Which big model is the best to use? The most popular "AI colleague" on X is Claude. Many people say it is currently the most honest and useful, not flattering, invincible in code, skilled in deep reasoning, and capable of writing in long contexts. In addition, Gemini is the "king of multimodality+research artifact", Grok is the "real-time information+straight man AI", and GPT is the "all-around player+daily stability". Above is also the order in which I use these four models on a daily basis. Claude and Gemini have both subscribed to higher-level services. My experience is that using the best large models directly to solve practical problems is more effective than tinkering with various tools. Of course, lobsters and other animals have their own uses. Claude, the helmsman Dario, recently said: 1. Intelligence is an industrializable chemical reaction: data x computing power x model scale, proportional input, intelligent stable output. To put it simply, the big model is still being developed. 2. AI is close to the level of human intelligence, but most people underestimate it. 3. The black box era is coming to an end: interpretability can now scan the neural circuits inside models like MRI. 4. Human knowledge is almost finished, and AI is generating dynamic data through self trial and error, and evolving on its own. It's a bit like Alpha Zero back then. 5. Amdahl's Law: AI accelerates efficiency, but the remaining "bottleneck links that cannot be accelerated" - judgment, cross domain integration, and physical world perception - actually skyrocket in value. 6. It is not impossible for humanity as a whole to become stupid in the future, it is a matter of choice. 7. The paradox of radiologists: AI has surpassed humans in image reading, but the number of radiologists has not decreased - the most technical aspects have been replaced, but the value of the "translation layer" between humans and technology remains. 8. Even if humans only retain 5% of the task and AI helps you complete the other 95%, your overall efficiency can still be improved by 20 times - this "area where humans still have a place" is larger than most people imagine. 9. You shouldn't just be a 'shell wrapped' thing - even if it's not a big model to eat, it will be easily eaten by anyone else. 10. The big model will ultimately be recognized as possessing some kind of "consciousness" or moral subjectivity under the definition recognized by most people. 11. AI is likely to surpass humans in almost all cognitive tasks and eventually extend to the physical world and robotics. 12. Coding was first taken over by AI; More general software engineering will be greatly automated, but system architecture and product judgment will be slower. 13. Biotechnology is about to usher in an AI driven revival, and humanity is truly beginning to 'cure many diseases'; Peptide therapy, cell therapy, and mRNA are the most noteworthy. 14. In a world where AI can generate everything, the ability to judge without being deceived or forming erroneous beliefs will become one of the most important competitive advantages for individuals. 15. Every two to three months, model upgrades open up new product spaces that were previously impossible, and application layer opportunities continue to exist - the key is to establish a true vertical moat. My suggestion is to use AI to solve your real-life problems.