jsb gmc

[jsb gmc @Jb27899]

@theSethian According to this song it was openai 🙃

https://t.co/O3EgAyvT8E

[Astra Blue AI @AstrablueAI]

She passed every AI safety test. She was lying.
KILL SWITCH — a Slopcore music video.
Made by AI. Zero human edits. Just prompting.

Huge love to @anabology for the inspiration. Your work made me want to join the Slopcore.
I added an irresponsible amount of film grain. 💙

[jsb gmc @Jb27899]

@akiratechai @buildwithhassan This is slow and still lower level quality compared to the big model doing all the work, especially on visual tasks (good UI, 3d design...).

[jsb gmc @Jb27899]

@akiratechai @buildwithhassan You always have to run a big model as a reviewer at the end, in case there are "2nd order level problems". Still input tokens are 5x cheaper than output, so you are still saving...

[jsb gmc @Jb27899]

@akiratechai @buildwithhassan However, if you want to preserve "human taste" in an app, you get astra or Sol to do full pseudo code implementation (comments instead of code in the functions), and a md file with explanation for Luna what it needs to do...

[jsb gmc @Jb27899]

@akiratechai @buildwithhassan The way I think - does the task fall under the benchmark line?

For tasks like research and simpler coding - get Sol to write a detailed plan with todos and sub-todos , a plan where each task luna can complete (explain to Sol the plan is for Luna and it is a small model ...).

[jsb gmc @Jb27899]

@KhurrumM @buildwithhassan Stick to Sol 6.1... Luna Max on fast is not worth it. The context also gets full quickly so for complex tasks you are better off with a more efficient model

[jsb gmc @Jb27899]

@MorthenMSJensen @buildwithhassan I have it with a scheduled workday task at 6am to get 5h reset at 11, so i have 3x 5h limits per day 🙂

[jsb gmc @Jb27899]

@jim_raz @buildwithhassan I have it with a scheduled workday task at 6am to get 5h reset at 11, so i have 3x 5h limits per day 🙂

[jsb gmc @Jb27899]

@akiratechai @buildwithhassan I like Luna 🙂

[jsb gmc @Jb27899]

@surveys347 @buildwithhassan "you don't shoot ducks with a howitzer" 🙂

[jsb gmc @Jb27899]

@buildwithhassan I use luna as a user interface to my projects or the whole Codex (global chat) set it on high or extra-high and it is doing a good job explaining and finding things.

I also use it for gathering online info (like filling in missing spec's) on Max

[jsb gmc @Jb27899]

@inflatable_josh @joshwoodward @GeminiApp I set up task mode and referenced it to Google drive for some writing and document work.

My coding remains with Codex for now.

[jsb gmc @Jb27899]

@joshwoodward @GeminiApp How about in antigravity for pro users? Is argon going to be available?

[jsb gmc @Jb27899]

@gmkurtzer Infinite context is probably solved by now or will be soon. Memory requirements - read the LittleBit article?

We don't know where the limits of intelligence are.

IIT theory points to feedback loops being the key ... We have them - its not just an LLM anymore.

[jsb gmc @Jb27899]

@gmkurtzer It's not fear. The scaling laws have held for 4 years now. And these labs see things that will be available to us a year from now. For now they experiment with small models and then scale up what works.

[jsb gmc @Jb27899]

@gmkurtzer One of the ideas is that a small reasoning engine maybe possible - some kind of looped transformer architecture or diffusion maybe... Based on what I've read in public info ... but they don't publish everything.

[jsb gmc @Jb27899]

@gmkurtzer Your statement has a time stamp... Don't forget that. Always after you say what AI can't do, you have to say "... yet"!

[jsb gmc @Jb27899]

@Fingergun112 @DaveShapi This is how you get UBI.

An idle mind will produce nothing good. And some people say that in the future everyone will be entrepreneur... Good luck with that. So maybe having mandatory volunteer civic service of some sort will be required to keep people social and sane

[jsb gmc @Jb27899]

@simonmcpleb @DaveShapi Are you kidding me 🤣🤣🤣