Julian Martinez

@AndrewHilaryUS me smoking my first newport at 15

The cigarette I smoked after my 15 hour flight to Australia was top 5 highest I’ve ever been. I had to sit down and hold on to a bench so I didn’t shake loose from the face of the earth a drift into outer space.

@SokeyeA just follow @leftcurvedev_

You actually need to be unemployed to catch up with AI No jokes

@leftcurvedev_ best series on x rn

🥊 Round 5 Qwen3.6 27B vs Claude Opus 4.5 Came up with a new prompt to settle the fight 🦩 "Salt Flat Mirror" challenge (inspired by Salar de Uyuni) One HTML file, full-page canvas, no libraries. Why this challenge is interesting: Perfect dual coherence. The sky and its reflection must stay perfectly aligned. The flamingos need to appear both above and below the horizon with accurate reflections. Most models either simplify the birds or completely skip the reflection. A genuinely tough final-round test! — 1st run [115s and 68s] 27B... No real depth, broken clouds, and the reflections were decent for the flamingos and clouds but failed to mirror the sky’s colors properly. The canvas almost looks broken (it isn’t) Opus... 🤯 The only real flaw is a duplicated sun reflection on the horizon. Other than that… my god, it’s beautiful. — 2nd run [108s and 66s] 27B... Way better! Depth, solid clouds, and proper reflections this time! However, all the flamingos are flying and the sun is having a seizure 😅 Opus!! 🚀 Nailed it twice in a row. Literally nothing to say, a total beast on this challenge. I’m genuinely impressed. Look at this water reflection.. — 🏆 Claude Opus 4.5 is the winner After five challenges, one thing stands out: even when Claude’s output has broken elements, the code structure and overall quality are clearly superior to the 27B. You can feel it. That said, Qwen 27B still delivered some seriously impressive results; especially when you remember it’s running locally on a heavy quantization (UD-IQ3_XXS) Yes, Opus 4.5 is a level above. But the most insane part? Anyone with a 16GB VRAM card can run the 27B. Fitting this much intelligence into that amount of VRAM is crazy! Wrapping up this fight, what did you guys think? What should the next matchup be? 👀 ↓ All the other challenges under this post

@leftcurvedev_ am running ollama with cline and now it’s impossible to go back to claude code

local ai haters in disbelief right now

@cassidoo @kalepail I firmly agree I spoke to someone today who couldn’t see the value in having qwen run locally when they could just use claude code. I had to remind them that these companies are for profit and eventually will likely be unaffordable for regular people

idk how else to say this, but... build your dream projects now. I feel like all the tools are giving away a LOT for free/cheap now. It's only gotten more pricey over time, and will keep getting more expensive. Your ideas are subsidized now, think of it as a fire sale and build!

Built an agent that bets on bitcoin automatically using ai watches price, runs grok, places real money on kalshi. no human needed. github.com/Julian-dev28/s…

@Syraavibes it’s too much get rid of the tv and replace the the pillow cases / bed sheets with a single mexican blanket from 1998

Men, be brutaly honest. is this enough?

@unusual_whales LFGGGGGG

BREAKING: A jury has found Live Nation and Ticketmaster to be an illegal monopoly that overcharges fans, per MorePerfectUnion

@SirmonyD001 Number…string!?

Random person: Age is just a number. Tech person: Actually it’s a string.

Most AI trading tools are a Magic 8-Ball with a GPU. I built something different. Sentient ROMA @SentientAGI runs a 6-agent pipeline on live @Kalshi BTC prediction markets — every 5 minutes: → Stage 1-2: deterministic data (no LLM) → Stage 3: sentiment on @AnthropicAI claude-haiku /@OpenAI gpt-4o-mini /@xAI grok-3-mini → Stage 4: ROMA recursive solve on claude-sonnet / gpt-4o / grok-3 → Stage 5-6: deterministic risk + execution One env var to switch providers. Full repo open. Repo: github.com/Julian-dev28/s… The framework goes further. Full breakdown: x.com/j_dev28/status… @SentientAGI @AnthropicAI @OpenAI @xAI @OpenRouter @huggingface @Kalshi

been building with a lot of agent frameworks lately and ROMA from @SentientAGI is the first one that really stuck with me you give it a goal and it handles the rest Atomizer → decides if the task needs decomposition Planner → breaks it into typed subtasks Executor → runs each one (in parallel when possible) Aggregator → synthesizes the results Verifier → validates the output against the original goal The whole thing sits on top of DSPy, so every stage is independently swappable meaning different models, tools, or strategies per step. just shipped an examples repo with four runnable demos. works out of the box with @OpenRouter @xai @OpenAI @AnthropicAI @GoogleDeepMind @FireworksAI_HQ drop in a key and run! github.com/Julian-dev28/r… throw something multi-step at it and let me know what happens

@naiivememe it’s me..i’m bro

calming down bro after buying fomo silver at $123 and now it’s $88

@unusual_whales everyone who bought “no” on polymarket

BREAKING: OMB confirms government shutdown starts at midnight

@chimpfone absolutely riveting work truly profound my life has changed for the better after reading this

>Wake up. >drink large 20oz coffee. > take 10 nootropics. > take nasal spray of more nootropics. > inject some peptides. >swallow 40 vitamins. >take some lions mane. >crack open energy drink. > watch Opus 4.5 code for 8hr. >talk to nobody. >go to sleep. Life is good.

You can just do things Polymarket agent? Go nuts. It'll take you like 20-30 minutes to get it rolling github.com/elizaOS-plugin…

🏆 WE DID IT! 🏆 Yamata just took 1st place in the Next-Gen Infrastructure track at the @okx @OKXDevs ETHCC Hackathon! Outstanding comprehensive trading ecosystem with impressive SDK upgrades. Multi-chain support across ETH, Polygon, Arbitrum, Optimism, and more. Thank you @DoraHacks for the incredible platform and to the judges for recognizing our vision of trading without trade-offs! This is just the beginning. The crypto DeGeneration deserves better infrastructure, and we're building it.

@dom_kwok that's a funny way to write 225 🤔

every 33 year old male should be able to bench 135 lbs

@EoliaWallet Congrats! Looking forward to what you build in the future

We won the hackathon! This is just the beginning of our journey.