GraphicalSob
@graphicalsob
Followers
2K
Following
1K
Statuses
1K
Building @pyano_fun, I believe in Edge AI because 90% can't afford Cloud AI. Specialized in social graphs, probabilistic models, and NLP.
Joined February 2022
@SuhailKakar - crypto has some of the smartest people in the world probably builders. - crypto also has some of the dumbest people in the world investors
0
0
1
Bang on. 99.9% of people don't care about privacy. 99.9% of people don't care about verifiable inference. 99.9% of people don't care about decentralized inference. 99.9% of people don't care about L1/L2. 99.9% of people don't care about proof of *#_###. They just want to get their shit done in the fastest, cheapest way.
0
0
0
On one hand, we say that AI inference needs to be validated. On the other hand, we’re willing to hand over control of our computers to AI. Lol. On one hand, we argue for decentralized inference, even though it’s slow. On the other hand, we want to use the same system to control AI inference and voice. Have you ever come across someone who responds with a 10 to 120-second delay every time you speak?
0
0
0
On one hand, we say that AI inference needs to be validated. On the other hand, we’re willing to hand over control of our computers to AI. Lol. On one hand, we argue for decentralized inference, even though it’s slow. On the other hand, we want to use the same system to control AI inference and voice. Have you ever come across someone who responds with a 10 to 120-second delay every time you speak?
Voice + Computer-Use Agents. You’ve been warned.
0
0
0
@0xNairolf But first, AI apps that actually solve a problem in crypto. People care about cheap AI inference; they don’t even know what verifiability means.
0
0
2
First of all, response time is directly proportional to user experience - the lesser the better. A swarm of agents doesn't make sense until LLMs' token generation speed exceeds 1000 tokens/second or more, and communication latencies are in milliseconds. LLM token generation speed remains constant in both cases - swarm versus single agent. Therefore, the only optimization left is reducing communication overhead and improving security. Hence, a single agent should handle all tasks with tool calling. Instead of using multiple agents, people should simply share tool calls through APIs.
0
0
0
First of all, response time is directly proportional to user experience - the lesser the better. A swarm of agents doesn't make sense until LLMs' token generation speed exceeds 1000 tokens/second or more, and communication latencies are in milliseconds. LLM token generation speed remains constant in both cases - swarm versus single agent. Therefore, the only optimization left is reducing communication overhead and improving security. Hence, a single agent should handle all tasks with tool calling. Instead of using multiple agents, people should simply share tool calls through APIs.
Been chatting to a few folks about Swarms vs. Single Agents Specifically: Is it better to use a swarm of agents to complete multiple tasks OR a single agent that you engineer to do it all the tasks itself? Why? Im trying to figure out if we need swarms or if we can just use a single agent to do many things. The argument for Single Agent: You can just do everything a swarm of agents in a *single agent* by making each swarm agent’s function a separate logic for the single agent. Which means when an agent needs to tap into a different function (e.g analyse sentiment on twitter) It can just access another part of its “brain” to do so Instead of hitting up a separate agent. It can just do all the different things itself. The argument for Agent Swarms At some point, it becomes inefficient for a single agent to process multiple functions for itself. Instead, hitting up individual agents, each focused on a niche function is a much better way to scale & coordinate around an objective. I also think swarms are kind of inevitable Different teams will build different agents to serve different functions The same way we don’t just have one product/company per sector. Open source + competition means we end up with an economy of agents. The “success” of this economy depends on how well these different agents work together to boost the “GDP” of that economy. Conclusion Ultimately I think we end with swarms, just by nature of a competitive, flourishing market. If anyone has a clearer argument please share.
0
0
1
Crypto needs AI applications solving real uses for people using decentralization or using crypto as a payment layer. Crypto doesn't need infrastructure to build AI apps. When you solve a real problem with a great number of users, then only you will identify what kind of infrastructure is required. Building infrastructure beforehand is just good for CT, not for actual crypto.
0
0
2
Add code, technical documentation, and you’ll have a fully functional AI tech team.
Any knowledge worker spends 50-80%+ of their time analyzing, synthesizing, and creating unstructured data - think about how much of your time is spent reading things like reports, documentation, PRDs, presentations and creating new variants of those things. I’m personally excited about using LLMs to create knowledge agents - agents that can automate all this knowledge work. This frees up everyone’s time and encourages more critical thinking, correct decision making, less unnecessary burnout. There are various use case categories that we’re seeing already - research assistants, automated workflows, report generation. A key focus for us at @llama_index is building the best platform and tooling so anyone can build these knowledge agents. 1. LlamaCloud as the knowledge management layer: 2. Our open-source framework as the agent orchestration layer:
0
0
0
@cryptopunk7213 No-code tools work well when the product pipeline is mature enough to standardize the workflow. Even no-code website builders are good for basic flows. AI agents are still in their early stages, so you're right that no-code AI agents might cause more harm than good.
0
0
1
Focus on repetitive tasks. Allow developers to manage AI agents. Make sense of overwhelming data streams.
Y Combinator JUST announced what startups they want to fund next in 2025. And it's mostly AI that replaces $100k/year job functions. My notes below in case it's helpful to you:
0
0
1
RT @gregisenberg: Y Combinator JUST announced what startups they want to fund next in 2025. And it's mostly AI that replaces $100k/year job…
0
1K
0
I recently learned in one of the Y Combinator podcasts: 1. Find a field that you have knowledge about and may be passionate about. Passion trumps knowledge. 2. Find a job role that you would like to automate or maybe enhance with AI. 3. Sit with that person for a day or two. List down the activities that they repeatedly do. 4. Start with automating the easiest of them.
Everyone says: Find a problem. Ok, how to?
0
0
0
Global betting and gambling market worth $500 billion - $1 trillion annually (2023), including regulated + unregulated markets. Only 2-5% of sports betting is successful, casino games even worse with <1% success. is a different story - 10% of users are making money, which beats any betting industry. More users joining=platform growth incoming
0
0
0