Xavier Lyu Profile Banner
Xavier Lyu Profile
Xavier Lyu

@xavi3rlu

1,045
Followers
584
Following
25
Media
103
Statuses

research @opentensor prev @panteracapital quant + fundamental

Manhattan
Joined March 2021
Don't wanna be here? Send us removal request.
@xavi3rlu
Xavier Lyu
3 months
life update: i recently left @panteracapital and joined @opentensor research team excited to dedicate all my time pushin' Ο„ with @const_reborn and @shibshib89
Tweet media one
18
18
182
@xavi3rlu
Xavier Lyu
2 months
Shining a light on three #Bittensor subnets that are driving cutting-edge research, all while being tested against rigorous scientific benchmarks. This is the innovation @opentensor loves to see πŸ‘‡πŸ§΅ (0/7)
9
60
231
@xavi3rlu
Xavier Lyu
1 year
Knowing other traders’ liquidation levels can be advantageous, similar to knowing other’s stop losses, so how do we surface such insight on @GMX_IO This 🧡summarizes my 2-part Medium posts where there will be more in-depth w/ full python code 1/9
11
22
117
@xavi3rlu
Xavier Lyu
1 month
#Bittensor subnets should focus on two key objectives in order: 1st: Optimize quantifiable metrics that exceed current benchmarks and advance the SoTA in research 2nd: Once this foundation is established, develop a viable, profitable product that attracts paying customers in the
@0x_Kun
Kun
1 month
@degentradingLSD The thing that wasnt addressed which needs to be addressed is whos paying to use it? You have to both beat the competition then you have to have enough people paying for it to be profitable Chatgpt is burning billions a year just to try win and get better and it not profitable
6
1
18
5
6
73
@xavi3rlu
Xavier Lyu
8 months
Running a node in @bittensor_ offers a differentiated way to impact the evolving field of decentralized AI, but it requires careful understanding and planning. Here is a practical guide on navigating this process πŸ‘‡ #bittensor
1
8
35
@xavi3rlu
Xavier Lyu
2 years
@guil_lambert left the following exercise in one of his Medium post. My solution here πŸ‘‡πŸ§΅ Nothing crazy, maybe it can help you with your financial engineering pset :P
Tweet media one
5
4
33
@xavi3rlu
Xavier Lyu
3 months
Tweet media one
3
0
21
@xavi3rlu
Xavier Lyu
2 months
(SN31: 2/2) SN31 is benchmarked against the well-studied CIFAR-10 dataset and they have produced models that outperform most SOTA models. SN31 plans to publish this result in a conference soon. If you would like to access their paper, please contact SN31 through Discord.
Tweet media one
1
1
15
@xavi3rlu
Xavier Lyu
2 months
(SN31: 1/2) SN31 (NAS Chain) incentivizes miners to discover the best neural network architectures that maximizes accuracy and minimize hardware-dependent characteristics such as # of parameters and FLOPs. The goal is to maximize the Pareto optimality between these variables.
Tweet media one
1
0
14
@xavi3rlu
Xavier Lyu
2 months
(SN34: 1/3) SN34’s ( @BitMindAI ) goal boils down to predicting if an image is real or fake on a scale of 0 to 1. Currently, there is a lack of open benchmarks for general synthetic image detection. DeepfakeBench is specifically used for faces. SN34 uses the BLIP-2 VLM to generate
Tweet media one
1
4
13
@xavi3rlu
Xavier Lyu
2 months
(SN34: 2/3) SN34 uses the base model architecture from DeepfakeBench and optimizes it against the dataset above. The team also introduced a novel approach called Content Aware Model Orchestration (CAMO) that enhances this by routing images to specialized expert models, improving
Tweet media one
1
1
11
@xavi3rlu
Xavier Lyu
2 months
(SN34: 3/3) SN34 is integrating more SOTA models like FLUX.1-schnell to perfect its synthetic image detection benchmark. Plus, new dashboards and graphs will soon help miners track their performance.
1
0
11
@xavi3rlu
Xavier Lyu
1 month
@xponentcrisis > Bittensor’s Discord is the perfect place for anyone with ADHD lmfao
1
0
11
@xavi3rlu
Xavier Lyu
3 years
On my flight to #AvalancheSummit , the guy is front of me is explaining Bitcoin to an old Spanish lady. The world is great
1
0
8
@xavi3rlu
Xavier Lyu
2 months
(SN40: 2/2) They evaluated the effectiveness of different text chunking solutions in a RAG setup, comparing SN40’s miners against Unstructured IO’s partitioner and AI21’s text segmentation API on the GPQA benchmark, which tests with graduate-level Google-proof questions.
Tweet media one
Tweet media two
1
0
8
@xavi3rlu
Xavier Lyu
3 months
0
0
8
@xavi3rlu
Xavier Lyu
2 months
(SN40: 1/2) SN40 (Chunking) incentivizes segmenting text data (with plans to expand to more media types) into the most semantically meaningful chunks to enhance performance in RAG pipelines. Currently, validators generate synthetic queries by selecting featured Wikipedia
1
1
8
@xavi3rlu
Xavier Lyu
2 years
It’s all because of video games smh
Tweet media one
0
0
7
@xavi3rlu
Xavier Lyu
1 month
@mxmsbt Would love to see SN44 disrupt Opta. Tons of teams, esp. youth, use Veo or Trace cameras -- imagine feeding that game footage to SN44's miners to generate advanced stats that only Premier League teams can afford. That would be dope!
Tweet media one
1
1
7
@xavi3rlu
Xavier Lyu
1 year
@tradeparadigm πŸ™πŸ™ and shamelessly shilling my tweet
@xavi3rlu
Xavier Lyu
1 year
Knowing other traders’ liquidation levels can be advantageous, similar to knowing other’s stop losses, so how do we surface such insight on @GMX_IO This 🧡summarizes my 2-part Medium posts where there will be more in-depth w/ full python code 1/9
11
22
117
1
0
5
@xavi3rlu
Xavier Lyu
1 year
2/9 Quick overview: leverage trading is a double-edged sword. It allows traders to amplify their exposure to an asset with borrowed funds, increasing potential P&L. Liquidation occurs when a trader's losses deplete his collateral, triggering an automatic closure of their position
1
0
6
@xavi3rlu
Xavier Lyu
1 year
it's interesting to see how the #FTX Exploiter routed through multiple wallets to get to #THORChain could there be another reason besides making tracing harder? 0xcbce eventually bridges to Threshold's tBTC, i didnt graph everything as it is getting too big and messy 1/2
Tweet media one
@lookonchain
Lookonchain
1 year
After @THORSwap suspended swaps, FTX Exploiter started swapping $ETH for $tBTC and cross-chained to the Bitcoin network through @TheTNetwork . So far, FTX Exploiter has swapped 75,636 $ETH($124M) for BTC assets and cross-chained to the Bitcoin network.
Tweet media one
Tweet media two
Tweet media three
Tweet media four
36
44
223
2
0
5
@xavi3rlu
Xavier Lyu
2 years
@ribbonfinance bae wake up, @ribbonfinance just dropped a new strategy!
Tweet media one
0
0
4
@xavi3rlu
Xavier Lyu
2 years
When Voyager said ur deposit is FDIC insured
Tweet media one
0
0
4
@xavi3rlu
Xavier Lyu
1 year
This is just a glimpse into the world of margin trading on DEX’s and there’s a lot more to explore. Stay tuned for part 2 with full python code and more related topics. Follow me on !
1
0
2
@xavi3rlu
Xavier Lyu
2 years
is there a way to see risky positions on aave/compound (eg. positions that are very close to liquidation). Similar to how @LiquityProtocol has their risky troves page:
1
0
2
@xavi3rlu
Xavier Lyu
2 years
yes, ive always wanted to go viral on @CitizenApp #nyc
Tweet media one
0
0
1
@xavi3rlu
Xavier Lyu
1 month
@JosephJacks_ 100% agree Allocating a validator’s weight is synonymous with a venture fund allocating capital
0
0
3
@xavi3rlu
Xavier Lyu
2 years
@chiajy2000 Introduction to Algorithm CLRS audiobook
1
0
3
@xavi3rlu
Xavier Lyu
2 years
"Ya the new NFT project is called AptosVerse. We are a revolutionary virtual real estate project with funding from top VCs like 3AC and Justin Sun. Currently no on-chain volume but the floor price is def gonna be 1b"
Tweet media one
1
0
3
@xavi3rlu
Xavier Lyu
5 months
Tweet media one
0
0
2
@xavi3rlu
Xavier Lyu
2 years
1
0
1
@xavi3rlu
Xavier Lyu
11 days
@CryptoDeFiGuy Sick headphones tho
0
0
2
@xavi3rlu
Xavier Lyu
7 months
0
0
1
@xavi3rlu
Xavier Lyu
1 year
also def recommend checking out @hyblockcapital the inspiration of this post was drawn from them, awesome tool
0
1
2
@xavi3rlu
Xavier Lyu
1 year
7/9 With these data, we can obtain a holistic view of the trading activity. Savvy quants can also plug these data into their models
Tweet media one
1
0
2
@xavi3rlu
Xavier Lyu
3 years
0
0
2
@xavi3rlu
Xavier Lyu
1 year
4/9 Therefore, we can use to filter for emitted events like these and parse their data. Below is from Ethereum JSON-RPC docs
Tweet media one
1
0
2
@xavi3rlu
Xavier Lyu
1 year
3/9 When you trade on GMX, at least 1/4 events are emitted: IncreasePosition, DecreasePosition, UpdatePosition, ClosePosition. All 4 serve a different purpose, so we need to retrieve all 4 between a timeframe to get the whole picture. Here’s what 2 of them would look like
Tweet media one
1
0
2
@xavi3rlu
Xavier Lyu
1 month
@angad_ai Thank you ser
0
0
2
@xavi3rlu
Xavier Lyu
1 month
@brodyadreon πŸ”₯
1
0
1
@xavi3rlu
Xavier Lyu
2 years
link to Lamber's article above
1
0
1
@xavi3rlu
Xavier Lyu
1 year
5/9 However, the on-chain data we scraped doesn’t include leverage and liquidation level, but it’s easy to calculate with the info we have!
1
0
1
@xavi3rlu
Xavier Lyu
1 year
6/9 On GMX, your collateral after P&L needs to be at least 1% notional or $5. This constraint needs to hold to not get liquidated, so we need to solve for the lower bound of exit price (p_j). It comes out to this, check out the medium post for full steps.
Tweet media one
1
0
1
@xavi3rlu
Xavier Lyu
11 months
@tokenomacs @stacy_muur πŸ‘πŸ‘πŸ”₯
0
0
1
@xavi3rlu
Xavier Lyu
2 years
@chiajy2000 My guess is that more FANG SWEs check their email when they get to their desk than other fields, plus as a SWE, unless ur a manager, most emails u get are notification emails instead of the ones u need to respond
0
0
1
@xavi3rlu
Xavier Lyu
2 years
@cz_binance non-binding LOI? so like the opposite of what elon musk did?
0
0
1
@xavi3rlu
Xavier Lyu
1 year
@JMSCavalheiro web3py and pyvis
0
0
1
@xavi3rlu
Xavier Lyu
2 years
we start off by translating S_t to its geometric brownian motion definition
Tweet media one
1
0
1
@xavi3rlu
Xavier Lyu
2 years
@CUBlockchain Martin Shkreli
0
0
1
@xavi3rlu
Xavier Lyu
1 year
@ColeGotTweets who u fighting
0
0
1
@xavi3rlu
Xavier Lyu
1 year
8/9 Beauty of DEX’s like GMX is that they provide a level of transparency & accessibility of data that is not possible with centralized exchanges. This allows traders to make more informed decisions, avoid potentially costly liquidations, & explore other’s activities to seek edge
1
0
1
@xavi3rlu
Xavier Lyu
3 years
@AdvayKoranne It’s probably already priced in
0
0
1
@xavi3rlu
Xavier Lyu
3 months
1
0
1