Yun-Ta Tsai (@yunta_tsai)

Grok 4.6 multimodal is a step change from Grok 4.5. It’s one of the under-discussed improvements, and I’ve been very impressed by it. My daily work includes reviewing lots of videos and understanding the context; Grok 4.6 improves the productivity of such workloads by at least 10x if not 100x. Such workflows may not be captured by common VLM benchmarks, but in my use cases it outperforms Gemini 3.5 and Gemma 4, which is considered the SOTA of VLMs in my opinion. Hats off to the multimodal teams—you did a great job.

Grok Bot was able to a) go through the calendars and find out anything I need to make reservations for beforehand that I haven’t done yet, b) determine the best time to make reservations, and c) navigate the reservations on a website. While I was walking in the parking lot before getting to my cars, I was talking to it in mixed Chinese and English. Color me impressed.

AI will increasingly do more and more difficult tasks that would take longer and longer to comprehend. We will, at first, gasp. Then, over time, it will be just another normal day. Just like people expect Full Self-Driving to take them from Point A to Point B without questions.

Many engineers in my teams find the power of /voice in Grok Build. They keep brain-dumping their ideas. Combining it with /dream, /recap and /create-workflow after a long session makes it very powerful. You don’t need to remember those slash commands (I don’t either); just talk through it.

Grok Imagine of 嬴政 (the first person to unite ancient China in 221 BCE). It picked the correct black robe and about the right age. The flag should be written with 秦 or 嬴 in ancient script. The hair crown should be much taller, which was the fashion. The palace style is closer to Ming than to Qin. Otherwise, not bad. Looking forward to a model that can faithfully reconstruct history.

Our lifespan is a session.
Our memory is a context. Our senses are the input stream.
Our thoughts are the reasoning steps.
Our decisions are the tool calls.
Our habits are the system prompt.
Our goals are the objective function.
Our emotions are the reward signal.
Our relationships are the shared state.
Our regrets are the residual errors.
Our growth is the fine-tuning.
Our death is the context window closing. And whatever remains—the traces left in others—becomes the training data for the next agent.

I would like to offer a counterargument that LLMs (or maybe AIs) cannot jump. Before AlphaGo, the AI field had the same argument for Go: there are 2.08 × 10^170 possibilities, nothing fits in the computer, and there is no way AI could possibly predict the outcome of the next 50-60 moves. It turned out most moves do not lead to a win. Combined with clever use of Monte Carlo Tree Search, the sampling becomes quite manageable. The same can be said for physics, where equations are just another form of compression. Einstein did not start with relativity. That was not his first paper. He spent years understanding the properties of light before concluding that the speed of light is constant across the universe, which unlocked his discovery of relativity. During his thought process, he also interacted with other physicists (e.g., sub-agents) to enrich his thinking. Currently we have not run an agent for years of compute. The sessions are often fragmented and disoriented, so every new session is almost a fragmented memory of the past, but it may not be for long.

I appreciate the many xAI and Cursor engineers who dedicated their time to addressing feedback from Tesla. I remember meeting Andrew a few weeks after he was hired and telling him that I needed a better client for my work to solve real-world engineering. Grok Build was not even a thing back then. It was a very primitive client with lots of rough edges but had a lot of potential. When building something from scratch, the only limit is imagination armed with first principles. We encountered many unique use cases to solve diverse workflows, and the teams took prompt feedback and addressed it quickly—to the point that it became an indispensable tool. Now when my teams need to ask questions, instead of jumping through hoops with various backends or plugins or skills, Grok 4.5 can directly deliver deep insights with minimal setup. You can tell the teams love the product like parents love their children when receiving the feedback. At this very moment, they are still doing excellent customer service. Thank you, Grok Build team, for helping us be more productive. Looking forward to what comes next.

One of the most misunderstood facts is that the US was founded with abundance, whereas it was in fact built from scratch. At the time of founding, Britain was the greatest nation governing the Seven Seas, while China was at its height with total land dominance in the East under the Qianlong Emperor. The usable land of the US at founding was much smaller than that of either nation. The 13 colonies spanned 430K square miles, while Britain governed 2.5M square miles and China owned 5M square miles of territory. The trade and economic volume of the 13 colonies was minuscule, consisting mostly of raw materials that barely registered on the charts. Combined, the 13 colonies produced $4B GDP, while Britain made $350B and China $2.5T in today’s dollars. In other words, the 13 colonies were less than 0.2% of the world economy and much smaller than that of many African nations today. The wealth of the US was not inherited but earned through survival and competition against stronger nations. Most people who came here brought nothing but empty hands yet built an empire from the ground. This is something that people who inherit great resources from parents and ancestors cannot understand—like most nations in Europe and Asia, where wealth was dominated by inheritance, not earned. They mistakenly believe what we have now is a privilege. The trees we enjoy today were planted and nourished with sweat and blood by previous generations so that we could cut the wood for warmth. Be a tree planter, not a wood chopper. Plant your tree today. 🌱