Sami Sytelä

Vibe-coding & enjoying the singularity | 3x ex-founder

[Sami Sytelä @Sami_Sytela]

@ns123abc Journalist? Not very neutral one.

[Sami Sytelä @Sami_Sytela]

But the model is only one part of the equation. You need a good harness to get the most out of it. And frontier models are optimized to work exceptionally well with certain harnesses. Not even gonna start with other major factors like hardware optimization. That's why global benchmarks are not very good way of actually measuring the full potential of frontier models.

[Sami Sytelä @Sami_Sytela]

@rezoundous Saving that for the last hour tbh 😅

[Sami Sytelä @Sami_Sytela]

@vansh22b @jonathan_wilke https://t.co/T6quTC3PTW points to a company named C2M Technologies. It seems to be a legit company with turkish founders. The company has been around for over 10 years. That's what I found out pretty easily. You can keep digging if you want.

[Sami Sytelä @Sami_Sytela]

So, I was given access to OpenAI's Daybreak Blue model after verification process that was casually initiated mid-chat while doing routine security review of my app. Anyways, any suggestions on how to get the most out of this model?

[Sami Sytelä @Sami_Sytela]

It's definitely GLM model but something bigger is going on. Maybe an acquisition by Alibaba or Tencent? I originally thought it would be acquisition by Google but that's not realistically possible because of trade-restrictions. Ofc, could be just something more boring like compute-renting deal..

[Sami Sytelä @Sami_Sytela]

@TokenGremlin Yep, it’s related to Sol and sub-agents somehow. Apparently the new computer history mode makes it even worse.

[Sami Sytelä @Sami_Sytela]

@FelipeFr1702 Luna is insanely cost-effective. Highly recommend that.

[Sami Sytelä @Sami_Sytela]

@RealEverNever @thsottiaux I believe Tibo is being sincere here. I understand the frustration but lets try to stay constructive. While Sol has some real usage-issues, Luna seems to be working fine, so I recommend using that while we wait for a fix.

[Sami Sytelä @Sami_Sytela]

@chetaslua This all makes sense if Google suddenly announces they have acquired https://t.co/7sbVVNTFIb.

[Sami Sytelä @Sami_Sytela]

@jonathan_wilke Absolutely love your creativity around this idea. Well done! 🥳

[Sami Sytelä @Sami_Sytela]

Just got my ”security clearance” upgraded on ChatGPT. Tbh, didn’t even know that was a thing. Apparently if you want to discuss about certain cyber topics, you need to go through a KYC verification process first.

[Sami Sytelä @Sami_Sytela]

@niklasdev Did you test other models? If they all produce similar results, then you know

[Sami Sytelä @Sami_Sytela]

@0xLalice @theo We got one for 15M too

[Sami Sytelä @Sami_Sytela]

@thsottiaux I do share the common concern regarding how fast Sol is eating up the usage. Used 20 % of my weekly limit in 2 hours with Sol High orchestrating and using Luna Max as sub-agents. With Luna Max doing everything I can run multiple projects 24/7 and stay at 14 % daily usage.

[Sami Sytelä @Sami_Sytela]

@BoldingBuilds @Da7_Tech I had a similar problem and was able to fix it: https://t.co/0taUDjf9G4

[Sami Sytelä @Sami_Sytela]

@prestonattebery Hey, hey... 40+ is not old! 😅

[Sami Sytelä @Sami_Sytela]

Submitted my first ever bug-report related to Codex, with a solution. If anyone else is having trouble with Codex freezing the entire computer, have a look at reported issues on Github/Codex.

[Sami Sytelä @Sami_Sytela]

@Da7_Tech Uhm, that's weird. I have 5-7 projects running 24/7 with Luna Max and limits are just fine. And I only got 100 $ sub.

[Sami Sytelä @Sami_Sytela]

@rezoundous Thanks to Luna I can now run multiple projects autonomously 24/7 without having usage problems. Codex performance is the real limit atm, but apparently it's getting fixed next week.