@hyperknot
Account based inGermany!
About this account
- Account based in
- Germany
- Connected via
- Germany App Store
X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
Building https://nitter.cf/t.co/CUfyhT0mBC and https://nitter.cf/t.co/GTLrvnmkaJ
Budapest
Joined July 2012
- Tweets949
- Following1.2K
- Followers3.6K
- Likes1.6K
Dear @pidotdev team (@mitsuhiko @badlogicgames)
I opened my 11th meaningful GitHub issue: auto-closed like always.
I've spent at least 10 minutes on each of those issues and many have contributed to meaningful discussion or PRs.
No matter how much I love pi and the core team's values, its such an extremely negative experience to try to contribute to this project.
Opus 5.5 is the first model in a very long time which I really enjoy using. It doesn't want to over-engineer, happy to remove and simplify, and is generally an nice partner to work with.
Last model which I enjoyed so much was Opus 4.6 from almost a year ago.
Opus happily finished the UI refining step, with visual feedback in a Playwright loop.
Do LLMs not feel pain when looking at this?
OpenAI researchers be like: 25% of the human cortex is dedicated to vision, why not do the same in GPT-6?
Mixture of Experts could also let vision parts stay dormant for non-vision tasks, so it doesn’t add much inference cost.
Astra can do segmentation
this is pure VLM result. no expert models (like SAM) were used
No other VLM even comes close to this quality
- high effort
- avg input tokens / image: 2,052
- avg output tokens / image: 4,685
- avg cost / image: $0.255
- median time / image: 78.3 s
↓ GPT-6 Astra segmentation deep dive
I've been using GPT 6 Astra (aka AGI is here) and I'm absolutely puzzled by the results I'm getting. Its basically unreadable for me. Missing whitespace from the code is one thing, linting fixes that, but the text it writes is unreadable.
Yes, it is smart, but its getting into an extremely unusable direction for many tasks, where previous models were uncomparably better. (Do you remember Opus 4.5/4.6? IMHO they were the peak of something we lost.)
I see extremely strong push before the IPO of OAI and Ant, my X timeline is basically AGI is here and vibeslop 3D demos, yet the most important core values of these models are getting lost. Of course, everything is benchmaxxed.
Even in @sam_paech 's post here, he says:
> Meanwhile, GPT-6-Astra has forgotten how to write in paragraphs. It's a sad state of affairs!
while the screenshot shows Astra on first place!
New models tested on creative writing:
GPT-6-Astra
Fable 5.1
Muse Spark 1.3
Gemini 3.8 flash
Subjectively, I strongly prefer Muse Spark 1.3 to both GPT-6 and Fable. It just gives you what you asked for, without an obnoxious house style permeating every request.
Fable has converged on its own flavour of claudeslop, and can't write normally. Meanwhile, GPT-6-Astra has forgotten how to write in paragraphs. It's a sad state of affairs!
Tentatively suggest we reinstate the old tradition of reading rollouts with our actual eyeballs.
Zsolt Ero retweeted
"Indie Hacker" used to be a few crazy people who built and sold their products online solo.
Nowadays Indie Hackers extract each other's money with bidding sites, worship air fryers, and boast of their 𝕏 payout.
Zsolt Ero retweeted
Electron hate is one of the most confidently wrong opinions in tech. People say it like they've cracked something open. "Discord is just a website." "VS Code is Chrome with a titlebar." Yeah, and your kitchen knife is just shaped metal. The framing tells you nothing about whether the thing is actually good.
Writing a real cross-platform native app is brutally hard, and not because the logic is complicated. Every platform has different UI conventions, different system APIs, different accessibility models, different font rendering, different input handling. Write a macOS app in Swift and it looks great on macOS and doesn't exist anywhere else. Want Windows? WinUI, WPF, take your pick, each with its own learning curve and its own special set of things that don't quite work right. Linux? Qt or GTK, both of which produce apps that feel slightly wrong on every platform they target, and you're maintaining all of this in parallel, same features across three codebases, three bug trackers, three build pipelines, three sets of platform-specific nonsense to debug Or use Electron with just oneOne codebase.
"Electron uses too much RAM." VS Code idles around 150-300MB on a typical project. Sounds bad until you check what else is open. Chrome with four tabs is using 800MB. Your JetBrains IDE, fully native, compiled to the JVM, is sitting at 1.2GB before you've opened a single file. The native Slack alternative someone built in Qt uses 90MB, sure, but it also hasn't shipped a new feature in two years and the emoji picker breaks on HiDPI and nobody is fixing it. Memory is cheap. The RAM argument is almost always made by people who don't look at what their "good" native apps are actually consuming.
Chromium is good. It is one of the most tested, most optimized pieces of software running on consumer hardware right now. The rendering is fast. V8 is fast. The security model has sandboxed processes and site isolation baked in, which is more than most native apps bother with. Embedding it in a desktop framework is not a betrayal of some pure native ideal. It's using a genuinely good piece of engineering for a job it's good at.
The app is not slop because it runs on Chromium. The app is slop if the team who built it didn't care. Those are different things. Maybe stop confusing them.
Zsolt Ero retweeted
Deep inner suffering inevitably arises when the human person is reduced to performance, consumption, or a statistical datum. Many young people today live under the yoke of expectations to perform, immersed in an exasperated competitiveness that generates anxiety, fear of not measuring up, and disorientation.
Zsolt Ero retweeted
Cold take on what comes next:
- OpenAI will flourish
- Anthropic will continue to be profitable
- Google will not catch up to Anthropic or OpenAI
- no chinese company will catch up to Anthropic or OpenAI
- the highest tier of intelligence will become a luxury product that only companies and multi-millionaires/billionaires can afford
- most of the companies that invested massively in them will have massive returns
- SpaceX’s AI will be fine and on par with Google by end of year
- Nvidia will become the first 10T company
Hot take on what comes next, after the sudden decline of tokenmaxxing:
- OpenAI will struggle
- with the decline of tokenmaxxing Anthropic will struggle (aside from this quarter) to make a profit
- Google will catch up to Anthropic
- some Chinese companies might, too
- LLMs will become commodities; margins will be very very thin
- Most of the companies that invested massively in them will struggle to make back their investments
- SpaceX’s AI efforts will flail
- Nvidia will eventually decline, once all of the above becomes widely recognized.
✨ I just replaced Mapbox on all my sites with OpenFreeMap by @hyperknot and my map bill is now $0
Mapbox's pricing is getting increasingly extortionary (which is fine, it's capitalism) but at some point you have to think, $857/month for what? A map? Really? A map is that expensive? How can loading a map be that expensive? It's just some PNG tiles you host somewhere? Why?
@OpenFreeMapOrg is 100% free and all you do is point your AI to openfreemap(dot)org and tell it to replace Mapbox with that
5 minutes and you save thousands $$$ per year!
Apparently @Cloudflare sponsors its bandwidth which is very cool and keeps it online!
Zsolt Ero retweeted
I am excited to announce that BigBoy Charging is now officially powered by OpenFreeMap!
This latest addition, combined with the incredible mapcn, completes my goal of being able to provide this service for free to all EV drivers 🔋⚡
Huge thanks to @hyperknot and @sainianmol16 for making this dream of mine a reality 🫶
Reminder before Sonnet 5 drops: SWE-bench tests a model’s ability to fix small Python bugs in 12 repos in one-shot with appropriate context fed to it. It’s not a measure of agentic coding ability.
I wrote in detail about a bunch of benchmarks and what they mean, link below.
Zsolt Ero retweeted
Made a site comparing the sizes of living things :)
The great Julius Csotonyi spent 5 months painting over 60 illustrations for the site, no ai used
This is the state of the art on ChatGPT when asked to create a map (topic was hot springs in Cyprus).
I guess interactive maps are safe for now.
I've been working with AI models daily since ChatGPT came out, and there is something new in Opus 4.5 what I haven't seen in anything else before.
Something in Opus 4.5 feels almost as big of a step as OpenAI o1 was.
I'm asking about a hypothetical experiment of making a vented paragliding airbag, which doesn't exist yet, and for the first time, I feel an LLM can "imagine" the world and communicate how physics behaves by making ASCII drawings and Python simulations.
GPT-5.1 is better in doing calculations internally, but it doesn't recognize it's own limitations.
Opus 4.5 doesn't even try, it makes a Python script instead, and asks me to run it and give back the results.
I didn't even ask for either illustrations or a script or mentioned Python or coding!
Some examples:
It drew a comparison chart in ASCII with the exact values of the Physics problem we are discussing (I didn't ask for it, yet it's amazing, even with the spacing bug!)
Finally it zero-shot a Python script, which generated this beautiful dashboard and run an optimization problem in the console (which I should copy-and-paste back).
For the first time, I feel an LLM really understands it's an LLM and communicates like a partner for the first time.
Opus 4.5 is something special.
