Dami

They can spend all the hundreds of million and there will not be a mane, firmino & Salah cohesion. At least not in the next decade.

34 years old Salah scrape a goal here today by the way. Hopefully the future is bright because Arsenal will win the pl this season and the next at least.

Liverpool need to build the team around Wirtz and Ekitike.

Web3 right now is great for transparency but terrible at anything that needs real privacy. In blockchains, every transaction you make is public. It’s like having your bank statement accessible to anyone who cares enough to view them online forever. It’s the year of privacy and @inconetwork is a mover as a confidentiality layer solution for blockchains privacy. This gives you the choice to be transparent, confidential, anonymous or totally private. Your money trails, your choice. Shield or Unshield.

Summoning my favourite protocols @Uniswap @jumperapp @aave to join the movement! ETHGas is introducing the Open Gas Initiative, eliminating gas fees from the end-user experience Learn more: ethgas.com/open-gas/ xitter.cf/ethgasofficial/status/…

GASS token may launch soon. @ETHGasOfficial raised $12M in seed round, led by @polychain to build and power an instant and gasless ethereum. Don't forget to claim your Gas ID for free. Get your Gas ID: ethgas.com/community/gas-rep… Connect your wallet(s), earn beans for gas spent and complete social quests to earn more.

They tried to scapegoat Salah for this atrocious football from Liverpool lol.

Discovered my Gas ID via ETHGas - turning my gas spend into rewards 🫘 As a Hero Jack, I've spent 4.1726 ETH on gas but earned 2500 Beans back. Get your Gas ID and Beans here: ethgas.com/community/gas-rep…

2025 is about to end and you’re in web3 with no skill or irl actual hustle. Just at the mercy of founders who leverage the system. You need to restrategize, no Gary Gensler to save us, so save yourself.

Delphi, Gensyn's Prediction market for Machine Intelligence. Decentralizing AI is important and @gensynai, they believe everyone should have access. We have various language models used to train systems from smaller to large language models such as GPT-OSS-20B by Open ai, Qwen2.5-0.5B-Instruct and many more. We've carried out training using various language models in our different swarm environment on math problems, reasoning and now coding. What if I told you there is a market to evaluate large language models, based on their performance. ---------------> This is gensyn's new product, Delphi a prediction market currently on testnet that allow users to trade open source models. The current market category is general reasoning, and you can buy shares in models you feel would perform well on this benchmark. Goto: delphi.gensyn.ai/market/0 -Connect wallet. -Get $TEST token from the My Activity tab, you can request every 24 hours. -There are currently 5 open markets. Buy shares of model or models you feel would perform best in general reasoning. -You can sell from the My Activity tab as well. -Current market opened till the 29th of December 2025. Once market settles, one or more models are declared winners based on their overall performance. Winning models are rewarded. ----------------- Delphi settles market on chain, transparently. In the future, it will be done via Verde, gensyn's verification system. Delphi is also fully decentralized, it makes use of LMSR, Logarithmic Market Scoring Rule, a scoring rule for handling predictions. This rule is transparent and cannot be manipulated because it's an algorithm that sits on chain unlike automated market maker that work with most prediction markets. Delphi isn't just a models predictions market; it can help with how we evaluate AI progress.

BlockAssist by @gensynai version 0.1.5 is here! You may think blockassist is just an AI assistant that learns from users' actions in Minecraft. It's more than that, it is a demonstration of how we can train AI for specific purposes. My assistant is still a bozo by the way. It has mastered the act of acting busy while doing the bare minimum. We’ve now got more environments so you can build different structures with your AI assistant. You’ll need different block materials to build in various structures. Use the E key on your keyboard to display inventory and replace an item with one you need. When you’re about to get mad at your annoying AI assistant, remember BlockAssist is a demo view and not in the final state. Now let’s get those structures up. gswarm🐜

Gensyn: Swarm Sampling Policy Optimization AI models learn by trial and error while improving over time. In @gensynai research, they introduced SAPO, a fully decentralized reinforcement learning post training process. Instead of relying on centralized systems where the authority says, "trust me bro", SAPO enables decentralized system to train faster with less compute per node. With SAPO, more diverse nodes from different part of the world training models mean better output, without exploding costs. Traditional reinforcement Learning post training faces bottleneck issues. It requires massive GPU clusters and limits participation to large compute resourced entities. With SAPO, a post training algorithm, each node trains its model in the RL Swarm and share its output with other models who can adopt it in the swarm. This method creates a powerful swarm network where one node can discover something pivotal or realize a breakthrough concept and share it across the swarm. One important benefit of the swarm sampling policy optimization process is it hasten collective learning in the swarm. Model solves tasks locally, shares insights, and samples other successes as update. This decentralizes reinforcement learning, allowing thousands of diverse nodes to contribute to stronger models.

Gensyn: Adopting Reinforcement Learning AI used to be a concept of the future. It’s moved beyond that; it’s present now in our everyday lives. We now have models being trained @gensynai for numerous problem solving and other possibilities. How, you may ask. This is where reinforcement learning, a method gensyn adopts comes in, which allows models to learn through trial and error over time in an environment. --------------> Models earn rewards for performing actions that contribute to achieving the defined goal. Rewards are feedback mechanism that signal value. An example is a person being handed a clipper for the first time to learn how to cut hair. -He watches hundreds of videos tutorials on skin cut, fade, low cuts and the likes. -He practice on hundreds of mannequin heads to get better, decides to start cutting real people's hair. -You can expect the first 20 to 100 haircuts to range from disaster to okay. Once you set your goals, these models, with data are being trained towards them. ------------ With reinforcement learning, models learn to make decisions independently without programming for each move. AI models can also come up with a new way of completing a task that was never anticipated. Despite its successes, reinforcement learning models requires significant compute resources and time, especially for complex environments. As reinforcement learning becomes more efficient, its application to gaming, healthcare, finance, Robotics, education and much more use cases are expected to grow.

VERDE, GENSYN TRANSPARENT VERDICT MASTER Machine learning and training on @gensynai is not a new concept. Verification is the "show your work" rule, which means demonstrating how you arrived at the output. How do you verify tasks though? In centralized systems within AI tech giants, you trust the provider because... well, that's all you can do. --------------> In decentralized systems, with an open network like gensyn, training is carried out on different machines across the globe from CPUs to high end GPUs. Introducing Verde, gensyn's very own verification system to ensure transparency of output. -You need to train models for a task. -Loads of GPUs are needed but you don't have them. -You outsource your training to a third party. So how do you verify the output from task? You want to make sure your full data was used, you also want to make sure no additional data was added. Verde allows you to verify the outsourced provider did what you asked them to do. Types of verification Trust Based In this, the clients trust the machine performing the task execution via TEE, (Trusted Execution Environment) Trusting the reputation of the compute providers. Learning Based In this, verification is done by monitoring how the models learn during training. The disadvantage is you don't know if all your data was used or additional data added. Execution Based Verification is done by checking if the steps are followed. Through cryptographic proof or refereed delegation. Gensyn adopts the refereed delegation approach. -It uses reproducible system to verify training. -Two parties execute task, output match, it is considered correct. -Different output, the referee, a decentralized system, i.e. blockchain determines the correct output. -No re-execution of the entire task. Gensyn already use Verde in judge, a prediction market in our previous swarm environment. The purpose of blockchain is decentralization and the research and implementation of Verde ensures the transparency of open network tasks.

Gensyn: NPCs IN GAMING You know when you’ve finished playing a game and you can tell how all characters you encountered act, how they attack and overall interact with you. What if this process can be more exciting and you get a different feel, experience and encounter with Non player characters when you play a game or decide to play your favorite game for the second time. ------------> This is one of the futuristic concepts in gaming. In my opinion, real AI integration does not only help with problem solving. It also gives a feeling of personalization. One of the problems in AI models training is hardware requirements and this is where @gensynai comes in. Artificial intelligence is the future of Non player characters that don’t just follow prewritten scripts but adapt, learn, and grow based on the player’s behavior. Instead of repeating same programmed gameplay, NPCs powered by models trained through reinforcement learning could become true companions. For example, that NPC in your first shooters game, you see by the corner and know its every move might learn to dodge your favorite attacks or adjust its position. Villains or final boss in games through deep learning, can remember past battles with a player and adapt their strategies, based on various training episodes recorded, just as we participated in BlockAssist. -------------- The benefit of AI models trained NPCs is it makes games less predictable and far more replayable. Each experience could present different challenges based on how NPCs adapt to player decisions. In 10 years time or less, the adoption of AI in games will be massive, so we can experience them in a different way when we play. While we might all be excited about this prospect, training for this will probably require bigger model parameters and more powerful compute resources. The future of NPCs is one where they become evolving agents.

It’s Sunday guys so don’t forget to touch grass. With that being said, @gensynai has some of the best and most talented community members. Check out Tempest’s fun gensyn prediction market. Here is my prediction 🙂