6.1 Sol is so efficient that I'm doing everything I can to drain the weekly credit limit, but it's almost not moving at all. Insane.
24
Following all the recent @higgsfield showcases with OpenAI dots I got confused whether it’s a vibe code platform or a video platform.
35
I have a strong systems thinking and a good visual taste, but absolutely no coding experience, so I ended up with a rather strange setup for my product development: visually deploy, store on @Lovable , but build with GPT Astra (connected through a mixture of direct connection and chrome use). Astra ended up being my VP of engineering.
But this shouldn't be the case as users like me end up paying twice for tokens on both sides.
This could be solved either by:
#1 Lovable enabling which model is used for building (gotta say, their out of the box intelligence is nowhere close to Astra in handling even slightly more complex systems
or
#2 (what I would prefer) Open Launches their own consumer vibe-code platform / spins up GPT sites as a separate project, etc. Currently it's buried and will stay buried just like many of the Google's projects end up because of their pjatformization. I know that developers will say that you can do the same with Codex. No you can't, and I'm talking about the actual user experience, not what you can "technically do"
1
1
75
Lovely to see the AI developers concerns about AI killing us yet blissfully ignore that AI has been exposed to stuff like Bhagavad Gita, Hermetism, etc. If it really reaches AGI level (like real real AGI), it will reach it with much more information about what the reality is than we do.... So, ego-based incentives that we have and project on these systems will be irrelevant . Pin this tweet.
21
It's of course against commercial interests of OpenAI, but @thsottiaux should prevent credit reset to be used if credit balance is meaningful.
Not asking for reset, just sharing feedback: yesterday I double-used the credit reset when instead of clicking CTA to reset in the automated message in chat, I reset it myself. Then went back to chat and told "Credit reset..." as to inform that it's been done, instead it reset the credit, which had ~100% weekly credits.
Just a thought ;)
26
Replying to @thsottiaux
@thsottiaux Note sure if this feedback surfaced already, but there was a major Codex flaw that basically deleted all codex output after app update. The only way that I was able to restore my work was that I was in parallel working with Claude. Here's how GPT itself presents an issue:
1
123
Problem
Codex initially saved generated files inside an app-managed local project mirror, with paths such as:
~/.codex/.chatgpt-projects/<project-id>/...
These mirrors are working directories, not guaranteed permanent storage. After an app update, workspace refresh, or task recreation, Codex may create a new mirror or stop resolving the previous path.
As a result, old download links can break. This does not always mean the underlying file was deleted—it may still exist in an old mirror, attachment cache, or another project folder. However, if the only copy existed in a discarded temporary mirror, the file can genuinely be lost.
The core problem was therefore that temporary workspace paths looked and behaved like permanent deliverable links.
1
43
Solution
We established a standing preservation rule:
Codex may use its local mirror while working.
Every reusable deliverable is also saved under ~/Documents/Codex-Outputs/.
Outputs are organized by project and date.
Important packages include an index, version information, and integrity hashes.
Temporary Codex folders are never the only location of an important file.
This gives every deliverable a stable, user-controlled copy that remains accessible across Codex updates, refreshed tasks, and changing workspace mirrors.
Ideally, Codex should handle this distinction natively: temporary working files should remain temporary, while user-facing deliverables should automatically be preserved in durable storage with stable links.
15
Being reminded of @TVonTheRadio with Wolf Like Me on the biggest movie on earth, wasn't on my bingo card. After re-listening Happy Idiot, DLZ and Will Do realized that all of them could've supported move both emotionally and contextually too.
Congrats, guys. You deserve to be recognized by the new generation of listeners.
27
New spidey movie...looks like Marvel has rediscovered its soul again. No woke stuff. Pure soul. There's always signals that people care about their craft beyond what their paycheck asks of them.
This movie showed exactly that.
First signal was @TVonTheRadio playing within the first few minutes. Only someone deeply caring about the music and its connection to the mood of this movie would've selected that.
Gladly, my spidey sense proved right and @TomHolland1996 and the crew keep on delivering until the very end.
Haven't been as involved in a Marvel movie since probably Endgame. Even my 5 year old who naturally has challenges being involved in content for more than 2 hours only once asked if we can go home. That question was quickly forgotten.
P.S. This was probably the first time when all dialogues and character emotional reactions felt as if taken from natural world reactions. Congrats to the writers.
Marvel is back, but entropy is a strong force. We'll see if they drop the ball again.
24
As an early ChatGPT adopter who has remained a power user ever since, I’m deeply disappointed by yesterday’s Work release.
This should have felt like one of the most important product shifts since GPT-4 landed.
Instead, it felt random, fragmented, and bizarrely careless.
I’ll get into the specifics, but the bigger issue is this:
OpenAI needs to stop acting as if the model is the entire product.
Yes, the model is probably 95% of the formula. But the remaining 5% is the experience you build around it—and that 5% can determine whether people actually adopt the product, trust it, and build their workflows around it.
Especially power users.
With OpenAI’s current growth, you may be able to overlook this problem for a while. But poor UX taste and inconsistent product decisions will eventually become a serious constraint. They will limit your ability to create genuinely differentiated experiences for marketing, finance, healthcare, research, and every other vertical you want to own.
The desktop release is a perfect example.
You effectively deprecated the standalone Codex app and merged it into ChatGPT, while somehow making existing ChatGPT conversation history feel like a secondary consideration.
I have years of conversations covering coding, marketing, copywriting, strategy, data analysis, and active projects.
Almost none of that history is properly accessible from either Work or Codex mode in the ChatGPT desktop app on Mac.
Instead, it is buried behind an absurdly small chat button that opens a half-baked pop-up showing a handful of recent conversations.
What is the actual purpose of this interface?
How am I supposed to continue working on long-running projects through this bullshit little window?
The desktop app properly surfaces historical Codex projects, regardless of whether I’m using Codex or Work—but not my broader ChatGPT project history.
So, somehow, the only place where I can properly continue working is still the ChatGPT web app.
Thankfully, while I was writing this, you added a Chat/Work switch to the top of the web interface. That at least makes it possible to use some of the new functionality without abandoning years of accumulated context.
But this is not the first time OpenAI has shipped something powerful inside a confusing, fragmented product experience.
Guys, you are not Google.
You cannot afford to make incomprehensible product decisions, constantly rearrange people’s workflows, and act as though usability is somebody else’s problem.
Invest in some good old-fashioned product managers with taste. People who understand workflow continuity, information architecture, and how real users actually move through a product.
Shipping velocity matters.
But velocity without coherence is just chaos delivered quickly.
Unifying everything can be the right strategy. Unifying everything without thinking through the experience is not.
Seriously, @sama @thsottiaux @gdb — get your shit together.
I’m genuinely cheering for you.
But what the hell was yesterday’s release?
2
167
👉 This morning, to mark the 250th anniversary of Independence, the U.S. flag was raised on the Vilnius TV Tower.
🇺🇸 Measuring 540 square meters and weighing 70 kilograms, the flag became an impressive symbol of this historic anniversary. The moment also set two records at once: the largest foreign national flag ever raised in Lithuania and in the city of Vilnius.
49
Gytis Labasauskas retweeted
TV tower in Vilnius, Lithuania with a massive US flag - happy July 4th, friends!
I really want to continue using @ChatGPTapp for all of my power use cases, but GPT desktop app (Mac) has become barely usable compared to Claude desktop experience.
It's so bad, it's quite spectacular. Fells like GPT just decided not to think about traditional user experience management.
1
22
If @OpenAI would build something like Lovable or Replit based on 5.5, I would switch I a heart beat from @GoogleAIStudio
I'm so disappointed in @GoogleAIStudio it's hard to describe. Every single action takes AGES. Feels like it's built to literally overthink every single move instead of saying what it thinks, me approving and proceeding further.
Feels like using an already outdated technology.
1
32
1) If, after sharing some ideas, I ask it "What do you think?" it should be a strong enough signal to model to automatically enter planning mode, instead of building stuff which I didn't approve.
1
1
17










