AaronCQL

Everybody and their mums building the infra. So who's building the damn apps?

@veerbal01 The "wrapper" that powers the model, like opencode or hermes. My favourite is Pi, where you can write your own custom extensions. Probably better to try out the first 2 first before deep diving into creating your own!

@AaronCQL harness means? any wrapper, custom provider?

The $10/mth plan on OpenCode Go with deepseek v4 flash feels like unlimited usage LOL

@veerbal01 Pair it with a harness you can control/tweak and it's amazing what these supposedly cheaper/weaker models can do

@AaronCQL you were right 👏, today, I tried to same thing (Perpatual Study) with OpenCode - Deepseek v4 pro max, it is pretty good and seems like it explain better than GPT 5.5 xHigh in some aspects loving it 😍

Been talking to a lot of you about how to make Jupiter Wallet even better. You gave feedback, we built it. Here are 4 massive upgrades we just rolled out to level up your experience: 👇

@readylayerone How much usage can you get using gpt 5.5?

@AaronCQL did not know they had deepseek v4 playing with gpt 5.5 on it and it's glorious

@veerbal01 This car wash test isn't a particularly good test of "intelligence" LOL. Even SOTA models trip on this sometimes (even opus 4.7 if you try enough times). Do a real-test of intelligence by replacing your current workflow and see if it fits your needs?

@AaronCQL I'd also like to use it but it's not particularly intelligent and I'm worried it might provide incorrect reasoning. This is especially concerning when I'm engaged in deep serious study or building something critical

Ah yes the enshitification continues. Props to them for making it sound like a good thing tho!

@GoldmanStash My m4 pro isn't fast enough 🤣. It's like 10+ tps only

@AaronCQL Have found the 27B dense with MTP and at Q8 quant to be really good. Day and night difference between the 35b moe

Been testing out Qwen 3.6 35B on an m4 pro macbook running at 40 tps. Not quite fast enough for agentic use cases, and not quite smart enough to replace opus/gpt. But definitely impressive enough that I'd want local/open-source LLMs to win. The more interesting observation is that local LLM architectures are at a crossroad now: dense vs moe. Hardware companies can't seem to decide which to optimise for either. GPUs going hard on bandwidth, Macs and Strix Halo going hard on capacity. Which side are you on?

@SherryYanJiang @rubanlah @brianchew @rachpradhan @sentrytoast Thanks, followed them all!

@AaronCQL @rubanlah @brianchew @rachpradhan @sentrytoast are pretty cool

there are a lot more cool ppl on twitter from sg than i thought, and im glad my tl is just started to reveal them

It almost certainly is this case. As unserious as crypto is, there's no way in hell any protocol/company will allow non-technical teams to push code to critical prod systems.

It's amazing to see the Chinese labs (Qwen, deepseek) doing everything that openai once stood for. How the tables have turned - the Chinese open sourcing everything...

@moothefarmer @kashdhanda @Geninsus It's OK, I'm already getting 16% without dming you

@AaronCQL @kashdhanda @Geninsus I’ll give both of you 10% if you DM me 👀

if you've got $100,000+ to lend out and want to earn ~6% fixed rate on your USDC please DM me

300mg of elemental Magnesium 30 minutes before bed has done more for me in reducing my insomnia and migraines, more than any other things I've tried

The final amount going into your wallet matters the most, not the quoted amount you see on any UI. We have been building our own gRPC & landing infra so that Jupiter's quotes react immediately to on-chain changes & your txn lands ⚡ fast.

@kashdhanda @Geninsus Deal

@AaronCQL @Geninsus i'll give you 8% to not DM me

Help I'm getting mogged by my minister

The current price for the 64gb ram in my thinkpad that I bought in 2024, is literally the same price that I bought the whole thinkpad for... Jeeeeez