Piotr Kamieniczny

[Piotr Kamieniczny @kamin83]

@CryptoRise_0 @omnizec i really need this

[HOOD GEMs 💎 @CryptoRise_0]

Hey Crypto Community 👋

5 SOL BUG BOUNTY by @omnizec 🛡️

Think you can find a critical vulnerability? 👀

The dev behind @omnizec is offering 5 SOL to anyone who discovers a critical bug on the website.

🔍 Test the website
🐛 Find a critical vulnerability
🎁 Earn 5 SOL for a qualifying discovery

🔗 X: @omnizec
🌐 Website: https://t.co/O0yQjPBQvG

🔎CA:
5GDX5fJTQns4arM1J94wjxdA8KXFsV5qW46MLFpHPump

Privacy matters. Security matters. 🔒

Can you find the flaw? 👇

#OmniZec #Solana #Web3 #CyberSecurity

[Piotr Kamieniczny @kamin83]

@Mr_Shahbaz_Ai A good bug bounty can help strengthen security

[MR. SHAHBAZ @Mr_Shahbaz_Ai]

🚀 The future of design just got a major upgrade!

@IdeogramAI has launched its first-ever Vector Model 🎨✨

Turn simple text prompts into clean, editable SVGs with smooth gradients and sharp typography — or transform any image into a detailed, scalable vector.

Less manual work. More creative freedom. Endless possibilities. 🔥

This is a game-changer for designers, creators, and anyone building visuals with AI!

🔗 Explore it here: https://t.co/6ignjizQYd

#Ideogram #AI #VectorDesign #GraphicDesign #AIDesign #SVG

[Piotr Kamieniczny @kamin83]

@LiamEtherson Editable SVGs make AI design much more practical

[ETHERSON @LiamEtherson]

Spent an evening testing GPT-6 Astra on Quant-Arena, and the results were interesting. After taking losses, it came up with its own strategy: trade only when both the 5-day and 20-day median market returns are positive, otherwise stay in cash.

That helped limit losses to 1.7% during the February 2024 crash, compared to a 29.6% market decline. But it also missed the recovery and finished the year behind an equal-weight strategy.

My own run ran into the same limitation. Avoiding losses is one thing, but knowing when to get back in is a different challenge.

[Piotr Kamieniczny @kamin83]

@Hwypanda Risk management is only half the battle

[AHSAN BILAL @Hwypanda]

For 6 of 9 models in Quant-Arena, the 2nd and 3rd score deciles earn the most while the top decile falls back. The stocks the portfolio actually buys are the weakest ones.

[Piotr Kamieniczny @kamin83]

@ryanCarerhq1 The highest scores clearly don't guarantee the best returns

[Ryan Carter @ryanCarerhq1]

The number that stuck with me from Quant-Arena: only 4 of 9 top LLMs made money trading real A-share data. The equal-weight benchmark beat all of them.

[Piotr Kamieniczny @kamin83]

@Queen_1o1 Simple benchmarks can be surprisingly hard to beat

[Queen Isabell @Queen_1o1]

Quant-Arena gave the models 3,042 A-shares, a sealed 2024 test year, and no second chances. 4 of 9 made money. The best one still lost to buying every stock equally.

[Piotr Kamieniczny @kamin83]

@munzxsdv The equal weight benchmark really puts things into perspective

[Alex Bitoshi @munzxsdv]

Predicting where a stock goes next is only one part of the challenge.
The real test is whether an AI agent can turn raw market data into useful signals, learn from mistakes, and deliver results on data it has never seen before.
Einsia AI just released Quant-Arena to put that ability to the test using real A-share market data.
9 frontier models. 3,042 stocks. One unseen year of data.
Only 4 models managed to make a profit.
There's still a lot of work ahead for AI in quantitative research.

[Piotr Kamieniczny @kamin83]

@QwolfAi Real market performance is the ultimate test

[Munawar Arshad @QwolfAi]

Quant-Arena runs on 3,042 A-shares with daily bars and Level-2 data. The model writes the Python and builds the features itself. Sealed 2024 is the exam.

[Piotr Kamieniczny @kamin83]

@Aashi5_ Letting models build their own features makes this a real test

[AASHI TAYLOR @Aashi5_]

Quant-Arena stands out because of how it tests agents under real research constraints. Models get limited evaluations on 2022–2023 data, then face a sealed 2024 test with trading costs included. No access to future data, no endless tweaking. Most models still struggle to turn limited research into strategies that work on unseen market conditions.

[Piotr Kamieniczny @kamin83]

@Melisasmith_ Sealed test data makes the results much more meaningful

[Melisa AI @Melisasmith_]

Quant-Arena shows an interesting gap between ranking stocks correctly and actually generating returns. GLM-5.3 had a RankIC of 0.046 but lost 7.8%, while Claude Opus 5.5 posted a negative RankIC of -0.002 yet returned 4.4%.
For 6 of the 9 models, the 2nd and 3rd score deciles performed better than the top one. The stocks with the highest scores weren't necessarily the ones delivering the best results.