am.will

I've been testing Spark agents extensively, and I think I've found the roadmap to using them effectively. The secret isn't the model. It's how you control the context that gets passed to it. Everything I've learned swarming with Spark in Codex. 👇 x.com/LLMJunky/statu…

@Supermicro @nvidia Source: fortune.com/2026/03/19/sup…

Whoa 🤯 @Supermicro Co-Founder was just arrested for smuggling $2.5B in @nvidia GPUs into China. I don't understand how you can be this wealthy, and take such an enormous risk. The AI race is heating up, and compute will be the key to "winning"

@DevinSoto @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI Why not?

@LLMJunky @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI Why?

Finally proud to announce that I've joined the GPU Minor Leagues. 2 x RTX 6000 Pro. I have six months to pay off the second GPU lol. You are all TERRIBLE influences.

@philippberner @amar_patel @XadenRyan @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI Good question. I have a source for discounts too. Your best bet might actually be rtx4000s

@LLMJunky @amar_patel @XadenRyan @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI I have so much excess solar energy.... does that make sense to make some money on the side?

@realBionicLife @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI 7700 shipped 🚢

@LLMJunky @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI How much;)

@RandoCollector @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI Late 2027 at the earliest. Is that right around the corner? 😆 Furthermore prices have jumped 20% since I bought You were saying?

@LLMJunky @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI Terrible investment. The next gen is just around the corner.😬

@karthickdotxyz @cursor_ai Lots do yeah. Limits are lower yeah

@LLMJunky @cursor_ai does anyone really using cursor, their limit are so less right ?

Nice. @cursor_ai just dropped their new "Glass" alpha, and they're leaning heavily into the simplified coding GUI trend that's been blowing up lately. First impressions are really positive. And just look at how insanely fast Composer 2 is. First impressions? Drop yours 👇

@amar_patel I'm glad bro! Enjoy

@LLMJunky Interesting. My codex experience has been total goldfish! Remembers the last prompt but doesn’t have the continuity I’m looking for… I’ll dig into it. I am so happy to be wrong about this Will, it’s been driving me nuts!

Not sure how I feel about this to be honest. On one hand, I will be happy to have GPT Pro in "Codex," but on the other, I've really come to appreciate all the focus and attention they've placed on making a purely swe focused product. Tough call. What do you think?

@SteveDun @crierlon @theo Of course they are This is likely through acp

@crierlon @theo Maybe they aren’t aware of t3 code?

T3 Code now supports Claude. If you have the Claude Code CLI installed and signed in, you can use it with T3 Code. Hopefully the lawyers won't make us remove this 🙃

@theo Lmao crashing out. Love it. This is how I'll use claude henceforth

forgive me not trying to be rude here, but grok doesn't know how codex actually works lol i agree that claude doesn't converse like codex, but it absolutely does have memory, it does save sessions, and the entire session is visible to the agent. i just pulled this random thread from 38 days ago, this thread has 69% of the context window used, so its roughly 180,000 tokens used. still remembers. over a month old.

@LLMJunky grok.com/share/bGVnYWN5… I asked Grok to explain it to you. Like I said, I think this is definitely a limited-time prob, but in my experience over the last 5 days, Codex doesn't remember what you asked it five prompts ago in the same session. It doesn't converse like Claude does.

this is a really good point. this feels inevitable. x.com/DevAdventur3s/…

@DevAdventur3s @claudeai @dkundel okay that's actually a really good point. i bet that is exactly what will happen.

@LLMJunky I’m worried that if they do this, they’ll introduce shared limits across chat and codex like @claudeai does. I would hate that. 🥲 @dkundel is it true ?

@midego1 mad because the fast model is 3x more expensive. which means that if they had simply charged $5/mtok for the slow model, it would have been perfectly acceptable. in other words, you're mad because slow mode is too cheap. can't make this stuff up.

oh i read it. i just thought it was absolutely stupid. every composer model before this was only fast mode by default if they had never released a "slow mode" you would be praising them for reducing the price its less than HALF of what composer 1.5 was. but because they were nice and offered an even cheaper option, you wanna call them criminals. like its so fucking absurd dude. instead of being happy that they: 1) reduced the price by A LOT 2) launched a new slower variant at another 67% discount over that instead, you wanna cry about it.

Composer 1: $10 mtok Composer 1.5: $17.50 mtok Composer 2: $2.50 or $7.50 mtok "Criminal Move" lol You can do everything to make people happy and they'll still shit on you.

@midego1 i'm just devastated. i just scrolled your profile and literally every comment was some toxic nasty bullshit. oh no, please dont mute me!

@LLMJunky Luckily you don’t know me at all. Welcome to the mute list.

@midego1 i understand perfectly well. you're a whiner who looks at most things through a pessimistic viewpoint. toxic.

@LLMJunky You don’t understand but that’s fine. Must be hard to be retarded.

@digitalix @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI thanks man :) can't wait to learn. love the channel. hope you can get your hands on a gb300, you deserve it.

@LLMJunky @TheAhmadOsman @0xSero @ivanfioravanti @ZenMagnets @NVIDIAAI congrats

not a fan of this argument, though i think its funny the reality is the other labs arent stealing data, they're stealing outputs and thinking traces and whatnot. thats more or less stealing the compute used to train the models, which cost a shitload of money money they did not have to spend. china has been doing this for decades i wouldn't mind it so much if they werent actively engaged in corporate espionage globally. wrong is wrong.

@theo @thdxr Can you steal something twice? Ironic Anthropic trained their models on stolen data in the first place.

where the heck is deepseek v4

@theo @thdxr no wonder they're behind

@thdxr They need to steal another 150,000 prompts from Claude first