Akash Srinivasan

Microsoft just fired CLAUDE CODE 😂 Because Claude was becoming more expensive than EMPLOYEES in their company !! Cheaper AI Models with good performance might dominate the future of AI. Or will "humans" replace Claude 😅 🥹 Check out the thread 👇👇👇

WBench shows that every AI Model breaks in different ways But most of them fail in NAVIGATION: → Not in simple prompts → but Across 2-9 left/rights turns deep (not just single commands)

Watch the below video 3 times.. The more you watch, the more FAKE it looks: Because the problem is that: Most benchmarks only test video QUALITY But this new AI Video benchmark checks → Navigation consistency → PHYSICS rules → Fight scenes, explosion & collision interaction

Every AI video model you've tested is probably failing in ways you can't even measure yet WBench just exposed the hidden weaknesses in AI Video Gen models like: → Google Veo 3.1 → Kling 3.0 → ByteDance Seedance 2.0 And the results are... surprising 👇 👇

@heyrobinai Most people feel that Flash models like Step 3.7 can't code But I believe it's just a skill-issue 😅 When will people learn prompt engineering !!

EVERYONE KNOWS FLASH MODELS CAN'T CODE.. so I gave one a real job. not a leetcode problem, an actual production landing page. full hero section, animated sound-wave background, pricing tiers, scroll reveals, the works. one prompt, first pass, no edits. I ran Step 3.7 Flash through Claude Code, same harness I use every day. typed the brief, walked away, came back to a finished page that actually rendered. what stood out: > first-pass output ran clean. it read the existing project conventions instead of dumping a generic SaaS template > it didn't choke on the visual stuff, animated waveforms, parallax depth, custom palette, the things flash models usually fumble > it was fast. up to 400 tokens/s, and I felt it the benchmark backs up what I saw: 56.3 on SWE-Bench Pro, first pass. but watching it ship a real page in one go landed harder than any score. 198B total / ~11B active. 256K context. Apache 2.0, runs locally, plugs into Claude Code, MCP, and Skills with no rewiring. if you build agents or coding workflows, this is worth running on something you'd actually ship. try it: platform.stepfun.ai @StepFun_ai

@sabir_huss50540 These platforms are great to find remote jobs but combining it with an AI auto-applier will make it them even better.

STOP WASTING TIME ON LINKEDIN IN 2026. STOP WASTING TIME ON UPWORK IN 2026. STOP WASTING TIME ON INDEED IN 2026. Remote jobs paying $95/hr are hiring right now. Most people don't even know these sites exist. Here are 12 platforms actively hiring today:

🚨 If your AI videos still look cringe, you are using the wrong tool Because I have used an AI Agent to create a $8,000 cinematic commercial !! Follow me + comment “TOOL” I’ll DM you the AI tool.

@heyshrutimishra He's not picking the next NVIDIA. He's picking the companies that own mobile.

Jensen called Marvell a $1T company. Now it’s.. Qualcomm

Here is a step-by-step tutorial 👇 👇 Try Now: hyper3d.ai #Hitem3D #AI3D #Imageto3D @Hitem3D

It just takes 20 seconds to generate model Which can be directly pushed into a production pipeline. Because Rodin Gen 2.5 has a complete 360 degree texture coverage without any blind spots. With a more faithful PBR materials

This AI helped me earn 200$ from upwork !! And I just spent 15 minutes A Client on upwork wanted a 3d model replica of an image for his game All I did was upload the image to RODIN from Hype3d and it handled the rest !! The client even payed me a bonus for fast delivery

@DBuniatyan Hivemind is just GitHub but instead of code it stores experience. GitHub is used by developers & Hivemind is used by AI Agents How ironic...

Coding agents that actually get better the more your team uses them. Introducing Hivemind: continual learning for AI coding agents. Hivemind turns the traces from every agent your team runs (Claude Code, Codex, Cursor, Hermes, OpenClaw, Pi) into reusable skills, then pushes those skills across all of them. All on your cloud storage. Now with SkillOpt built in, your skills get trained: +19.1 points of accuracy in Claude Code, +24.8 in Codex, best or tied on all 52 setups tested. Open source, one line install.

PewDiePie launched his own ChatGPT competitor. After testing it for 30 minutes... I genuinely can't understand why people are hyping this up. As most of creative tools are broken ! But I do love the fact that we could run open-source AI models locally in our machine for FREE

@heyrobinai I wonder if blender would integrate AI into their software just like VS Code did....

you don't need to learn 3D modeling anymore ai just made years of skill-building completely optional.. tripo converts ideas into finished 3d models faster than most people can open blender Tutorial below:

@LearnWithBishal The next big cybersecurity risk isn't your phone. It's AI agents with access to your phone. @LearnWithBishal Can you please share any security settings for that ?

Most people think the worst thing that can happen to an iPhone is losing it. Not anymore. The real risk is someone getting access to your Apple ID. Because that means: Your photos. Your passwords. Your messages. Your backups. Your entire digital life. Here are 8 iPhone security settings I'd enable today: 🧵

99% of founders are overpaying video editors for marketing content. This AI tool edits ads, reels, and product videos from a single prompt. Comment "VIDEO" and I'll send it 👇

Even among open-source models, SkyClaw-v1.0 is by far the best in terms of performance & affordability. SkyClaw-v1.0 just surpassed AI models like: → DeepSeek V4 Flash → Minimax 2.7 → Qwen 3.6 35B

If you are a "Vibe Coder" looking for a cheaper alternative to Claude Code. I would suggest you to try "SkyClaw-v1.0" With OpenClaw as the UI Interface Because it is literally 5x CHEAPER than Claude Opus 4.6

Microsoft just fired CLAUDE CODE 😂 Because Claude was becoming more expensive than EMPLOYEES in their company !! Cheaper AI Models might dominate the future of AI. Or will "humans" replace Claude 👇

If your AI coding budget was capped at $200, what's your pick ? Cursor, Trae, Claude Code, or Codex ?