jorl17 · HackerTrans

5 dagen geledendiscuss

This was a much better experience than I expected. Rather unbelievable!

Side-effect of the data: clearly the model is better than I normally am at playing, as it spontaneously did several things I had not told it to do and wouldn't really know how to do (at least not with a keyboard).

Really remarkable, congrats!

jorl17

11 dagen geledendiscuss

I only use superpowers when I want to stick with composer2.5(fast) for everything. If I use it with other models it's terrible. With composer2.5 it is slightly better, though not much.

jorl17

19 dagen geledendiscuss

Sandy might remember.

jorl17

20 dagen geledendiscuss

I'm from Portugal and I start losing it at 25. 30 degrees is insane.

Last summer my house got to 39, and I didn't have AC (it was broken). I think I'm still recovering.

jorl17

22 dagen geledendiscuss

Claude code (desktop) and Codex (desktop) are both absolutely dogshit pieces of software. I can't pick which one is worse. I'd be sort of ashamed to say I actively worked on them, regardless of how they can empower people. Cursor's new UI is similarly terrible. They're all slowly getting better, but too slow for my taste.

They are incredibly slow in unpredictable ways, eat up memory at an insane rate, and just feel like they were built with no regards for UX. Like they crammed together all the engineers with no idea of how to build a coherent and predictable UI and let them loose on the product without proper designers.

The other day Codex (desktop) was eating up 70GB of RAM on my machine. What had I done? Literally nothing. I opened it and let it update once.

Another one with Codex was when I had a specific conversation where no activity was happening and which would make the app spin up all of my CPU cores, rendering it barely usable. It would take seconds to react to anything or update the UI. The conversation wasn't even in focus!!! Restarting the app wouldn't help. After I archived it, it suddenly got better

Claude Code Desktop used to be so, soo, soo slow and eat up so much RAM. It was unusable for anything other than playing around when I first tried it. It also didn't communicate any of what it would do. Using it was like living in a world with no affordances, constantly afraid of interacting with them and being faced with some sort of destructive action. Still, it has definitely been improving in terms of the UI experience.

Cursor's new agents mode suffers from similar issues. Obscenely slow, hogging CPU without anything going on, breaking with existing UX patterns (some of them already well implemented in their other, more polished, previous version), confusing buttons and labels which don't explain what to do and that sometimes do destructive operations on your code.

My favorite cursor absurdity is that if you use their workflow to create a worktree and the worktree setup script fails, the following happens:

1. The agent has no idea that it failed, let alone have any logs of the failure

2. Often you yourself don't get access to the logs of what failed in that script. Don't ask me, half the time it just says it failed with no further logs.

3. When you do get the logs, you cannot copy them in ANY way. You can't even select them. I have had to resort to taking a screenshot to do OCR on it

I've also had cursor repeatedly have concurrency/race condition bugs when creating multiple worktrees in parallel. I have 5 tasks, I spin them up all together so they can create 5 worktrees and they crash with random internal cursor errors. Wasn't the point of this abhorrent new UI you've stuffed me with to enable parallelism?

It's like people aren't even testing the shit they ship. Which I guess they aren't.

I'm a big believer in AI and think it is changing the world and will continue to do so, but I almost get offended at how bad these products for which I am paying (sometimes quite a lot!) are. There's "move fast and break stuff" and then there's "build crap to call stuff".

jorl17

29 dagen geledendiscuss

Same on Zen

jorl17

vorige maanddiscuss

This is genius! I can already see improved versions of this idea making it big.

jorl17

vorige maanddiscuss

FYI your link 404s.

Seems like you copied the ellipsified version, so what we get is https://github.com/gitsense/gsc-cli/blob/main/internal/cli/r

jorl17

vorige maanddiscuss

So, in the past I've shared that I evaluate AI models by feeding them my ever-growing large collection of personal poems that span well over 800 poems (1000 depending on how you count) and over 250k tokens.

What I do is feed it some initial prompt asking it to simply discuss what can be said when faced with this unedited, unseen collection of poetry. I ask the model to evaluate who the author is (or claims to be), what they went through in life, if there are different chronological poetic "phases" or different types of poetry. I request an analysis of the body of work and of the author themselves. In the more recent versions of the prompt I ask it to dive deep. Then I add the poems, chronologically sorted, with an index, a title, and a date (and subpoems, if they have them).

Crucially: Since ~70% of my poetry (or thereabouts) is in portuguese, I ask this in portuguese, and I get back an analysis in (european) portuguese. Earlier models couldn't even do that properly.

In the past, I couldn't use such prompts, and had to use longer, more guiding ones. I also couldn't even feed all of my poetry to the models because they just did not have enough context.

I'll go ahead and state that Claude Fable is undoubtedly the best model I have seen, though I cannot put a number on how significant a leap it is -- perhaps because my benchmark does not allow me to evaluate that anymore. I would say it is a significant leap over Opus 4.6, though -- a new level of understanding. Okay, I'll try to put a number: if Opus 4.6 was a 16/20, this is a 17.5/20. These numbers are pointless, but I had to try.

It made one (1) relevant mistake I could identify (where it messed up the names of two relevant people in my life who I have not talked to in over 5 years).

I'm impressed by how it just feels like it's getting the person behind the poetry, and how nearly every statement it makes is correct -- and when it isn't I am completely aware that no one could know based on the poetry alone (bar that one mistake I mentioned -- and that's very needle in a haystack, like deducing the name of a person based on a poem based on another poem with hundreds of other poems in between!)

It's really hard to explain, but it just finds more correct connections between the poems and explain much better my (recollection of) a state of mind when writing poetry. This is also the first time where it really unravels some key concepts of my poetry in a way that seemed almost effortless: it lays bare the poems and what they imply about the meaning of some of my concepts. Other good models understood these concepts, but this feels like it's on another level, as if it's making it simpler as it speaks, rather than the opposite -- like a good teacher.

When it is explaining several topics related to my poetry and myself, it cites poems which even I had already forgotten but which it is entirely right to select.

I am actually feeling a bit emotional with how much it "understands" of me here. It's somewhat incredible how LLMs have progressed from the lack of comprehension of a couple of poems paired together, going through realizing a body of work has some guiding principles and cohesion, to truly figuring out these deep concepts and intricate connections which I know for a fact would take months of someone's life to unearth. Every major breakthrough feels like my soul is being spliced together by an AI model out of these hundreds of tiny pieces of me. I can't put into words how unbelievable this feels, and this Fable analysis, like others before it, is on a new level.

Let me put it this way: there are several poems in my collection which one can try to "guess" the meaning or context of. But I don't think many people would get it, because they would have had to know me really well and to be following along my life as it went. Even then, they could very well fail to attribute such meaning. And, with each new major release, models have gotten much better at guessing.

Before Opus, they would guess incorrectly often, and in many scenarios where I thought it was rather obvious that they were wrong. I think a human spending time looking at the poetry would quickly dismiss the proposed ideas of the model.

With Opus, it was the first time that I would almost always say: "Ok, the model got this wrong, but I think many humans would make the same 'mistake', and it wouldn't surprise me if everyone just assumed what Opus did".

Now, with Fable, there are very, very, very few sentences in this very long answer it produced where I can say: "Yeah you got that wrong, but I get it". In almost every situation it is mapping concepts, ideas, interpretations and cause-and-effect correctly. Yes, it is hard to "guess" what I thought, or was going through, or how X connected to Y -- but this model is doing it, incredibly consistently. I know I'll get the usual naysayers to these posts who think I'm just shilling a model, but this is the truth: what is being done here is amazing and I don't believe I know any person around me who would find this out about myself reading all of my poetry.

I often write poetry from the point of view of other people (some of which I do not know) and models (even Opus) have this tendency to make the opinions in poems as my own. Fable is the first that looks at a particular poem here and says "maybe this is not the author's opinion, who knows". The literal first model. It then immediately fails to do so with another poem, assuming it was about myself, but it's clear, undeniable progress. And like I said: I think most people would not _know_ which poems are truly about myself or not.

I've written word after word here, and yet words elude me to convey what this model represents to me. How it's almost always right, how it sees my fractured bits as a sort of cohesive whole, and how it just seems to "understand everything better". That's just it: it just seems like it really understood everything better. Like Opus before it, and like Gemini 2.5 pro before it. Out of the tens of thousands of verses, it picks some which no other model had picked and which I feel truly represent some of my best work. Older models seemed to sort of have a "hole" in its knowledge in the middle of the corpus, where they knew what was there but in a sort of hazy/foggy way. This model seems to recall every part of the corpus with the same precision.

For context:

- Opus 4.7/4.8 were a noticeable downgrade over Opus 4.6. They wrote more, in a harder to parse way, and they made up more. Still, All Opus models are clearly superior to everyone else by a large margin

- Sonnet-level models have a slight edge above the best of the other models. But they make too many mistakes, don't grasp several concepts, mix up their dates and timelines. 3 years ago I would have been blown away by Sonnet models but today they are inferior.

- Gemini models have a unique way of approaching the request, where they try to literally interpret my poetry as a mathematical theory. This sort of makes sense if you look at some poems, but it is surely laughable, as if someone one day actually has access to all of it, no one in their right mind would do so. This is a shame, because the first big breakthrough with LLMs and my poetry, to me, came with 2.5 pro, which was the first model that could look at the whole corpus as a cohesive whole without getting lost in the middle of it or making things up.

- GPT models have improved over time and also have this sort of alien-like language, sometimes being a bit too blunt in their analysis, but I can't say they are meaningfully superior to Gemini models.

I am very pleased to see progress in this area again, as Opus 4.7/4.8 were NOT progress and I was worried that we had hit a plateau here, but I can't say that.

In all honesty, the level of understanding and cohesion that Anthropic's models (Opus and above) have over my poetry means I fear my benchmark may be hitting its limits, as I don't know if there's anything a model could do that would wow me and lead me to say "this is a major breakthrough". Perhaps Mythos is a major breakthrough and I don't know. I can't find much that's wrong with it, but I also couldn't with Opus.

As I have in the past, I will periodically probe the model again and see how coherent it is. For now, I'm very happy to see an improvement.

What surprised me the most was that even though I set the thinking budget to xhigh (in OpenRouter), this model instantly started replying without showing a thinking block. I thought it just had the thinking hidden but that is not the case, as some replies showed thinking and anyway the first reply was blazingly fast. (I will try Opus 4.6 without thinking now, just to see if it changes it for the better -- maybe that was just it. I'll edit the message if it shows improvement).

jorl17

vorige maanddiscuss

Opus 4.7/4.8 often over-engineers on my setups, plus:

- It talks a LOT more like GPT models. You know: wrinkle, shape, gate, coarse, scope, gap, path, production-ready-workflow-of-the-day, and so on -- "that's expected, a consequence of the previous like-driven workflow". If I wanted to get a headache using AI I would have gone with GPT in the first place!

- It outputs text in a much harder way to follow along. I can't exactly say what it is. Maybe a bit of everything? Bolds are missing, bullet points are gone, paragraphs are bland and too long, and it doesn't feel like a model programming with me, but rather a somewhat full of themselves grandpa developer looking down on me. It's very weird to describe this, but it is definitely how I feel.

Granted this can totally be because of the way it reacts to the prompts now. We've got a rather large corpus of skills and "rules and good practices" that Opus 4.6 responded to great, and maybe the new models just get turned into this when fed with them....I don't know.

Either way, with Opus 4.6 being as good as it is, I need Fable to be a significant step up to justify a price increase. if it can get me to babysit opus a little bit less on some stuff, it might be worth it. Otherwise, I'm very happy with Opus 4.6 and hope they don't deprecate it.

jorl17

vorige maanddiscuss

Yes I also noticed it is pretty slow, which sort of defeated the purpose of using it for me.

Usually I'm working on a large task, typically with Opus, while also having a bunch of smaller tasks in their own independent worktrees. Those still need supervision, but less. My goal was to get deepseek to drive the cost of those down, but it was too slow and unreliable...

jorl17

vorige maanddiscuss

Not OP, but: I guess in a similar fashion to when I google things or read other websites: I don’t, but I use my instinct, judgement, experience…

Very often I do catch LLMs, even the best such as Opus, confidently saying wrong things about areas in theory I know little of. And sometimes I fail to catch them and only realize that later on….sort of like…how I learned my whole career? So many wrong abstractions, tools, and so many hard earned lessons. With LLMs it’s the same, but the process is much faster. For critical decisions I don’t blindly trust an LLM, for example.

jorl17

vorige maanddiscuss

Out of curiosity, what is your stack? And is this in a legacy project or new one?

I have tried using deep seek flash and pro but they make amateur mistakes. Sonnet level at best.

However v4 flash is absolutely amazing as a generalist model and it’s what we’re using on a product built on top of LLMs. I wish I could code with it but it’s not going to happen anytime soon

jorl17

vorige maanddiscuss

I’m digging into deeper / more complex problems, now. On top of that, I’m also building products faster for our startups, so I am filling in much more of a product role than merely an engineering one. But, really, it is both — and I’m absolutely loving it!

Also, with the added speed I can produce things more in line with the quality I’ve always wanted to add (many more tests, for example).

jorl17

vorige maanddiscuss

You had me at the NAND gate in AoE II's editor.

jorl17

vorige maanddiscuss

Your website is ok. The blue-on-black text in the footer is very blurry but that's also part of the CRT-filter design.

Since most of the actual text is not "white on dark", but "dark on blue", it's very readable.

The blue version is fine. The best of all 3 by a large margin

The amber version signals me that I should stop looking at it eventually, though I hang on for a while

The green version hurts my eyes. I want to leave as soon as possible

I think the green is brighter to me than the amber and, on top of that, the black text on green background somehow also gets hard to read.

So you definitely made the right choice going with that default!

jorl17

vorige maanddiscuss

There are many tools for this, and I use the one that I tried first, so there are probably better-suited alternatives out there.

I run MacWhisper, and I paired it with BetterTouchTool so it triggers on any input when I double tap the fn/globe icon.

Obviously all of my transcriptions through it are entirely local. I usually use the Large V3 Turbo model, though in the beginning I used Parakeet v3, which was slightly faster but produced more mistakes (and kept a lot of filler words -- 'ahhm', 'hummm').

However, if I'm interacting with the Claude or ChatGPT/Codex apps, I often use their voice recognition instead, because it tends to be more accurate, especially with punctuation, albeit significantly slower. OpenAI's is noticeably better than Anthropic but I feel like that gap has closed a bit recently (might be all in my head, though).

Like I said I don't really care about mistakes in the transcription. If you try to read it, it feels like a fever dream, but the LLMs get it.

If I say "taken" it may have "take and" If I say "all the while calling the method" it might have "although a while. while. call in the met of". This is a rather extreme example but I've seen them happen. The repetition of words happens because I'm talking with "humns and ahs" and do repeat words or just the ends of words. It's very rare for the models, especially Opus, to have any issue with this transcription. When they do, they tend to signal to me they didn't get it, or I catch them in the act. But, like I said, it really is very very rare.

As an example, I've got quite a significant feature to work on, which would have probably taken me weeks to design and implement, and I've used this exact method today to ink out the plan:

- I have spent the last couple of days researching the feature in my off-time and just "thinking about it in the background" (think: I fall asleep thinking of it -- a habit I've always had)

- I spent ~25 minutes brainstorming out loud. The transcript ended with ~17.000 characters and ~3.000 words.

- I sent that transcript, in cursor, to Opus 4.6-High with instructions on how to iterate on it and how I want to work while planning

- I then spent about 1.5 hours with it iterating and building the actual plan (and supporting technical decision document, which points at the FULL transcript of the whole interaction). Many of my original ideas made it to the final plan, others got scrapped or simplified, and others still got added. It contains a mixture of my ideas, Opus' ideas and our push-back on "each other".

- Now I have a multi-step plan, with at least 8 distinct stages to implement this massive feature which I know for a fact would have taken me weeks to implement, and I expect to implement it in at most 3 days, but very likely it will be a day and a half.

Final context (with regards to your Claude Code question): My main development environment is Cursor, though for personal projects I also use Codex and Claude code. For the initial "researching of the feature in my off-time" I often have interactions with ChatGPT and Claude where they have no access to the codebase, and I have them go find out what the state of the art on specific topics is. All of these interactions also involve me using my voice to talk to them (though nowadays I don't typically use their voice mode, I just let them reply in text). Then I brood over that.

jorl17

vorige maanddiscuss

While I think this is true

> If you use GenAI on things that you couldn't approach alone, it's an incredible tool.

I think this isn't true in all cases

> If you use it on stuff that you're pretty good at, it's not a gamechanger (and if you're an expert, it's a minor boost at best).

I think even then there's a divide.

I mostly work greenfield projects (and love it!). For these, AI has been a literal game changer. Our projects are built faster, with one or two orders of magnitude more automated tests, and all quality metrics are up.

Meanwhile, nearly all of my friends complain that AI doesn't help them. But they mostly work in very large existing codebases.

Still, even in large projects I think AI (the expensive variant) has been a complete gamechanger for me. Sure, I spend a lot on tokens, but I just feel happier and enjoy what I do more. The singalong people say about "thinking at a higher abstraction level" is what I feel. I really am thinking about architecture and larger patterns, instead of the boring nitty-gritty (which wasn't boring at all when I was a kid learning to code!...)

I think a key factor in all of this, to me, has been dictation. Most of the time, I don't write -- I use voice-to-text. I don't even read what comes out of it -- the LLMs get it (it is mostly unintelligible to anyone else) .

This means when I'm planning a big feature, I give a gigantic brain dump to the LLM in perfect stream of consciousness way, going through ideas, pros and cons, edge cases, what exists, what doesn't exist, where I'm sure of something, where I'm not sure and want the LLM to browse the state-of-the-art. Sometimes I spend 20 minutes just talking to the microphone before I send the first prompt. When I pair that with Opus, I find that I am able to build much faster and to go through alternative designs much more frequently as well.

I keep trying to tell all my friends: use voice to text and braindump to the computer. But they refuse... I couldn't imagine having to type everything nowadays. Even though I'm a fast typer, it's still much slower than the speed of my thought, which, granted, is still faster than the speed of my voice.

In effect, I filter much less, but I've come to think that's positive for the good LLMs: I throw all the edge cases and what ifs I'm thinking about -- all those years of experience dealing with similar systems.

If I wanted to go back to work in-office, that would be my major problem: I need to be able to talk with my computer all the time, loudly, and pacing through my room.

jorl17

vorige maanddiscuss

Hear, hear.

Back when I attempted to be an edgy college guy, I carried my Gentoo with green foreground on a black background. EVERYWHERE. My pcmanfm (yes, I was one of those) looked glorious in true matrix style (and I did have a matrix screensaver). I did it because it was "cool", not because I felt that dark mode was better.

Then when I changed to MacOS, since there wasn't native dark mode, I don't think I ever _thought_ of even changing it. Things just looked great and I had no complaints.

And, as I've aged, dark mode started to actually hurt my eyes.

There's a special kind of dark mode which I can never put my finger on and literally makes my head hurt. I can feel my light adjusting itself to the change and the blurriness settling in. Every letter seems to transfer some of its weight onto neighboring letters, even those in a previous paragraph with quite a large vertical gap! I don't see the letters overlap but it's like my mind is telling me that they ARE overlapping. It's bizarre but it's the best description I can give: my brain is convinced they overlap, even though my eyes disagree.

I can never focus on those websites and have to quit immediately. Usually pitch black is bad, but I've seen some websites make it work. I once read it has to do with astigmatism and ever since then I've paroted that, but I have friends with astigmatism who scoff at my white/light mode.

Examples (most of them from random googling):

- This is terrible. The green makes my head hurt and it's very hard to focus on it. https://anilkody.framer.website/?ref=darkmodedesign

- This is ok. https://www.danielsantos.co/

- Our company's website is fine https://www.cron.studio/

- Also ok https://superset.sh/

- Unbearable: https://www.omnius.so/web-development

I'm a guy who loves to have the maximum brightness, and incredibly bright lights. I'm not kidding: if I work against a black wall I'll go crazy. Back at my parent's place I pointed 4 different strong ceiling lights at the place where I used to have the computer to make sure it was LIT.

It sort of sucks because there's an increasing amount of dark-mode only websites and I've had to occasionally apply custom styles to them just to browse...

jorl17

vorige maanddiscuss

This is exactly it.

I've said many times that I believe "using the computer will transparently involve having it write and run code for you" (and if you're not technical you won't even know it!). What you're saying goes in that direction as well.

I feel that it's often better for us to create purpose-built tools for our lives, and with every model release, the complexity of those tools grows.

These are really personal tools: they solve a problem that other people might have, but are very tied to your own specific way of working, and would be hard to explain or adapt to someone else. So: shop jigs.

I have about 10 custom scripts and programs that are like this -- I haven't felt like this since college! Back then I had all the time in the world to customize my setup...now I have agents!

In a way, I want to show this to all my friends, but whenever I mentally trace how that would go, I realize they wouldn't really understand a bunch of the quirks they have, because they are _my_ quirks. They're reasonably complex pieces of tech that solve my problems very well, which are themselves particular versions of broader problems, and which I (at least for now) have no interest in supporting.

It's so clear we're heading in this direction, and yet so many people still believe code will be for the elites. Maybe production-code...As for the rest, I think soon your mom and dad are going to have their computer running code it wrote to serve them. Security-wise it's scary, but it's exciting to think about!

comments