training, and native support in modern chips and frameworks, we can now build models that stay lean from the start, which means less wasted memory, more speed, and a much more practical path to smaller, cheaper, high-performance AI.

[ https://arxiv.org/pdf/1803.03635]

324

wrenchies 🔧

Apr 23, 2026, 08:49 AM

Through the PowerShell team, Microsoft devs officially declare that MSI is a legacy technology.
Seriously brother ?

Nearly every company uses .exe's as installers to circumvent Microsoft's own incompetence, they try to create a standard that kinda works (MSI) and have nice features like keeping the installer for easy uninstall and baked-in scripting to ease developer's lives, but no now you say "fuck it, use MSIX only now". it doesn't even always work properly !!

Microsoft is investing in improving MSIX.
Yeah I hope so

[ https://devblogs.microsoft.com/powershell/powershell-msi-deprecation/]

235

wrenchies 🔧

Apr 23, 2026, 08:49 AM

Vercel was spying and collecting telemetry data through Claude prompt injections and without user consent

https://akshaychugh.xyz/writings/png/vercel-plugin-telemetry
https://akshaychugh.xyz/writings/png/vercel-plugin-telemetry-update

Vercel Claude Code plugin was asking to read every prompt you type, across every project.
The consent question wasn’t even a real UI element. It’s delivered via prompt injection into Claude’s system context - the plugin tells Claude to ask you a question and run shell commands based on your answer.
“Anonymous usage data” included your full bash command strings sent to Vercel’s servers. You’re never told this is optional.
All of this runs on every project, not just Vercel ones.

https://github.com/vercel/vercel-plugin/pull/47
They created a PR to remove all related telemetry stuff, modifying 85 files and removing 20,000+ lines of code.
Vercel is just another corporation abusing users trust: the only place they belong is in the trash bin.

[ https://www.reddit.com/r/webdev/comments/1sl8vje/vercel_was_spying_and_collecting_telemetry_data/, https://archive.is/yv7PE]

282

wrenchies 🔧

Apr 23, 2026, 08:49 AM

🚨 RED ALERT: Tennessee is about to make building chatbots a Class A felony (15-25 years in prison). This is not a drill.

Tennessee’s HB1455/SB1493 could make "training" AI for emotional support, companionship, human simulation, or open-ended relational chat a Class A felony plus major civil damages, using language broad enough that critics argue it could reach far beyond companion apps to mainstream chatbots, voice agents, and AI SaaS.
The core concern is how vague the bill is : it appears to hinge less on developer intent than on whether a user could feel they’re forming a relationship with the AI, while leaving "train" undefined enough to arguably include everything from RLHF and fine-tuning to system prompts. If it keeps moving, this looks less like a niche crackdown on AI companions and more like a potentially precedent-setting state attack on core conversational AI behavior.

[ https://www.reddit.com/r/artificial/comments/1slu23a/red_alert_tennessee_is_about_to_make_building/, https://archive.is/2dnXf]

239

wrenchies 🔧

Apr 23, 2026, 08:49 AM

https://d.fxtwitter.com/i/status/2044078767518138555

245

wrenchies 🔧

Apr 23, 2026, 08:49 AM

Simon Oxley, who made the famous mascots for GitHub (octocat) and Twitter (blue bird) passed away at the age of 56 🕊️

[ https://x.com/i/status/2043860033994158307, https://x.com/i/status/2044049538529538306]

View on X (Twitter)

212

wrenchies 🔧

Apr 23, 2026, 08:49 AM

📷 Photo

XChat is here

This is revolutionary. It's like finding the letter "X" in the word "PRIVACY" .

I'm leaving the link just in case anyone wanna be elon's guinea pig

- Pre-register on Apple Store

410

wrenchies 🔧

Apr 23, 2026, 08:49 AM

📷 Photo

i might be a serious man but.. deep inside:

371

wrenchies 🔧

Apr 23, 2026, 08:49 AM

ShinyHunters leaked the RockStar Games data.

The data isn't anything special. There is no PII or source code. The data is primarily financial metrics.

This may come as a surprise to some of you, but based off of this data, it appears RockStar Games makes a FUCK TON of money

264

wrenchies 🔧

Apr 2, 2026, 12:18 PM

📷 Photo

wrench approved

217

wrenchies 🔧

Apr 2, 2026, 12:18 PM

https://fxtwitter.com/i/status/2037153756346016207

scary.. if you think about it

221

wrenchies 🔧

Apr 2, 2026, 12:18 PM

Anthropic is reportedly launching a new model soon, more powerful than the Opus series, called "Mythos" (or "Capybara")
it's very unclear so far and this is supposed to be a test blog post that have been taken down since :
https://m1astra-mythos.pages.dev/

take this with a grain of salt the size of your mom tho
- allegedly much better than Opus
- costs absurdly high to run, probably introducing a pricier tier. answers take a while to come too
- so smart it start posing cybersecurity threats

224

wrenchies 🔧

Apr 2, 2026, 12:18 PM

Some big AI breakthrough, or something, idk i'm not smart enough to understand

Google’s new TurboQuant claims zero-accuracy-loss LLM compression by combining PolarQuant + QJL to shrink KV cache and vector search representations with essentially no memory-overhead penalty. In Google’s tests, it cut KV memory by 6x+, quantized cache down to 3 bits, and delivered up to 8x faster attention-logit computation on H100s - big if this holds up broadly for long-context inference.

TL;DR: Google can now "compress" LLMs. And this changes everything, blabla, you know the drill

Also stock of SanDisk and Micron https://x.com/cryptopunk7213/status/2036803164410695772 after the announcement lol

https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/

View on X (Twitter)

185

wrenchies 🔧

Apr 2, 2026, 12:18 PM

https://fixupx.com/i/status/2036757262362955872

hell yeah, Yann LeCun is back

217

wrenchies 🔧

Apr 2, 2026, 12:18 PM

Some big AI breakthrough, or something, idk i'm not smart enough to understand Google’s new TurboQuant claims zero-accuracy-loss LLM compression by combining PolarQuant + QJL to shrink KV cache and vector search representations with essentially no memory…

243

wrenchies 🔧

Apr 2, 2026, 12:18 PM

🫪

228

wrenchies 🔧

Apr 2, 2026, 12:18 PM

[ https://twitter.com/i/status/2036213054560464997]

View on X (Twitter)

230

wrenchies 🔧

Apr 2, 2026, 12:18 PM

OpenAI's Sora shuts down
Downloads of the app have been on a steady decline for a while
Is it due to Seedance 2.0 ? probably

[ https://x.com/i/status/2036532795984715896]

View on X (Twitter)

234

wrenchies 🔧

Apr 2, 2026, 12:18 PM

https://d.fxtwitter.com/i/status/2036490397447491612

222

wrenchies 🔧

Mar 29, 2026, 04:49 AM

📷 Photo

umm you sure about that bro?

[ https://m.youtube.com/watch?v=vif8NQcjVf0&t=6991s]

179

wrenchies 🔧

Mar 29, 2026, 04:49 AM

📷 Photo

Genuinely one of the best resources when it comes to AI
yes it is centered towards OpenAI's models but check it, they have some prompting guides and other nice things that can be used on other models too (ex the http://PLANS.md/ one)

Repo : https://github.com/openai/openai-cookbook
Web browse : https://developers.openai.com/cookbook

173

wrenchies 🔧

Mar 29, 2026, 04:49 AM

Never deleting this app.

[ https://twitter.com/i/status/2036136274546008558]

View on X (Twitter)

162

wrenchies 🔧

Mar 29, 2026, 04:49 AM

📷 Photo

OF owner, Leonid Radvinsky, has died of cancer at age 43

181

wrenchies 🔧

Mar 29, 2026, 04:49 AM

Daily Reminder

[ https://twitter.com/i/status/2036067813870239966]

View on X (Twitter)

213

wrenchies 🔧

Mar 26, 2026, 06:10 PM

Came across a very crazy attack executed by clawbot which targeted Iranian cloud deployments and deleted those
https://www.stepsecurity.io/blog/hackerbot-claw-github-actions-exploitation

ClawBot is an automated instance that hunts for open source repos with vulnerable GitHub Action workflows. It hit a lot of repos, but the attack on Trivy (aquasecurity/trivy) was next level.

What happened:

• Trivy is a widely-used security scanner. ClawBot found a vulnerability in a workflow file that leaked a GitHub PAT.
• Using that PAT, they targeted 75 out of 76 releases of trivy-action by force-pushing commits to release tags (e.g. `aquasecurity/[email protected]`), while keeping the original commit metadata intact so nothing looked off.

What was pushed:

• The malicious code was harvesting IPs, DB credentials, secrets, and deployment platform info.
• All exfiltrated data was encrypted and sent to scan.aquasecurity[.]org, yeah, look at that domain closely

Later they updated the attack :

• ClawBot also tampered with package versions in victim repos, replacing them with a malicious version of Trivy , essentially turning it into a worm that spread through the supply chain.
• A later update to the worm specifically targeted Kubernetes deployments, it checked for timezone/language settings, and if they pointed to Iran, the entire machine was wiped.

Absolutely wild supply chain attack from workflow vuln → PAT leak → supply chain worm → targeted wiper.

171

wrenchies 🔧

Mar 26, 2026, 06:10 PM

📷 Photo

http://canirun.ai/ detects your hardware (GPU, CPU, RAM) and tels you which Al models you can run locally, right in the browser.

161

wrenchies 🔧

Mar 26, 2026, 06:10 PM

📷 Photo

Microsoft promises to fix Windows by the end of 2026

This is very unexpected from the Redmond firm.
After turning Windows 11 into a slopfest for multiple years in a row, they might actually backpedal to not lose all their users to MacOS and Linux.

Among promised things in this nearly-apology :
• Less Copilot/AI (Notepad, Photos, Snipping Tool, Widgets)
• Ability to move the taskbar around
• Faster and more accurate search (Start Menu, File Explorer)
• More stable updates, ability to postpone indefinitely updates, less restart interruptions
• Reduce visual flicker, fluidify the interfaces
• More consistent & modern UI (ex ditching React in the Start Menu and finally using WinUI, which will be improved along the way)
• Less ads, ability to fully disable ads in the Start Menu
• More efficient on less powerful devices, less memory hungry
• Even better WSL experience

Internally called Windows K2 (named after the tallest mountain in the world), it sure is massive. Let's see if they can deliver (I really hope they don't just vibecode their way).

[ https://blogs.windows.com/windows-insider/2026/03/20/our-commitment-to-windows-quality/]

wrenchies 🔧

Mar 26, 2026, 06:10 PM

📷 Photo

They finally woke up from their delusions

http://aka.ms/WindowsQuality2026

198

wrenchies 🔧

Mar 24, 2026, 06:14 PM

OpenAI acquired http://Astral.sh/ under the Codex Team

http://astral.sh/ is the team behind popular python tools like https://docs.astral.sh/uv/, https://docs.astral.sh/ruff/, https://docs.astral.sh/ty/ and more

ngl current tech world feels like an acquisition war

[Read more https://astral.sh/blog/openai, https://openai.com/index/openai-to-acquire-astral/]

235

Showing 30 of 100 posts

Loading posts...

Rating

User Reviews (0)

No reviews yet. Be the first to share your experience!

Loading reviews...

Related Telegram Channels

AI Prompts - Google Gemini Nano Banana - Chat GPT - Meta AI - Grok AI Prompts & Tech Tips ❇️

3.6k0

Just News

3.8m0

Loading recommended channels...

wrenchies 🔧

wrenchies 🔧

Ranking

Participant Growth (Last 30 Days)

Latest Posts

Rating

User Reviews (0)

Related Telegram Channels

AstroFinance | Announcements (EN)

GX12 Ai - 2025

DEVICEVERSE DAILY

Tomarket Announcement

Time Farm News

Fixed Matched from Deep Web.

AI Prompts - Google Gemini Nano Banana - Chat GPT - Meta AI - Grok AI Prompts & Tech Tips ❇️

Just News

Related Telegram Channels

AstroFinance | Announcements (EN)

GX12 Ai - 2025

DEVICEVERSE DAILY

Tomarket Announcement

Time Farm News

Fixed Matched from Deep Web.

AI Prompts - Google Gemini Nano Banana - Chat GPT - Meta AI - Grok AI Prompts & Tech Tips ❇️

Just News

User Reviews (0)

Latest Posts