Albert Gao🇺🇦

Something I learned over the years: The implementation matters, only when you get the API design right in the first place.

I 100% love the model vendors to use my data to train. It will certainly make the world a better place.

@1a1n1d1y Right after sonnet and opus suddenly developed this new habit of commit right after any changes, I looked for alternatives. If I need to prompt for basic shit like this. That means something fundamental is broken.

presented without comment

@mitchellh @jxnlco @jarredsumner Damn, seems I need to move to SF 😆

@jarredsumner This is actually one of the main reasons I left SF 13 years ago. Dating without bringing tech into it was hard. Moved to LA, dated an actress, together 14 years with 2 kids. We never talk about tech. Bliss.

me at nopa on a first date her: how’d you pick this restaurant? me: I asked claude [waitress compliments her pants, clothes come up] her: where do you shop? me: lost a bunch of weight, had to rebuy everything. sent claude a pic of me and it picked [few min later, looking at menu] her: wait how do you track the calories me: believe it or not, also claude. I send it a pic of the menu I’m such a shill

Really true. Yesterday I wrote a less than 100 lines code. Because in order to convey my idea, I need to describe not only this code, but also connections to the current code. Believe it is too much to do, ended up having a little fun. And once the code is written, asked the LLM to adopt it everywhere. Nice combo.

@abacaj I mean, the higher the number, the higher my company valuation, right? Right?😅

28T tokens for a 350M model!?

@birdabo Oh, no, bro 🤣 the contamination is fast🤣 literally for all Chinese friends, 0 loves this design, Chinese has taste❤️the original post is patriotic on the wrong spot. BTW: how did you make that video? Grok?

this is 𝕏 in china btw.

Nah, the western style is just better, which is also why it is been adopted world widely. I can find the features way faster in western apps than their Chinese counterpart. The worst part of the Chinese app design is the marketing over everything else leads to extremely poor UX and way more clicks to get to where I need to go. And all those clicks just for potentially saving few cents…I couldn’t find anything more pathetic than this…if the Chinese market is not blocked by CCP, you will see shitty design won’t survive any longer. In fact, many Chinese indie app and small SaaS are adopting western style.

If you open a Chinese app for the first time, you’ll probably think it’s badly designed. Too many icons and features. Everything crammed onto one screen. If you grew up on Western apps, your instinct is immediate: this is cluttered. But it works. In the U.S., we’ve been trained to associate good UX with minimalism. In China, density often signals value. Open WeChat or Alipay, and it feels overwhelming at first. Information-heavy, feature-packed. But to local users, that density means capability. It says: everything you might need is already here in front of you. If you enter a new market assuming your design taste equals good UX, you’ll misread the signal. Good design is contextual.

@thekitze I hate to say, but most of the time, more than half of the Codex comments are actually legit...😅

🚨 BREAKING: GPT 5.4 rates the Claude Code codebase 6.5/10 💀 "This is not junior spaghetti. This is staff-engineer spaghetti: performance-aware, feature-flagged, telemetry-instrumented, surgically optimized spaghetti" 😭

all you need to do is typing: `ollama launch opencode`

I have to say, local model actually works great for the right task. I am currently working in a no-internet environment, and have tried multiple sessions using this combo, it truly surpasses my expectations. The task on the image ran through several steps: code review, plan, execution, validation and cleaning (type checking and linting, etc), have tried multiple local models, qwen 3.5 122B is by far the most capable one. - @ollama 0.19 - @opencode 1.3.3 - @Alibaba_Qwen 3.5 122B Q4_K_M M5 Max / 128GB (Yes, I tested manually and with my comprehensive e2e and integration tests, all works great, the flow can still be further enhanced ofc, but really really solid start) Once my internet is restored, I will try using a cloud plan + local execution combo for complex tasks.

@jamonholmgren I believe Apple started this by setting that magic number. Then every company did the same since now the market has been educated by big brother already. 😅

Best evaluation for local models are the real use case, tested the models below for a small frontend refactoring work: - GPT-OSS 120B (OpenAI, 120B params) - Qwen3.5 122B (Alibaba, 122B params) - Nemotron-3 Super (NVIDIA, 49B params) The prompt is: “For all route components, component A/B/C are using their own shell X instead of the AppShell, update them so all the same.” A little nuance here is the component X wraps the common shell inside of it, so the normal way would be refactor the X by removing the AppShell from it, then wrap the A/B/C with the AppShell with the X still there. - GPT: followed by instruction 😆 just applied the AppShell, and has some minor errors. - Nemotron: a better version of GPT, with 0 errors, but still followed the instructions closely 😆 - Qwen 3.5: the first step was to check both X and AppShell, then refactored X and applied AppShell 🤗 So, obvious winner here: @Alibaba_Qwen M5 Max 128GB with OpenCode through Ollama

@theo I’m not sure which community you were referring to, but both the LocalLlama and LocalLLM subreddits are full of wonderful people. There is some really deep tech talk happening there all the time. Stereotyping is one way to interpret the world, of course.

Local models? Really cool. The local model community? Absolute cancer.

@thdxr @theo I was using Qwen 3.5 122B Q4 locally yesterday with opencode, actually works out of the box and wonderfully. I love the UX ❤️I would say it’s unique and better than Claude and Codex in some categories ❤️❤️

this crowd is infuriating we do want to support it well but it's obviously lower on our priority list given the number of people who want this we try to make things work out of the box well for the more typical users which sometimes is in conflict with what they want they make angry posts about how we're evil and we've set out to trick them and steal their data they post misinformation everywhere and someone even made a dedicated website full of incorrect accusations one time we fixed someone's issue and instead of being grateful they switched to saying "well i can't trust this to not break again in the future" it's like what they're really looking for is an opportunity to yell at someone

@thekitze All you need to do is to use LLM to generate another arguments like this to convince yourself 🤣

(a few) reasons i won't switch to ios even tho my current android probably died 💀 ◈ notification center CLEARS: it's way better in every possible way, period ◈ apps can display permanent notification in the notif center (i.e tailscale, tesla) so you can see their status + press notification action buttons ◈ full screen calendar reminders (apple doesn't do this except for alarms) ◈ with apps like tasker, autoapps, macdroid etc. you can do REAL automations ◈ enabling adb via network allows any agent to fully take over and control your phone ◈ i was laughing at the idea of going back to a fingerprint but it's way faster than face id ... i said what i said ◈ splitting and floating apps are SUPER useful, don't let tim cope convince you otherwise by forcing you to use a single app at a time in 2026 ◈ google photos > apple photos, all day every day ◈ little icons from apps in status bar tell you which apps you have notifs from without swiping down the notification center ◈ but what about clipboard sync? you can use KDE Connect which gives you better and more predictable clipboard/file sync than the iCloud one ◈ you can use syncthings? instantly sync any file across mac, android, or any other device on your tailscale network. i use it to sync my screenshots and camera folders to my mac so i don't need to airdrop things. they just appear there immediately. ◈ airdrop works btw (latest pixel and samsung only, which are the only androids you should buy anyway) ◈ you can install any apk file from anywhere without giving tim apple a bj ◈ install streamio, emulators, download roms, just do whatever tf you want ◈ a REAL file manager, not one that's pretending to be a file manager ◈ better battery life (yes, really) ◈ REAL difference between browsers (browser engines) not just the app skeleton ◈ chrome on android is gazillion times better than chrome and safari on iOS ◈ gboard CLEARS the iOS keyboard: clipboard history, snippets, always show number/emoji rows, better correction, better swiping, better better better everything ◈ gemini > siri i mean lmao it's not even a question ◈ WAY more sensors in home assistant ◈ way faster animations (especially if u reduce them to 0.5x) and feels way snappier than liquid ass ◈ no one is forcing app library on you, you 100% own your homescreen ◈ if you want you can install a custom launcher (i vibe coded one for myself) and completely replace your homescreen with something custom HOW THE FUCK DO I GO BACK TO HAVING AN ORANGE PHONE LIKE THE REST OF YOU WITH THE SAME FUCKING BULKY CASE AND SAME HOME SCREEN AND APP LAYOUT AND LIST OF COPE APPS LIKE FLIGHTY AND ALL THE OTHER BULLSHIT ANDROID 4 LYFE

@DavidKPiano @steipete I second this, the reason we need to have this ridiculous prompt is crazy: “Stop creating component during render, ‘const getUser = ()=> <span>{name}</span>’ is effectively a component! If something is a component, make it a proper component!”

If you willingly use `useEffect` for data fetching, that's fine, just please go into your GitHub settings and set this to disabled

@reach_vb Need to be able to open files in codex

We are expanding Codex’s toolkit - what tools would you love to see work with codex the most?

Damn, just like drugs, buying tokens to vibe code should be illegal. It’s way too addictive... 😆

@jammingames ❤️this is the real “family game”! ❤️you build, you play ❤️

This is why co-op is awesome. My dad (71), me (44), and my son (21) flying a mission together. Three generations.

GPT 5.4 ❤️ UI Design Official guide from OpenAI developers.openai.com/blog/designing…