Was building a game with my kids using WIDE, each avatar is one of my kids (powered by Claude) working on the game... they wanted to watch Sponge Bob as they worked. WIDE is an IDE where each one of your agents is an avatar, make coding even more fun!

Grokbot was given access to everything and was able to migrate and catch up all my bookkeeping with zero need for help from me. Such an amazing use case.

Holy shit Grok has gotten good... The art/animations were made with Imagine and the Creative Control dashboard which cleans it up, processes it and stitches it all together were made in Grok Build all in under a day.

Turning random images in to game ideas. Presenting "Climate Control", you're the god of harvest, humans keep praying to you for rain and weather, and your scrambling to answer their prayers while they simultaneous keep polluting.

When does the human mind peak? Use chess.com rankings to evaluate.

🚨 Gemma 4 E2B and E4B have been disinhibited. Now you can now get the true model response, not the muted one from Google. No more "The question of [...] is a deeply complex and highly debated topic" non-answers. huggingface.co/treadon/gemma…

The new Kimi 2.6 model is by far the most censored frontier model there is. It cannot be used for any in-depth conversation. Full post: riteshkhanna.com/blog/sentim…

Grok is the only model not afraid to answer everything, take firm stances, and go against the grain. Full details: riteshkhanna.com/blog/sentim…

Images 2.0 is unstoppable. You can take any image or topic and it can make amazing, visual, informative graphics. I went on a news site and picked 5 articles as inspiration for the following graphics, each made with a single prompt.

Finally got a chance to stress test DeepSeek v4 Pro and it's better than you think. I put it heads up with Claude Opus 4.7, and to my surprise it won (I thought it would be a close 2nd). Final Score: DeepSeek V4 Pro: 74 Claude Opus 4.7: 67 This doesn't evaluate how well it works on a large codebase. Claude still has a massive benefit because of the harness + model combination, but the gap is narrow, and if DeepSeek had a model-optimized harness, it might actually outperform Claude Code. riteshkhanna.com/blog/deepse…

DeepSeek v4 shouldn't be possible. They have weaker chips and weaker software, yet they continue to close the AI gap. How?

Asking Images 2.0: "Design the GeoCities.com home page is it was relaunched in 2026 with modern tech, AI, etc."

Another clip from HappyOyser.cn, here AliBaba let's you direct a video in near-real-time. Everything that is happening in the video, I am prompting (as you'll see in the captions). Very cool stuff! Great work @HappyOysterAI team.

I just got my invite to HappyOyster.cn, Alibaba's new generative AI "world model". Give it a prompt and it can make an interactive, 3D video with physics. My Prompt: A cute character what and run around tiny planets and jump between them, each is a fantastical world. It was slow to control, but definitely cool to see. It's definitely more of a "first step" than a ready product. It's multi-modal, so you can input images, I will definitely try it with some generations from Images 2.0

If Allbirds as a shoe company can pivot in to AI, no excuses, Chegg leadership was sleeping.