Asif Masood

301 posts

Asif Masood

@A51F221B

Application Security Engineer

Islamabad, Pakistan Katılım Ekim 2016

1.3K Takip Edilen329 Takipçiler

Asif Masood@A51F221B·1d

I've seen so many people comparing AI to the industrial revolution, but I think it's a false equivalence. AI can’t be compared to the industrial revolution or anything else; this is the first time in human history a technology can match or exceed human intelligence.

English

Asif Masood@A51F221B·1d

So excited to watch Real Madrid this season

English

Asif Masood@A51F221B·3d

It's embarrassing to admit, but I found out today that by default Claude Code deletes chats after 30 days of inactivity. That's why I sometimes couldn’t find chats in the terminal, which I thought was just a UI bug messing with the chat history.

English

Asif Masood@A51F221B·3d

I've been using Claude Code in the terminal for so long, but I've finally seen the light and switched to the desktop app, which is way more convenient in every way.

English

Asif Masood@A51F221B·5d

@Tur24Tur looking forward to your results !

English

286

Tur.js@Tur24Tur·5d

My account was whitelisted today, charged $111 and got $22 bonus credits on top. 200 concurrency, 5K RPM, 3M TPM, no daily cap. More than enough to stress test K3 for cybersecurity work.

Tur.js@Tur24Tur

I'm still hitting rate limits when using Kimi K3 with Claude Code Harness. I tried using it to find an XSS vulnerability in a vulnerable web application, but it couldn't find one and kept getting stuck when using its own CLI harness. Even with my own harness, I'm still running into rate limits. I still don't understand the hype around Kimi K3. So far, I think the best Chinese model is DeepSeek V4 Pro. I'm sure the team is cooking something big, so I'm looking forward to seeing what they release next.

English

11.2K

Asif Masood@A51F221B·7 Tem

I think it's going to be a France vs England final.

English

Asif Masood@A51F221B·26 Haz

Okay, I'll just say it: using ChatGPT/Codex feels like using a Samsung, while using Claude/Code feels like using an iPhone.

English

Asif Masood@A51F221B·3 Haz

Yeah i take my words back, the patch that they applied is even more hilarious, i don’t think anyone from security team is even involved because how else would you explain this😂😂

English

Asif Masood@A51F221B·3 Haz

The Meta AI support assistant account takeover is hilarious. How could their security team have missed this? But I also empathize with them—working at such a huge scale and getting everything absolutely right is not always easy.

English

Asif Masood@A51F221B·28 May

So, the new Siri is basically a harness that can be plugged into any model.

English

Asif Masood@A51F221B·28 May

The AI race is turning more and more into one about price rather than which model is SOTA. If Chinese models are even 80-85% of what SOTA models (Claude, GPT 5.5) are at 1/10th the cost, users will still switch to them because of the cost.

English

Asif Masood@A51F221B·28 May

@Tur24Tur @XiaomiMiMo its a really cool idea to benchmark all the models yourself in this way. I would be really interested in the final results on which model performed better on these tasks.

English

271

Tur.js@Tur24Tur·28 May

@A51F221B @XiaomiMiMo Currently running an experiment gave MiMo, deepdeek V4 pro, and claude opus the same web challenge with multiple chained vulnerabilities multiple steps required to reach the flag First model to submit it wins.

English

371

Tur.js@Tur24Tur·28 May

Timeline kept pushing @XiaomiMiMo at me. Ignored it. Kept ignoring. Finally clicked. Selected the Lite plan: $4.62, 4.1B credits. Model: MiMo-V2.5-Pro. Claude Code setup is two lines: ANTHROPIC_BASE_URL=token-plan-sgp.xiaomimimo.com/anthropic and ANTHROPIC_AUTH_TOKEN=your-key. Tested it on real security tasks. No restrictions, got the job done. Sometimes the algorithm is actually right.

Xiaomi MiMo@XiaomiMiMo

🚀 Better inference efficiency, lower costs, broader access. MiMo-V2.5 Series API pricing is now permanently reduced — by up to 99% compared to previous pricing. ✨ Unified pricing across all context lengths. MiMo Token Plans have also been upgraded: • 5–8× more usable tokens at the same price • Simpler and more transparent billing rules 🎁 As a thank-you to current users, all current Token Plan credits will be fully reset. 🎧 MiMo-V2.5-TTS remains free for a limited time. ⏰ Effective May 26 at 6:00 PM PDT. These improvements are powered by continued inference optimization and serving efficiency upgrades across the MiMo stack. 🛠️ We’ll also publish a detailed technical blog on the inference optimizations later — stay tuned.

English

458

52.4K

Asif Masood@A51F221B·27 May

I read somewhere today that according to quantum physics, decisions that happen in the future are affecting your present. I can’t stop thinking about it.

English

Asif Masood@A51F221B·24 May

Why are there so many tire lockups happening in this F1 season

English

Asif Masood@A51F221B·24 May

For additional context, I am talking about creative writing or writing about any large topic in general. LLMs are still really bad at that (bulk writing), at least in my experience.

English

Asif Masood@A51F221B·24 May

For a technology that's entirely based on the generation of text or prediction of the next token, LLMs are surprisingly bad at it even now. I am talking about current SOTA models like GPT 5.5 and OPUS 4.7

English

Asif Masood@A51F221B·20 May

José Mourinho is the man to do it.

Asif Masood@A51F221B·17 May

@bryan_johnson inventing religion from first principles

English

Bryan Johnson@bryan_johnson·17 May

I've been praying the past few weeks. Unsure why. There's good evidence behind prayer. It mimics breathwork, calming the nervous system, dropping cortisol, and quieting the brain. Daily prayers show lower depression, anxiety, and pain. I'd like to develop a prayer practice. Growing up, the protocol was written for me. Explaining whom to pray to, the structure of the prayer, and the boundary conditions. I don't really know how to pray now.

English

1.9K

341

10.6K

1.3M

Asif Masood@A51F221B·17 May

SAST/DAST vendors may be lagging right now, but once they ship AI (and most are actively building it), they’ll likely outperform raw LLMs like Claude or Codex. The edge? A much stronger, security-first harness around the model.

English

Keşfet

@Tur24Tur @XiaomiMiMo @elonmusk @BarackObama @taylorswift13 @cristiano @BillGates @NASA