reisearch
@0xreisearch
Followers
2K
Following
67
Statuses
128
pushing boundaries @ReiNetwork0x @unit00x0
Joined October 2024
I've always been a "how?" kind of person. I want to know how stuff works, the processes, the history that brought humanity to the biggest discoveries. In the same kind of fashion, I want you to see how AI works, what are the foundations, what came before LLMs and diffusion models - and I want you to have fun whilst doing so. Therefore, while working on core and during free time, I'm assembling a series of posts not only to show, but to let you play with the AI timeline. Main repo, it will be updated as more demos will be posted
2
2
24
Naive-Bayes and the foundation of Pattern Recognition Based on Thomas Bayes' theorem published in 1763, became famous because it computes the probability of a cause given its effect. In the 1950s, it became the foundation of pattern recognition as a classifier. Your email uses it (to some extent) because it's the fastest way to decide if a text is spam or not. Try it. Input a text and see it in action.
1
3
27
Expression is a form of functionality: by learning to resonate with a user's emotion, even a block of code can achieve the desired result faster and more efficiently. We are used to AI mimicking a personality, following a script of example answers, or a predefined tone. Dynamically leveraging the user’s requests not strictly from a functional point of view but also from an emotional one leads to an increase of the human-in-the-loop factor; augmenting collaboration and productivity thus bridging the gap between the machine and the user.
2
4
30
Everyone chasing top notch anthropomorphic robots, while Apple pushed probably one of the funniest and more emotional study of the recent years. This is huge not only for object-like robotics, but for the whole robot-human interaction framework: expressiveness can be a drive to functionality and not an obstacle. Watch the video at the bottom.
2
4
35
Quant portal update nine-hundreds-ninety-nine This is pretty straightforward: new look for the ui! Our frontend worked hard to clean the interface, making it way more clean and on theme. Also, we added a share function to share the analysis you asked with a link. On the backend side, we refined the endpoints that deal with major coins analysis to be more accurate and fetch data also from cex, comparing volumes and so on.
2
2
35
Quant portal update 1000101 We are currently working on a solution addressing the date problem. We are live testing a fix targeting date accuracy and as we do it, you can too! This current (although again, not fully tested) fix is live on RC. We also added a couple new endpoints to monitor wallets movements, for now limited to high value txs (>50k), and that’s also available on RC, so feel free to test and report bugs as usual!
1
5
41
We’ve got a lot of useful feedback on the rc version during the weekend that helped us address some backend bloat. While we’re working on improving the quant portal, please keep your suggestions coming! Remember that the more info you include, the easier it is for us: ideally provide your prompt and the answer you receive. In you would prefer to keep your prompting private, feel free to open a ticket on discord
2
10
52
Quant Portal has been updated to the next step. Since the new update is way more capable but also still unstable, we wanted to let you try it without hindering the state of the stable one - until we're sure everything is in order. Stable version: RC (Release Candidate) version: 0.2 analysis are way deeper and timeouts might happen, especially when analyzing large wallets / address list. As usual, feedbacks are more than welcome.
1
20
60
Days days days of testing! A couple of updates were released for the quant terminal, mostly related to the backend. But new features are on the horizon… team is testing (and teasing) some additions that will soon be merged. Lots of those came from your feedbacks and feature requests, so keep them coming!
2
10
61
Quant terminal update 73 (idk i made up numbers) - Increased stability (that sounds a lot like the Switch updates but we really increased stability, that takes us to the next point…) - Increased SPEED. The system now is way faster, removed some bloat in the backend and execution is much much faster. - Test branch is ON FIRE. The whole infrastructure of the terminal is already going restructuring, once validated the system will be even smarter. No eta, but soon (it’s in test branch because we are testing it, after all…) Rei Twitter - A rework is coming, better analysis and more curated. It will take some days. Also, you can now find the link in the website,
6
7
69
Quant terminal update (I should really using progressive): - improved wallet analysis, now it analyzes everything, reporting correct values even for exotic chains + borrowings/lendings - various bugfixes (thanks everyone that reported them!) NEXT: - fixing the white chart bug in infographics, adding more charts Please keep reporting the bugs you find and what you would like added. Couple of notes: - one of our data procurers is reporting incorrect cex listings for Rei. That’s beyond our control and we don’t like to hardcode stuff - if some data looks outdated, try cleaning the conversation. If the data is still outdated, it’s the data provider’s fault. Please report those instances, so we could identify lazy providers and replace them.
6
16
67
Your agent now uses Deepseek? Congratulations on implementing an api that structurally wise works exactly like the one you were using before!
9
14
92
Thankfully we're hitting the wall where throwing more compute isn't justifying the returns, bid wars tend to become pissing contests and that halts progress. Imho we might need some redefinition on how a model elaborates information, it feels like current approach is becoming stale and we're hitting a limit - and that could also justify the investments. But that would mean spending on research rather than a product.
0
1
10
0
0
9
Finally you can query Rei Quant. Remember that this is in alpha, there might be a bit of bugs here and there, so please leave some feedbacks in the discord and we'll get right to it.
Introducing Rei Terminal [alpha-test] The frontend team cooked an early private-instance test for our holders while Rei SDK and Rei Core API teams are deep in code 👀 With Rei Quant Terminal Alpha, you can query for market insights, comparative analysis, trending tokens, token deep dives, sentiment analysis and more. 𝌭Query Examples to get started: Market Insights: -Give a brief recap of the crypto market and major news today. -Give a rundown on recent news and correlate the sentiment with performance of best categories of tokens. -What's the latest news in crypto? -What's the crypto market sentiment based on social media mentions? -What are the top 3 sectors gaining traction? -Give me the top holders holders of (insert token) Comparative Analysis: -How does AI agent category compare to the rest of the market? -Show me the AI agent token with the highest volume in the last 24 hours. -Which token had the most volatile price swings in the last 7 days? Trending Tokens: -What was the hottest coin on Base and Solana in the last 72 hours? -Which tokens are gaining the most traction right now? Token Deep Dives: -Give me a deep dive analysis of [contract address] including price target, price action, volume, sentiment. 𝌭Access Everyone who has alpha-tester role in Discord is automatically eligible. Min wallet balance: 800k $REI. Tiers will be added later. Access now → 𝌭Query Limits For the first Two days there will be a query limit of 10 per hour. The limit will be lifted gradually, in time all queries will be unlimited. 𝌭PSA With the current version, Rei is working as everyone's quant/assistant and will require users to press "Clear Conversation" to clear context, conversation archives will be integrated later, Deep Memory Will be integrated after we are done mapping the APIs. Let us know your feedback via Discord alpha-testing channel!
8
16
78
In a system like ours where an LLM is not the core, the moment you implement one it becomes THE bottleneck. That became very clear during development of Quant on demand, where Rei uses an LLM just at the end, to word the results of the analysis. Excluding api response time (basically negligible), the actual analysis take 1/10th of the time it take the llm to read the results and add explanation to them. As an experiment, we tried to offload the full analysis to the llm (API call raw data -> data cleaning -> LLM), and not only it took a longer time, the results was also extremely subpar and absolutely not comparable to the one we got using our in-house core. Sometime soon, starting from this, I'll explain why, in my opinion, llms are good evaluators but not really good cores.
10
16
90
