Dither AI: All You Need to Know

‍Dither AI aims to develop the world’s most powerful AI time series model using blockchain data.

The Time Series Model Challenge

What we find most interesting about Dither is its work on time-series AI models.

Time series analysis has always been important in areas like finance, healthcare, and environmental science. It relies on special statistical and machine learning models to handle data that changes over time.

Transformer models have typically struggled with time series forecasting because:

• A nuanced understanding of time dependencies, lags, and seasonality is needed.
• The continuous nature of time series data, usually marked with timestamps, fundamentally differs from the textual data that LLMs handle.
• The presence of noise and outliers can distort predictions if not properly managed.
• There is a relative lack of high-quality time series data for training.

It wasn’t until February of this year that Google released TimesFM, demonstrating competitive performance in Large Time Series Models (LTM).

Dithering with On-Chain Data

Dither’s solution to the challenges of quality and quantity in time series data is using blockchain data. Blockchains are a natural reservoir of time series data, making them ideal for training foundational LTMs.

And it’s paying off.

‍

The chart above shows how Dither’s model compares to top LTMs using the Mean Average Error (MAE) metric on the Electricity Transformer Temperature (ETT) dataset, a standard benchmark for LTMs.

Remarkably, Dither’s Seer1.5 model has a 30% smaller error rate than TimesFM, despite being 6.5 times smaller.

Why does this matter?

• Time series models apply to many sectors - finance, healthcare, agriculture, etc. Breakthroughs in long-term forecasting would have a significant impact on all these sectors.
• On-chain data, with its sheer volume and time-stamped nature, is perfect for training LTMs.
• The success of Dither’s model against top-performing ones like TimesFM shows that even a small team with fewer resources can achieve competitive results in the right setting.

Seerbot — Your AI Companion in the Trenches

Dither’s first product is Seerbot, a Telegram-based bot designed to send trading signals for Solana tokens using Dither’s time series model. There are two versions of Seerbot: Standard and Premium.

The Standard version focuses on small-cap Solana tokens. If Seerbot detects a high probability of a successful trade within the next 24 hours, it sends out a signal.

These signals include market cap, liquidity, security, clustering, and whale demographics. Seerbot integrates with popular Telegram bots like Bonk, Trojan, and Photon.

Here’s an example of a Seerbot standard signal:

Let’s check out how GRACIE fared.

‍

How did it perform? Not so well, in this case—down only from the time of the signal ☠️. The signal had a 25% chance of bearish movement. So Dither’s AI doesn’t get it right all the time; it’s probabilistic.

A group of community members conducted rigorous backtesting of SeerBot in addition to building a user-friendly front end for analyzing historical classification data. This front end will be released soon. Their findings affirm the unequivocal effectiveness of SeerBot, as depicted… x.com/i/web/status/1…
— Dither AI (@Dither_Solana) 11:51 PM • Mar 10, 2024

Dither published a backtesting analysis back in March 2024.

It’s worth noting that backtesting was done during a bull market, and no similar studies have been conducted since. In the current market climate, many Seerbot signals fall under “Slow to limited movement.”

The Premium version of Seerbot monitors large-cap tokens over longer time frames and offers more detailed entry and exit recommendations. For instance, on August 29, Seerbot issued an exit signal for the NEIRO token, forecasting a 58.09% chance of a 25% price drop, a 41.75% chance of a 50% drop, and a 0.16% chance of a drop greater than 50%.

Example of a premium signal on Seerbot

FYI: Seerbot isn’t meant to replace your judgment. It’s a tool to aid in your decision-making process.

Tokenomics in a Nutshell

DITH was fair-launched in February 2024, during the peak of the memecoin boom. 4% of the total supply is allocated to the dev wallet. The total token supply is 100 million, all of which is already in circulation.

It started trading at $0.080 on Jupiter, Meteora, and Raydium Solana dexes and has since fallen to $0.045 — a fully diluted valuation (FDV) of $4.5M.

🌈 Research-Level Alpha

If you’re into trading low-cap memecoins, Dither AI offers reasonably priced access to its trading signals.

There are two tiers for access to Seerbot. Standard requires a minimum holding of 2500 DITH tokens (~$137.50), and Premium requires 50000 DITH tokens (~$2,750).

Any new features will be beta-tested by Premium holders first

According to their dashboard, there are 380 premium holders with 354 slots left available. You can find out more from their public Telegram group here.

The Team

Not much is known about the team, as they remain anonymous. However, they consistently make an effort to engage with the community daily, and they have a track record of regularly releasing updates.

Our Thoughts

• Time series AI models have the potential to be incredibly powerful when implemented correctly. Just think about the possibilities of making accurate financial forecasts.
• Seerbot has a decent track record, with some users reporting profitable trades from deep in the Solana trenches.
• We are generally not fans of utility tokens that require staking a certain amount to unlock specific benefits, especially for trading signals. The more people who access the same signals, the less valuable they become, which can create friction among token holders.
• With a relatively low FDV, Dither could be undervalued if it continues to generate +EV trading signals—something that will need to be proven by consistent historical performance over time.

If you’re interested in trading algos using AI, we recently wrote a research piece on Taoshi, the algo trading subnet on Bittensor.

‍

Cheers,

Teng Yan & Joshua

‍

This research is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.