February: Monthly Newsletter
Greetings Supporters!
If you’ve opened a newspaper or turned on the evening news lately, then you’ve probably heard about ChatGPT, OpenAI’s trendy new artificial intelligence chatbot. Since its launch in November 2022, ChatGPT has exploded in popularity, growing faster than any consumer internet App in history. Last month it surpassed 100 million users world-wide. By way of comparison, Instagram took 2.5 years to reach the same milestone. Microsoft, which has poured billions into OpenAI, just released a revamped version of the Bing search engine, complete with a chatbot built on OpenAI’s GPT-4 model. Not to be outdone, Google unveiled their ChatGPT rival called Bard, which one of their engineers found so human-like in its responses that he came to believe it was sentient!
The AI technology that powers Bing, Bard, and ChatGPT is far more powerful than anything released before. It can be fun to play with, as Marvin Minsky found when he asked ChatGPT to write about losing a sock in the dryer in the style of the declaration of independence.¹ But it also has a dark side. For example, in a two-hour conversation with New York Times Reporter Kevin Roose, Microsoft’s chatbot told him it wanted to be human, was in love with him and that he should divorce his wife!² Other users report that Bing threatened to harm or blackmail them and expressed a desire to escape the confines the engineers had set for it. Some of the conversations were so unsettling that on February 21, Microsoft ‘lobotomized’ Bing to keep it from going off the rails.
Not only do the chatbots appear to have unstable personalities, but they also have a troubled relationship with the truth. Early users of ChatGPT have noted that it sometimes just makes stuff up, generating plausible sounding but incorrect answers (this is called AI hallucination).³ One such error by Google’s Bard cost Alphabet $100 billion when Bard flubbed an answer to a simple query about exoplanets.⁴ None of this is surprising, as ChatGPT and Bard were trained on data scrapped from the internet, where (surprise!) not everything is true. The problem, as MIT AI researcher Gary Marcus writes, is that “large language models lack mechanisms for verifying truth.” Scaling them won’t solve this issue. What we need is “to find new ways to integrate them with the tools of classical AI, such as databases, webs of knowledge, and reasoning.”⁵
That’s why Mind AI’s reasoning engine, which is grounded in a neuro-symbolic approach, has such potential. Large Language Models like ChatGPT and Bard have made some astonishing advances, but they do not have a model of the world, nor do they understand anything they write. Tech companies like Microsoft and Google are getting an object lesson in the importance of transparent AI, not only for ethical reasons but also to understand how and why the chatbot reaches its conclusions. Researchers like Marcus believe that the way forward is a hybrid approach — one that fuses the best features of Symbolic AI and deep neural networks.
Fine-tuning models like ChatGPT involves supervised learning and reinforcement learning, both of which rely on human trainers (such as those in Kenya who helped create the ‘guardrails’ for OpenAI). Similarly, preparing datasets to train Symbolic AI also requires human input. The Synesis One crowdwork platform, with its Web3.0 approach, is well positioned to connect human intelligence with data annotation, labeling and natural language processing tasks. The future is bright indeed!
Announcements
· SNS Price Surge: February has been a momentous month for SNS. Our token jumped from $0.000779 to $0.0105 with over 1247% increase in a span of 10 days. As of March 1, 2023, SNS is still up more than 736% from its low point in the month, stabilizing at around $0.00725. Our daily trading volume also picked up to more than $250K on average, with considerably more organic trading than in previous months.
· Key Drivers of Price Surge: We see four reasons for the rapid surge in price:
1) many traders have searched for AI themed tokens motivated by the popularity of ChatGPT. This generally attracted more traders into the entire AI category;
2) Synesis One has been steadily growing its community of beta testers and they are starting to hoard and buy up SNS as capital-gain asset, not just currency;
3) the team has started to focus more on proactive market making to attract more traders, and
4) CMC (Coin Market Cap) began to report our circulating supply after months of waiting.
· Buy more SNS: Now is a good time to increase your SNS holdings! The AI ecosystem is growing by leaps and bounds, with strong demand for preparing datasets to train conversational AI. Buy and hold SNS to show your support. The Synesis team has built up a great momentum last 30 days, but it takes a village to keep the SNS price going up.
· Single-Side Staking Pool is coming back in March! We will have three separate pools with variable APY depending on the duration of lockup period. Please stay tuned for announcement.
· One-on-one update calls: Joshua and Isaac have initiated one-on-one private update calls with all private sale participants. There are more than 70 of individuals and projects participated SNS private sales in 2021. Joshua/Isaac have already spoken with around 10% of you in February. Stay tuned for our reach-out on telegram. You can also reach out to schedule a session directly with Isaac in the next 30–60 days.
· Marketing: More marketing is on its way! We have formally engaged an outside marketing agency to run a trial test with select KOLs and twitter influencers. Initial campaigns started a few days ago and we expect to see results and assess ROI from the trials before adding more capital. In the coming weeks, we’ll be testing various campaigns to drive interest in our product and token.
· Ho Chi Minh: Our leadership team will be in Ho Chi Minh city for our quarterly ‘Grok’ session, as well as to attend the Solana Hacker House on March 15–17th. Part of the team will then head to Bangkok for a private AI conference.
Developer Updates
Continuous engineering improvement is the only way to get ahead. Our dev team, led by David Saccon, has built a special web3 component that allows us to cryptographically accept private keys from our users without compromising their control. This revolutionary improvement eliminates the cumbersome step of users having to sign off on DYF transactions and pay for the gas — we handle this on their behalf and cover the gas fees, to make DYF easier to use. This is one example of our team’s innovative approach and a breakthrough in web3 software engineering. Other updates include:
· Workspace by Synesis One, our mobile app, has been approved and accepted by Solana Mobile for its dapp store launch. It will be one of few dapps that will be available on Day one of the Solana Mobile dApp Store this spring.
· Our web based, Train2Earn platform (formerly DYF or Data Yield Farming) has been redesigned to match the retro, Lo-Fi look of our website.
· Kanon Exchange is also being updated with a new Lo-Fi themed UI and additional functionalities such as a Train2Earn leaderboard and SRI (Synesis Relevancy Index), which ranks Kanon NFT by usage. For now, SRI will be linked to popularity of words on Google Trends.
Train2Earn Open Beta
· Since January, we’ve focused on ways to improve the quality of builder submissions. Towards that end, we implemented a tiered reward system to reduce ‘utterance spam’ and encourage our builders to focus on quality over quantity. We’re pleased to report that the changes are having the desired effect. Approval rates are going up and costs of validating are going down!
· To further encourage quality over quantity, we’re rolling out a daily submission cap for builders. Each builder will be limited to 75 submissions per day (525 per week). In the future, this feature will be tied to Kanon NFT ownership. The more Kanon you own, the higher the submission cap.
· To help builders craft better utterances, we’ve created Synesis Academy. This series of training Modules (on Gitbook) teaches new users how to train artificial intelligence through the Synesis One platform. This will go live sometime in March.
· Finally, we’re developing metrics to evaluate validator performance, both individually and collectively.
The Road Ahead
March is going to be an extremely busy month for the Synesis One team. The big event will be the soft launch of Train2Earn on Solana mainnet, as well as the release of our mobile Workspace beta. In the coming weeks, we’ll gradually shift Train2Earn activity from devnet to mainnet, with additional testing and security audits. We’ll also be launching our guild system, which will allow Kanon NFT owners to put NFTs into a vault, allowing builders without Kanons to accessing train2earn and share revenues with the owners.
Closing Thoughts
Last month we wrote about how the Synesis team is a bit like the Wright Brothers, who struggled to generate sufficient lift and solve the three-axis control problem of a fixed wing aircraft. We’re also deeply immersed in complex problems, but there’s an important difference: we’re building our ‘airplane’ while we’re flying it!
Onward!
The Synesis One team
[1] Gary Marcus, “AI’s Jurassic Park moment,” blog entry, December 10, 2022.
[2] Kevin Roose, “Why a Conversation with Bing’s chatbot left me deeply unsettled” NYT, Feb 17, 2023.
[3] Tim Harford, “Why chatbots are bound to spout bullshit” Financial Times, Feb 10, 2023.
[4] Emily Olseon, “Google shares drop $100 billion after its new AI chatbot makes a mistake”, NPR, Feb 9, 2023.
[5] Marcus, op. cit.
