Well done @Charles_Leclerc great race, made the tyres last when McLaren couldn’t
1
112
lee retweeted
What a bloody legend @dapperlaughs
This man was born for a time such as this. Laughing AT the bastards is the only way we’re going to make it through what’s coming
Well done everyone buying tickets to go see him. 😘
lee retweeted
Go where the cocks flock 🐔
t.me/barnyardportal
First 300 cocks to like + rt this tweet + join with proof posted in the comments get a special $CAWK 🪂
Drop wallet addy
Just picked up the Air Max 1 ‘86 Royal Blue that was released today 🔵 #nikeairmax #airmax1
1
160
lee retweeted
Today we are migrating to a new Gated Telegram protected by @MatricaLabs
SeerBot gets another upgrade with some new features
The burn and Premium now have official dates: TBA on our spaces this Sunday nitter.cf/i/spaces/1zqKVqbbOoZxB
lee retweeted
Going to launch a Solana memecoin in 24 hrs specifically for designers in web3.
You know what to do
50% LP | 40% PRESALE | 10% AIRDROP
Comment your sol address and retweet to get in
lee retweeted
There are two ways to operate a business: lead the market or follow trends. This morning Dither has done another first.
We have the only vision model operating on @solana and it is trained to distinguish relative quality of tokens
lee retweeted
While the exact Premium release date remains undisclosed, today marks the commencement of onboarding premium users. Individuals possessing over 50k DITH can connect with Mattness on Telegram. This month, our focus is on collaboratively refining the premium experience with your needs in mind.
Seer's capabilities undergo a transformative evolution each month. It's a gradual progression achieved through sustained effort rather than an abrupt change in a single day.
Premium is not a single release. It's not static. It begins with large-cap classifications. It will continue as Solana's premier forecasting AI. This marks our beginning.
There will be an AI semantic sniper to meticulously test signals within the metadata of newly launched tokens & an upcoming utility of burning DITH for both standard and Premium usage, along with other unannounced future releases.
Burning mechanisms will be discussed later today. Right now, we thank our premium holders and look forward to continiously build Solana's most advanced AI trading tools.
Exciting qualifying so far! Come on @LandoNorris let’s get that pole #F1
lee retweeted
A reflection on the the launch of Seer & @Dither_Solana.
Celebrating some of the community's gains and sharing thoughts on a slowly building momentum.
I have @degentokenbase to thank in part for opening my eyes to the "meme coin as trojan horse" model of building.
1/20 Read
lee retweeted
In the past day alone the DITH-SOL & DITH-USDC @Dither_solana Meteora pools have done 20% of all of @meteoraAG ‘s total 24h swap volume 🤯
lee retweeted
SeerBot is beginning to be upgraded. Right now it is UI and backend. Soon it will be AI.
This has been an enormous undertaking to finish building out the integrations for our classification system. Let's take a look at what is coming and what is planned.
We began by training a new modality (time series) for our 2.7B parameter LLM with memecoin data. This resulted in a bespoke embedding layer and a final projection layer back into time series which sandwiches an LLM. We then removed all but the embedding layer so we could embed time-series data into vector space.
The next step involved further training of this embedding model to cluster tokens that would be successful and those that would not or would rug. The method of training is known as triplet loss. It takes one example then compares it to a known positive and negative example. As you can see from the attached training run the model will learn the similarities over several epochs (training runs).
The final step is to find the centroids associated with those clusters. That's jargon for we figure out if new sequences of data are associated with the positive (buy targets) or negative (possible rugs). This is about the most simple classification schema you can get but there are many parts and each set needs to be run, tested and tuned.
Dither project is approaching its gpt-1 moment. With the built out infrastructure, I can go from open source llm to classification model in 8 hours. If I have the pretrained model ready and only am working with triplet loss training then I can fine tune the model in 2 hours. This means everyday we will be updating the SeerBot once the AI is released.
After we have the tuning all dialed in we will make the models larger and more comprehensive. We will soon train an embedding model based on a 7B parameter LLM model. Then we will start working with the larger models for projections instead of classifications. The goal is to iterate as fast as humanely possible on our AI models. Get ready














