Phyrex|Nov 16, 2026 07:29
I've been using AI for collaborative work for a long time. This time, I conducted a test where three different AIs took over a project, and I compared the results:
1. ChatGPT: Its work ethic is excellent—it's like the class monitor or student council president—but its delivery capability is abysmal. Small pages are fine, but large programs always encounter problems in the crucial final step, and these are often unsolvable issues.
ChatGPT's biggest drawback is its inflexibility; it sticks to one path and never communicates with other AIs when it gets stuck. Therefore, for me, ChatGPT is best suited for basic infrastructure.
A single large cloud task can consume a whole week's worth of Ultra credits in 24 hours.
2. Grok, or Grok Bot, is the kind of unsung hero. You can clearly see a struggling student trying incredibly hard, working diligently and delivering decent results, but it's like giving a 30% answer to a 100% question. I know many of you are used to using Bots, but Bots have a major pitfall.
In Bot work, you can't choose the workload intensity. In simpler terms, Grok has five intensity levels, but a Bot isn't in any of them. For a Bot, it's like a secretary; the smaller the task, the better the delivery and the higher the accuracy, simply because their "intelligence" is limited.
And I haven't even had time to test the higher intensity Grok tasks before I used up my weekly quota. So next time, I'll mainly test Grok, not Grok Bots. And I also advise those doing large-scale development: don't use Bots; they're a nightmare.
3. Claude: Frankly, my impression of Fable 5.1 (medium level) was quite average. It's extremely resource-intensive, seemingly three times faster than the highest level ChatGPT and twice as resource-intensive as Gork BOT. Its delivery capability is mediocre, making it very suitable for finishing tasks, essentially cleaning up after ChatGPT and Gork BOT.
Upgrading to Fable 5.2, I started using the advanced version. It's a true powerhouse, a quiet powerhouse. Almost any difficult assignment can be solved with Claude, but it's incredibly expensive.
However, Claude isn't without its problems. Compared to the comprehensive capabilities of ChatGPT and Gork, Claude is very poor at image creation. It can handle image processing, but it can't generate images independently; it must be used in conjunction with other tools.
4. Muse is a complex device. For me, Muse's biggest drawback is its need for human assistance. It's extremely cautious about permissions, like a timid little girl in class—she might be a good learner, but she's incredibly fearful and always expects you to make decisions. The most frustrating thing is that I have to agree to every connection to GitHub.
Every write operation requires approval; there's no skipping. So, while Muse might work locally, it's absolutely unsuitable for frequently modifying GitHub.
On the other hand, Muse's remote assistance feature is interesting. Tasks can be handed over via email and GitHub, making it essentially an excellent collaborator. It's very simple; just set up Muse's email address and send commands via email. Muse will reply with the completed email, but you need to delete it frequently, and you can't use your own frequently used email address, otherwise you'll constantly get reminders.
From my current work situation, ChatGPT handles the basic setup, Gork and Muse handle the implementation layer, and Claude handles acceptance, oversight, and cleanup—it's a good combination.
One @Gate, more markets for trading.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink