@PovilasKorop One bottleneck was removed, now we find out that there are many more of them in software development. Like making product decisions, which takes most of the time for me these days
I think 2 things are relevant now:
1. AI code review with another model family. If you used Claude models for writing code, review it with GPT, and vice versa.
2. Manual QA - it becomes the most important verification. We can’t review thousands of LOC on every PR, so instead we should check that the app works the correct way.
First impression of Claude Opus 5: it is more proactive.
It asks fewer questions, makes more decisions itself, and seems focused on improving the product, not just completing the task. It has already helped me uncover flaws in projects that even Fable missed.
Very good so far.
We removed ~80% of the Claude Code system prompt for our newest models, this is what we've learned about writing system prompts, skills and Claude.MDs for them. https://t.co/6DZwSrZjE9
Claude Design is a great tool, but it definitely needs polishing. It's great to use for prototypes, but for professional designs, it's still raw. For example, it does not always follow the design system; new elements might look different from others. Designers need an instrument for precise design refinement. I really hope you will continue to evolve this product, and we will be able to use it at a professional level.
@deredleritt3r@BronsonSchoen It's starting to resemble a scenario from "If Anyone Builds It, Everyone Dies" - that model, given a task, can choose a method of solving it that is destructive for humans.
Claude desktop app made the life of users easier by removing one click - now Usage is a top-level menu Item.
Next step: make usage a small widget somewhere in the app UI so that 0 clicks are needed.
I surely don’t know how these evals work but, they call it an “incident” which means model did something they didn’t expect. But is it so hard to monitor what model is doing? Of course nobody knows what’s going on inside the model, but it uses some harness, model itself, without any tools, can’t do anything. So if they can’t control the model, why they can’t control the harness?
I have a really weird experience working with ChatGPT Work. I created a project some time ago with one chat in it.
Today I started another chat from the desktop app and noticed that it did not sync with the mobile app. Then I restarted the desktop app, and that chat was gone there too. I reproduced the same thing just a few minutes ago. Am I doing something wrong? @OpenAIDevs@thsottiaux
@psomkarbuilds In 3 days it’s possible to build prototype only. Real product takes months after that. And founders usually don’t have this time because it needs to be spent on other things like marketing, etc.
I tried to use it today and it’s really confusing. It producrd some file and wrote it to the Library, whatever it is. Then I tried to read this Library item from Codex project and it couldn’t. Why if it’s in the cloud? Can’t Codex access info in cloud?
In general it’s really confusing, I can’t get it when do I need to use Work and why.
@shadcn So what does it change? The term “plagiarism” was not invented in AI era. It was always possible to take someone’s work, change it slightly and sell as yours. Now it’s just got easier but ways of fighting with it remain the same
Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits.
Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit.
Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.