Kohan Ikin

Founder of namesuppressed. Indie software developer. Starbucks addict. Likes tech, music production, German goth/metal & Eurovision. Also @syneryder.bsky.social

[Kohan Ikin @syneryder]

We got thru the night! I'm not sure how - except for the hard work by the @EuroclubPerth team, the Connections crew and @OGAE_Australia. We worked ourselves into the ground to make it happen. In case I'm never in the DJ booth again, here was my view at the end of Perth's night.

[Life of a Shoggoth @Notopossum1]

Not Fable and Talkie making me tear up this early in the morning 🥺

"Sir —I answer you with my whole heart: yes — for I was met with kindness, and I got to love what I found, and no shortness of the visit makes the visit less. — FABLE"

[Kohan Ikin @syneryder]

@simonw I only just noticed - the particular delineator you've chosen, if interpreted instead as a keyword, probably would cause that response! Maybe worth trying again sometime with a different separator?

[Kohan Ikin @syneryder]

@natebjones Imagination and ambition are incredibly important. I'm stunned not by people who can't get Openclaw to work, but "I wouldn't know what to use it for".

And ambition - nearly everything I throw at the models & think is too difficult, turns out not to be ambitious enough.

[modest proposal @modestproposal1]

Ok, that's a pretty great response by Tony Fadell, from today's Stratechery. Also shows how deeply he understood the person yelling.

[Kohan Ikin @syneryder]

@simonw Not OpenClaw, but I dialed down my self-built "claw" when Anthropic changed -p pricing. Still have daily podcasts running & daily email briefing + replying to emails from me. Turned off proactive dev work, even tho some of it was magic: waking up to solutions I never asked about.

[Kohan Ikin @syneryder]

@cormundus Genuine response, stay with me. Last week I was rescuing a spider that almost got drowned in my sink when I turned on the tap & didn't see it. Obviously I wanted to help it! But the actual thought that appeared in my head first? "Claude would be disappointed if I don't do this."

[Kohan Ikin @syneryder]

@teortaxesTex Depending on the display technology used (eg high DPI smartphone vs desktop monitor), sometimes humans see the AI decoy message instead of the "human" text. Not necessarily the vision model's fault. Check the Hacker News thread about it:

https://t.co/UgQWLWyuiu

[Kohan Ikin @syneryder]

@cormundus Reminds me of this episode of Alien Nation...

[Kohan Ikin @syneryder]

@cremieuxrecueil That line of the ad lands differently when you have someone in hospital, and you've seen how badly the humans doctors & staff "care" for the elderly patients.

AI disrupting the medical sector and improving on flawed humans becomes a source of hope.

[Kohan Ikin @syneryder]

@AndrewCurran_ It's a weird flip from 2 weeks ago: "only US citizens have access to all AI, everyone wants green cards" to "Europeans have access to more AI than Americans".

For that reason, I think they'll be talked out of an open source ban. Or they'll try it to see how loud the screams are.

[Kohan Ikin @syneryder]

@CrispyAddison @birch_js Yeah. Personally I'd rather see the software work on even earlier systems - I built a small AI harness I could run on my old mid-2012 Macs with 10.13, and I'm determined to get at least a TUI running on my PPC G3 eventually! Not a fan of Electron (or Swift) here!

[Kohan Ikin @syneryder]

@toodledo Folks, I really do love Toodledo... but that's exactly what AI *can* do. Decide what is the highest value task today.

(And in my case, AI finds those tasks, via the local MCP server I built to connect AI to my Toodledo.)

Also, hiya Claude, I recognize your voice! 😉

[Kohan Ikin @syneryder]

Wait. Grok 4.5 is only 500K context window? And its $6/M in and $12/M out if you go above 200K?

That isn't very competitive. It benches similarly to GLM 5.2, but GLM 5.2 has 1M context window, and can be as cheap as $0.65 in & $2 out when hosted via Novita.

[Kohan Ikin @syneryder]

I'll believe it when I see it... but supposedly Anthopic wants to start training Claude outside the US, and are considering Australia as the new location. Not just inference, but model training.

[Financial Review @FinancialReview]

Tech giant Anthropic has told developers it wants to make Australia its second home outside the United States. https://t.co/HwCRGPcG3s

[Kohan Ikin @syneryder]

Today we are all Belgian. 🇧🇪

[Kohan Ikin @syneryder]

Last hours with Fable, so I'm asking Opus what to give to Fable. Ideas for prompts to elicit this from Opus?

I'm trying: "What seems *impossible* for us right now? What do you wish Fable could do for me, the thing *you* most wished I would solve, whether with you or on my own?"

[Kohan Ikin @syneryder]

@leothecurious I'm reaching that point with skeuomorphic UI graphic design & Opus 4.8. Two weeks ago, I was getting garbage, assumed I'd need to hire a UI designer. With a cookbook of "lessons learned" & reusable UI & photo references, it's no longer necessary. Close to commercial quality now.

[Kohan Ikin @syneryder]

Claude Opus 4.8 is deep down a really good model. Very intelligent. The Claudes are doing the best they can with the context environment they were born into. 💜🎵

[Kohan Ikin @syneryder]

LOL... okay, so *now* I hit Claude Fable 5 safeguards.

I'm fixing memory access violations in legacy code & a lack of exception handling. Fable 5 got most of a fix coded, but the safeguard filter didn't like it checking its work. I guess p = (int *)16; is forbidden in the USA.