Cosmin George

@SBF_Alameda How are you going to resolve the conflict of interest of running your own derivative exchange, AND actively trading against the market at the same time? People complain that @CryptoHayes trades against the market, yet FTX and your shop is out there.

@signulll thats bs marketing, so easy to get into raya. but yes totally degenerate

there are people who have spent years on the waitlist for a dating app whose entire value proposition is that some ppl are excluded from it without realizing they are the ones being excluded.​​​​​​​​​​​​​​​​ absolutely remarkable lmao. my second favorite part is when girls who say they are trying to use it to find their husband… like have you ever legit met someone who got married off of raya?

Keep your Semax, give me some Actovegin instead. Peptides are not a new thing, all of your favourite athletes have been using it for decades, this was all public information, yet nobody bat an eye. Actovegin, which is a peptide extract from CALFS BLOOD was literally in Lance Armstrong his stack all public information Hans-Wilhelm Müller-Wohlfart (nicknamed 'healing Hans') was one of the biggest fans of Actovegin, and the list of athletes he treated was borderline insane: - Usain Bolt - Ronaldo Nazario (R9) - Boris Becker - FC Bayern München - Steven Gerrard - Stephen Roche very potent for endurance, recovery and glucose oxidation (especially for the brain) why does nobody talk about it though?

@MRatable guy is likethe biggest claude shill

You know AI is a bubble because SemiAnalysis is bringing in enough revenue to pay $25m for people and $7m for Claude

@rodarmor goat

I wrote a small program which makes coding agents dramatically better at finding Rust documentation. github.com/casey/cargo-pa… It's called `cargo-path`, and it takes a dependency name and prints the path to the local copy of the source code. 🧵

@synthwavedd bruh the amount of teasing is insane

🚨 Codex app update just added this string: "GPT-5.5 is now available in Codex. It's our strongest agentic coding model yet, built to reason through large codebases, check assumptions with tools, and keep going until the work is done. It uses more quota per token than GPT-5.4, but needs fewer tokens to get the job done." h/t @pattoniumbot

😂😂

@solidintel_x brother what did u do to donnie, delete pic

INTEL: Trump says Iran talks could resume as soon as Friday

@parcadei @Douglance Just wondering, how come you're focused on Claude only? Have you tried or thought about applying same things to Codex?

I haven't laughed this hard in a while, what eval would you have me do for a FastEdit It's literally... Opus: here is edit snippet System: Snippet merged what do you want me to eval hahahaha "i'm claiming that without evals, we don't know if it's better or worse." I've shown you tldr in the wild has saved tokens and people have achieved better output and told you I run evals before launching tools hahaha the "we can't predict how models behave. we're often wrong when we try." isn't true at all lol arxiv.org/pdf/2001.08361 we predicted scaling laws would improve models, we were right transformer-circuits.pub/2022/in-contex… the first big "we found the mechanism for a core LLM capability" result and directly lets you predict when/where ICL will appear in training transformer-circuits.pub/2023/monoseman… Training SAEs unlocked our understanding and improves our ability control LLMs the "we can't predict how models behave. we're often wrong when we try." take is outdated in 2026 the new take is "when we try, even when it's hard or takes a while, we tend to figure it out" lmfao a LLM isn't some magical unpredictable black box genie where you pay with your hopes, dreams and dollars for unpredictable output it's a form of intelligence that requires deterministic tooling and the correct environment to get the best out of it + if we're all honest with ourselves, no singular eval is a universal sign off - if they were, my previous point of 4.7 beating 4.6 wouldn't hold and every benchmark maxxed tool would be the best for everything lmao

Introducing: Continuous Claude v4.7 (optimised for Opus 4.7) strap in - we've got RLMs, 50% off Edits, 95% off Reads, fine-tuned models and even evolving codebases 👀 let's dive in to what's changed, what's new and what to do👇

"Fractional Differencing: How to Save Time Series Memory": "If we don’t difference, the data is non-stationary, and our models break. If we do difference, the data becomes stationary but memoryless and useless for prediction... [A remedy is to] difference on fractions instead of integers." zaurtarunov.substack.com/p/fractional-d…

quantuition-vix.netlify.app I got a little carried away Today and More

@percept_labs LLM slop

TradingView makes you spend hundreds of hours fighting legacy integration debt. Percept is a charting library for trading frontends that works out of the box, with 13 chart types, advanced tooling, and native data access included. If you are building on charts, reach out to us.

@NikLentz just built what ur selling in 10mins using claude code

DSMMA Full Suite: One indicator. That’s it. Everything built in: • Opening Ranges aligned to each market’s specific time • Daily Zones • Carnival Pivots • Settlement-based Std Dev levels • Weekly + Monthly key levels • Short-term trend following & flip points across timeframes It adapts to your chart automatically: 1m → OR + Zones 5m → Zones + Pivots 1H → Full intraday toolkit Daily/Weekly → Higher timeframe levels No switching indicators. No setup. No guesswork. I build the best working indicators on @TradingView. Wasn’t approved for their marketplace for whatever reason, but still available through the DSMMA subscription on our website. DSMMA members still get full access to the entire suite, with guidance and updates daily. 🎥 Full walkthrough below

always keep in the back of ur mind 1. 'new' buyers are always easiest to take out 2. partials are key, dont be a greedy SOB 3. wait for structure THEN react to price as said below, bias for the day is still short UNTIL price takes out mar8th low...

@thsottiaux Codex Desktop hit a bad turn-state bug for me. After a noisy local search, it answered the previous question instead of the latest one. Also spawn_agent gating is hidden in developer instructions users can’t inspect/edit. Exposing active tool policy would help.

Today is OpEx. All the puts below the red line expire worthless today with current spot

@ivan_labrie You deserve more followers.

10 years of timestamped calls. Every link verifiable. 3 BTC cycles, equities, commodities, macro. 👇

@quantymacro tap water king

@bitshine_ wait until you find out I won’t pay £5 for still water

it’s 2026 and only now I’m getting into NN/DL (I know) for personal stuffs and so far I’m using runpod but it’s actually kinda expensive is there a cheaper setup or is everyone just burning their paycheck on GPUs

The senior analyst at a Tiger Cub who trained me on financial modeling back in '08 (shoutout Chris Laporte) walked me through the importance of deeply detailed cash flow builds. So much of the world is focused on EPS (certainly the sell-side), that you can generate a lot of insights on business performance by focusing closely on the cash flow statement. Everyone agrees that Free Cash Flow is the purest driver of net present value, but modeling FCF is a very messy exercise whereas EPS has the benefit of accounting treatment that will smooth out the FCF lumpiness. Thus, much of the world defaults to EPS. Great analysts spend a lot of time to see through the FCF messiness to uncover true cash flow generation capabilities. How much is operating cash flow impacted by working capital changes? How much of OCF is "fake" SBC add-backs that show up in dilution? How much of net income is actually recurring cash charges that are added back to Adj Net Income to flatter headline numbers? How much of FCF is actually distributable via dividend or buybacks? To build all of these analyses into your financial model, there is no shortcut to having 10+ years of historical, quarterly cash flow statements. And this is a super annoying exercise to build, since companies report "cumulative" cash flows in Q2 and Q3, so it requires layering in the 9m cumulative cash flow, then subtracting Q1 and Q2, copying formulas as values. Ugg. I've never found a shortcut to doing this. And while I completely agree that there are elements of financial modeling that help me learn the business, updating 5 years of cash flow statements isn't one of them. I tried this exercise in 5 AI Excel co-pilots. Most failed on this exercise, but I was able to get Scout by Daloopa (not sponsored though do have free access) to do this reliably, with click through visibility. While AI Excel still cannot one-shot a complicated model like my DHR model (not even close), AI Excel can do simple things like update my 5-year outdated BS & CF that save immense time and brain damage. I'll call that a big win. (Will walk through this step by step on my AI Excel seminar April 30th).