Ivthos

A classic American thinker for a new era. I post ideas, rants and odd stories. Pronounced: if those.

[Ivthos @ivthos]

@yannschaub @rezoundous I’ve been using Astra more because Fable cannot comprehend things that Astra intuitively gets.

[Ivthos @ivthos]

@brithume The US used to hang pirates. Just saying…

[Inevitable West @Inevitablewest]

🚨BREAKING: The historic Limoges town hall has just completely collapsed after far-left rioters set it on fire

It was built in 1876. It survived World Wars. It did not survive modern France.

[Kosher @koshercockney]

Holy fucking shit.

Paris right now, this evening.

Police vehicles try to escape with destroyed tyres.

[Ivthos @ivthos]

@thsottiaux What is ChatGPT Space? It's Notion without databases...

[Ivthos @ivthos]

@emollick But that would be like shutting off the entire Internet at this point. 🥺

[Ivthos @ivthos]

A note on my “monkey and lake” comment. I have Claude migrating a database that has been offline for 6 years. When that completes (probably in a day or two) I’ll post some historical context about legal right pushes for animals and bodies of waters. That seems relatable to AI, consciousness and who becomes liable for its actions. Also I dislike “AI” and super intelligence terms; it’s all just technology and technology doesn’t have a conscious silly rabbits. 🐰 (Unlike actual rabbits.)

[Ivthos @ivthos]

I can’t help but think that AI will cheapen life. Super Intelligence exceeds human intelligence. Humans laid off and replaced by AI. At a point it gets depressing.

[Ivthos @ivthos]

@BrianRoemmele @BigelowInst I’m reminded of past times where people tried to get a monkey and a lake declared as human so they could inherit and own property (in the UK think). Maybe if the monkey or the lake had a bunch of GPUs they would have qualified as human then? Eh, probably not.

[Ivthos @ivthos]

@dory_benami @BillAckman One down another 45 million to go. Could take a while. 🤔

[Ivthos @ivthos]

@jaystratton @HarperCollins @WmMorrowBooks I preordered the Audible version today 👍

[Ivthos @ivthos]

@elonmusk How is this not a religion at this point?

[Ivthos @ivthos]

@emollick I re-optimize my prompts for every new Opus model. A prompt for 4.6 didn’t work well for 4.8.

[Ivthos @ivthos]

@emollick X feels like it is practically all bots. *sigh*

[Ivthos @ivthos]

The number of times Grok Bot or Dot woke up and tried to sell me things is zero.

[Ivthos @ivthos]

At what point is Meta Muse's proactive nature just advertising... It knows I have 6 libraries from Fracture Sounds, but still. @Fracture_Sounds @Meta

[Ivthos @ivthos]

The desire to refer to Meta Muse as Metamucil is so strong.

[Ivthos @ivthos]

I rolled back the sandbox permissions. Not my ideal fix but will work for now.

[Ivthos @ivthos]

Codex is trash again... I asked it to review a file, and it asked for permission to read. Agentic agent down...

[Ivthos @ivthos]

I built a similar research desk with 12 Grok bots. Ultimately I scrapped it for one Claude Cloud Routine.
Negatives to my Grok bot pipeline:
- didn't scale (requests had to get queued up by a front desk bot), at least 5 bots were involved for every request (subject experts were pulled in as needed)
- difficult to reproduce the setup or transfer it to a different cursor account
- Grok's MCP protocol has an annoyingly low limit at 40 KB limiting it's ability to write large reports (had to use APIs, python and manage an .env)
- I ended up needing a 3rd party auditor bot to grade the efficacy and clean up mismatched bot instructions as my manager bot couldn't track everything

Meanwhile, the Claude Routine is git managed, can run in parallel and is reproducible. I can check the routine's quality by running different versions and comparing outputs. I use Grok bots for long running conversations but not for automation since a well constructed multi step routine could handle that better. Claude's report was also many times better as Grok got stuck in stereotyped roles for the type of research it was to do.