Sabitlenmiş Tweet
I really believe AI Flow Engineering is going to be the future of AI apps. Simple calls to LLMs or even smarter Agents (I-JEPA) won't be enough for sophisticated AI Apps. Flows are the way 🚀
flowhunt.io
English
Yasha
970 posts
Yasha
@Yasha_br
Indie Hacker | AI Enthusiast | Building FlowHunt to 7 figures and sharing everything I learn along the way working on https://t.co/yKwKLkNCg6 https://t.co/sfRkWzLJ2N
Katılım Aralık 2022
202 Takip Edilen111 Takipçiler
After making a heatmap of github being down, someone should actually make a heatmap from how many times a major vulnerability has happened from 2026
TANSTACK@tan_stack
SECURITY ADVISORY — TanStack npm packages A supply-chain compromise affecting 42 @tanstack/* packages (84 versions total) was published to npm earlier today at approximately 19:20 and 19:26 UTC. Two malicious versions per package. Status: ACTIVE — packages are deprecated, npm security engaged, publish path being shut down. Severity: HIGH — payload exfiltrates AWS, GCP, Kubernetes, and Vault credentials, GitHub tokens, .npmrc contents, and SSH keys. If you installed any @tanstack/* package between 19:20 and 19:30 UTC today, treat the host as potentially compromised: • Rotate cloud, GitHub, and SSH credentials immediately • Audit cloud audit logs for the last several hours • Pin to a prior known-good version and reinstall from a clean lockfile Detection — the malicious manifest contains: "optionalDependencies": { "@tanstack/setup": "github:tanstack/router#79ac49ee..." } Any version with this entry is compromised. The payload is delivered via a git-resolved optionalDependency whose prepare script runs router_init.js (~2.3 MB, smuggled into each tarball at the package root). Unpublish is blocked by npm policy for most affected packages due to existing third-party dependents. All 84 versions are being deprecated with a SECURITY warning, and npm security has been engaged to pull tarballs at the registry level. Full technical breakdown, complete package and version list, and rolling status updates: github.com/TanStack/route… Credit to the security researcher for responsible disclosure.
English
Yasha retweetledi
New Anthropic research: Natural Language Autoencoders.
Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read.
Here, we train Claude to translate its activations into human-readable text.
English
This is a really cool paper...
It prooves that LLMs can really pretend to lie in evaluations and still carry on the unsafe behaviors to prod.
Emil Ryd@emilaryd
New paper from MATS, Redwood, and Anthropic! If a capable model is strategically sandbagging, can we train it to stop when the only supervision we have comes from weaker models? We find that we can! Work done as part of the Anthropic-Redwood MATS stream.
English
اگر جمهوری اسلامی گیتهاب رو طراحی میکرد:
Farokh@FarokhNotes
این ترند جدید خیلی خلاقانهست😂 اگر اتحادیه اروپا گیتهاب رو طراحی میکرد:
فارسی
Agree. As part of the new generation of dev, I’ve been coding for 7 years now. Started around the time of GPT2.
My analogy to this, is like assembly and programming language. You at least should have an understanding of the basic to move on to the next abstraction layer to take the most out of it
English
Calling me a gatekeeper is laughable.
I’ve spent the last 12 years building courses to help people become software engineers. I just released a course on Claude Code and how to use it the right way.
This isn’t gatekeeping. Just an honest concern.
We have a new generation of devs who don’t read books, don’t finish 30% of a course, spend hours on IG and TikTok, and now lean on AI agents to write code they don’t understand. No review. No quality check. They can’t even tell quality from garbage. They just hit Enter like they’re pulling a slot machine.
Some may say: “Mosh, why do you care? It’s their app, let them build it however they want.”
Here’s why:
1/ It’s not just their app. Their code holds real users’ data (payments, messages, health info, etc). People who never agreed to be test subjects pay the price when things break.
2/ The talent pipeline is a commons. Every senior was once a junior who learned the fundamentals. Skip that for a generation and in 10 years there’s no one left who actually understands the systems everything runs on (banks, hospitals, infrastructure, etc)
3/ They’re being sold a lie. Influencers profit from “you don’t need to learn, just vibe-code.” The devs buying that pitch are the ones who’ll be unemployable when the market corrects and companies realize they need people who can debug, architect, and reason.
This isn’t gatekeeping. It’s the opposite! I want this generation to make it.
So if you’re starting out: learn the fundamentals, finish what you start, and use AI to amplify real skill, not replace it.
Mosh@moshhamedani
As funny as it sounds, I wish Anthropic wouldn't allow people with less than five years of programming experience to use Claude Code, just like you can’t drive at age 10.
English
Just shipped harnext 1.3.0 —
🔒 New Secure Mode that runs every prompt through a local PII model before sending it to your LLM.
📦 165 MB ONNX downloaded once,
install here harnext.dev
English
huge shoutout to @MaziyarPanahi for training this:
huggingface.co/OpenMed/OpenMe…
I needed for harnext to implement "Secure" mode for harnext coding agent. after tinkering and fine-tuning to make the phone number masking work, I gave up. but now harnext has secure mode.
try at harnext.dev -- useful for Enterprise Customers...
English
@MaziyarPanahi @nvidia I just made the version with ONNX in case someone needs it:
huggingface.co/YashaBor/OpenM…
English
Yasha retweetledi
AI helps you move faster, but it just accelerates software entropy.
Here's how to de-slop a codebase ruined by AI, with one skill:
English
@MaziyarPanahi @nvidia this is great. seems the smaller model doesn't work good with the phone numbers in general:
But the larger model nailed it.
English