lexomis

me like to build

[lexomis @lexomis]

Sonnet 5 gonna be the Otherside-ing of Anthropics goodwill in the community.

[lexomis @lexomis]

Sonnet before I started overthinking everything and just turned into a worse Opus was a really good model (4.4/4.5 ish).

Since then I can't stand how it codes.

[lexomis @lexomis]

Sonnet 5 just gonna overthink everything like Opus does. But slightly faster.

Please labs I beg you. Stop thinking the only customers want to do is 4m goal turn loops.

Give us fast, light weight models that just do. Well. And don't overthink everything.

[lexomis @lexomis]

i'll be paying anthropic less money next month.

so should you.

[lexomis @lexomis]

I love how insightful Armstrong thinks he's being when he says the most banal shit.

[lexomis @lexomis]

If I was CMO of Ollama I'd be having a field day with all this closed source access drama.

Oh and y'all really should get an Ollama subscription for half of one of your max subs on the closed source guys.

[lexomis @lexomis]

In my build system I can switch models with a simple flip and not lose context. I keep five tuned models (3 open source, 2 closed) for all 7 of my primary agents.

Y'all need to stop being so married to one company. If you're gonna marry something, marry the tech.

[lexomis @lexomis]

use the agents for what they're good at. but, as an engineer or a designer or a business human or a lawyer or a marketer or whatever it is you do, don't forget to use yourself for what you're good at.

otherwise just /goal it and be done.

[lexomis @lexomis]

a loop isn't good because it produces .... things.

a loop isn't good because it produces .... lots.

a loop isn't even good because it produces what you want.

a loop is good when it produces the right thing.

[lexomis @lexomis]

this only happens when you think really clearly about where and how to insert YOURSELF.

not where you insert more models producing more content than you can review. the content needs your judgement. but in the right places

[lexomis @lexomis]

everybody's all goo goo for glm pops but I've been banging the kimi drum for a while now.

welcome y'all to your multi model future.

[Mat Velloso @matvelloso]

Back in my day a thing that could rewrite its own code and access anything on your machine was called polymorphic virus

[lexomis @lexomis]

every time i try to use deepseek i end up thinking .... why.

[lexomis @lexomis]

Honest question. Where can I learn about the methodologies of the various AI benchmarks. None of them seem to actually incentivize anything I actually want.

Is there a comparison site anywhere?

[lexomis @lexomis]

i really wish labs would focus on big context window, lots of inputs, fast processing, a few turns, and crunched output.

so much of what i do would benefit from models optimized for that instead of all these long horizon models.

[lexomis @lexomis]

all of them.

[Ashley Ha @ashleybchae]

Let’s settle this debate once and for all

[lexomis @lexomis]

the irony of it all is that ai agents is what is actually gonna decentralize us so much more than crypto ever did.

when i got into crypto the ethheads wanted to see the end of the cloud. nobody ever achieved shit in that regard. but ai agents are making very real inroads.

[lexomis @lexomis]

judge a trade by whether it followed the plan rather than by its outcome

#til

[lexomis @lexomis]

@tradingview For fucks sake they added goddamn Dexes on Avalanche nobody fucking uses but not the most important dex in all of crypto. Come on man.

[lexomis @lexomis]

It's super duper annoying that @tradingview doesn't have Hyperliquid feeds yet.