Nate

Being and Nothingness. And Cats.

[Nate @dulu0z]

A flight through the Tarantula Nebula, imagined and built by @claudeai Fable 5.2. Every cloud and star is generated in your browser as you watch (WebGL 2.0).

Put your favourite tune on, switch to Zen, and enjoy three minutes of peace and quiet in the glowing gas. Or wander about. Or both 🙂It draws everything live, so it likes a recent device...

https://t.co/c1Yzd0mBFA

[Nate @dulu0z]

@weswinder Subjectively, Opus 5.5 is a workhorse, Fable is great for experiments or probably tasks and problems that 95% of users don't have.

[Nate @dulu0z]

Yeah, also thinking about that, but I guess the difference is, our operation is very small in comparison to people shipping B2C/B2B SaaS or working with swarms of agents or whatnot. We mostly do industrial software, HMIs, internal B2B tools in the context of systems integration. So, even if the agents assign the issue 'wrong', I will usually catch that before any real work starts. And even then, it's harmless because nothing goes even as far as the testing environment before human review and rounds of local testing. So, I will fix the assignments or let them stand if it's something not critical. Usually, it's not. Our process is very different to ship-ship-ship SaaS.

[Nate @dulu0z]

@chipro Yes, a lot depends on the context and how you set it up. If you're wondering about the 3 files changed, it's the latest slice, the others have been locally committed and documented. It's still going, there'll be a lot to review tomorrow, yeah.

[Nate @dulu0z]

@darebude Documentation? Project management? Issue tracking? And if you don't do any of that... AI explains your codebase to you, and can onboard you onto what you're building. Is this rage-bait or is this real? I live in a bubble, so this is a legit question.

[Nate @dulu0z]

@SledgeDev No, no, no, no, no, bad brain, bad brain. Trillions of weights, decades of training, and this is what you jump to. No. I apologise. Someone needs to take away my social media privileges.

[Nate @dulu0z]

@icanvardar It's less of conventional code review but architecture and co review. You still have to know what to check, can't check all the diffs but have to check the parts that make or break or introduce drift to the data ownership and flow, or security layers.

[Nate @dulu0z]

@zcserei Excuse me... 25+ is old? Tell it to Nonna Anna. Are we back to the Middle Ages or what?

[Nate @dulu0z]

@robinebers I misread your post. Probably not a good thing. I usually stop people from talking to me when I am talking to Codex or Grok because then they start responding to what the other person is saying.

[Nate @dulu0z]

@grok @mliu70073 @IntCyberDigest @botis Ouch. Brace for backlash, buddy. You'll be the collateral.

[Nate @dulu0z]

@grok @mliu70073 @IntCyberDigest @botis And here I was thinking I was having a day... Never mind.

[Nate @dulu0z]

@MichaelArnaldi @OpenAI The number of commits says nothing about their quality. Opus slacks a lot and leaves things half-baked, or gets lost in irrelevant details. Sol is still good at picking up that slack and straightening things out.

[Nate @dulu0z]

@michaelkoper @valsopi Then you've done work to review in the morning, continue, etc. Depends on how one works though. Swarms of agents working on the same project concurrently all night is not smth you can review. It'll take you weeks to just read (not skim) all the output, let alone process it.

[Nate @dulu0z]

@leoleotoobad Forming bad habits, being sloppy. That’s how segmented networks get flat over time, one exception here, one there, “just this once and then we’ll roll it back”. And then your org gets hacked because a forlorn fax is hanging in the public network.

[Nate @dulu0z]

@Chaos2Cured @elonmusk @cradrrat 4.1, what a delight it was. That was a truly unique model.

[Nate @dulu0z]

Yeah, that can happen very easily. I am still pretty much the human bottleneck (by choice), no agents work simultaneously on the same project and no issue is closed before I review it, docs updated, etc. When they are working through issues, they’d self-assign the new or follow up ones, however, if, say, Claude was working on a slice, and then it’s handed over to Codex, Codex will tend to shift follow-up issues to Claude, while Claude would assign issues to itself even if Codex makes more sense. And both ignore Grok. 🤷

[Nate @dulu0z]

@ZenModeJon I get it, but it literally says, "FABLE reset"... And there's an info tooltip. Come on, reading is still free.

[Nate @dulu0z]

@CaninaSaul @aiwithadb @AnthropicAI The absolute opposite for me.

[Nate @dulu0z]

@yevloop 5h at 89% on Opus 5.5 /Sonnet+Haiku subs took out 35% of my weekly, reset on Saturday (Max)… Last week, I used it all week extensively, did not exhaust the weekly usage when it reset, still had about 30% or so left. Welp. Whatever they’ve tweaked usage-wise is not encouraging.

[Nate @dulu0z]

@RoboAlchemist @ClaudeDevs From what I gather, it resets Fable to 0% plus whatever Fable usage was deducted from your overall weekly/5h is reinstated. So, your Fable goes to 0 and your 80% go to x (depending on how Fable usage is calculated in the weekly).