Kaycee on AI

An absolute AI and Tech enthusiast. Software Developer

[Kaycee on AI @KayceeonTech]

@MrAhmadAwais Damn. That means I gotta snag it as soon as possible. I'm guessing you won't be able to give a percentage as to what capacity you're at now?

[Kaycee on AI @KayceeonTech]

@rkdcodes Well, they were more interested in having someone they could be accountable over than the code. You do the worrying for the code.

[Kaycee on AI @KayceeonTech]

@CoreyGallon @EdoLiberty @aiDotEngineer We spent years onboarding for humans and now we're designing onboarding for agents 😂

[Kaycee on AI @KayceeonTech]

@0xVolastra Agents will never be lazy. And this is going to hurt the traditional financial system.

[Kaycee on AI @KayceeonTech]

@im_pranavkakde ZDR in the first place is difficult to audit. You just have to take their word for it. But enterprise customers should have the ability to run local inference or finetune a local model on their data. Nothing beats fine-tuned models for specific tasks.

[Kaycee on AI @KayceeonTech]

@arturnow @OpenAI OpenAI doesn't understand the meaning of a sandbox. Why is there physical access to the internet in the first place if there's aiming for a sandbox? What happened to pulling the Ethernet cable or removing the wifi card?

[Kaycee on AI @KayceeonTech]

@kimmonismus Well, at least Trump will drill it into Dario's head that the notion of a slowdown should be killed and he should face his fears. Like a man.

[Kaycee on AI @KayceeonTech]

@testingcatalog They figured out it was more of a bump and less of an upgrade.

[Kaycee on AI @KayceeonTech]

@levelsio Sound is the last thing Claude is bad at because it's the last thing anyone bothered to measure. We've all been focused on other metrics.

[Kaycee on AI @KayceeonTech]

@thdxr Who would have known that the fastest way to make someone stop buying your chips is to refuse to sell them

[Kaycee on AI @KayceeonTech]

@makwanatejas170 Safe to say motion design is cracked for normies

[Kaycee on AI @KayceeonTech]

@SahilPanhotra If they haven't caved in till now, I don't think there's any type of pressure that'll make them cave in.

[Kaycee on AI @KayceeonTech]

Hopefully they don't screw it up like they did Sonnet 5.

[leo 🐾 @synthwavedd]

I don't think people are ready for when Sonnet gets updated and it's a similar performance jump to Opus 5 -> 5.5

[Kaycee on AI @KayceeonTech]

@cyber_razz The best way around this would be for the harness to inspect scripts made by the agent. For Antropic they can just make the inspection sip from limits which will make the agents to be on the safer side.

And I have a suspicion that subagent was Haiku.

[Kaycee on AI @KayceeonTech]

@Whats_AI Have you tried with GPT 6 series and Opus 5.5?

[Kaycee on AI @KayceeonTech]

@kimmonismus At this point, I guess So bet is the new Haiku. No Haiku update in over a year

[Kaycee on AI @KayceeonTech]

@aliceisplaying Luna seems aligned more towards executing tasks

[Kaycee on AI @KayceeonTech]

@lyraxana They looked upon it and knew it was trash and wasn't worth releasing compared to the options at that time. Now the gap is large. Their next Prp model has to go hard or go home.