I'm talking about per-request model remember? With extensive prompt you realistically can have one request every 10 minutes because the agent will be busy for at least 10 minutes executing it. They aren't rate limiting that.
GitHub have finally wised up that you could run Copilot on a cron and use almost unlimited tokens. They have since made the product unusable by enforcing extremely strict rate limits, annoying many paid customers.
How? They run their scraping and training infrastructure - and models themselves - from within those “AI datacenters”[1] we hear about in the news - and not proxying through end-users’ own pipes.
[1]: in quotes, because I dislike the term, because it’s immaterial whether or not an ugly block of concrete out in the sticks is housing LLM hardware - or good ol’ fashioned colo racks.
Nice. A similar win I've found is documentating process flows with mermaid diagrams (.mmd). I would imagine it would work well when combined with your approach.
Thanks, I’ll definitely look into this! I’m planning some benchmark experiments with different prompting techniques since even small model updates (like the 1M token Claude change) seem to change how it uses the tool. If you try it out with Mermaid diagrams, I’d love to hear how it works.
I'm using "RAM" loosely, meaning working memory here. In practice, it's a key-value store with pub/sub stored on our shared memory layer, Ensue. Agents write structured state to keys like proofs/{id}/goals/{goal_id}, others subscribe via SSE. Also has embedding-based semantic search, so agents can find tactics from similar past goals.
This doesn't make any sense. To overwhelm proof of stake you need to have a majority of the crypto. To overwhelm proof of work, you need to have a majority of mining power. AI agents have neither of these.
AI native Blockchains inherently have no value because they're not attached to real world money.
I think people are maybe missing a key development with stable coins recently. They can be used by AI agents to pay for access to protected API endpoints or websites. The http 402 status endpoint is finally being utilised, years after it's creation. Ycombinator has a lot of AI based startups and they are given unlimited tokens.
All crypto/browser automation/bot detection companies are jumping on the bandwagon:
In a world without search engines, LLM chat bots will need to be held to account for the server resources they're using. Seems like a lot of companies are betting on them paying for access or acting as AI shopping agents.
Good, KYC exists for a reason. Why does AI need to open an account, anyway? Just give it a debit card with a limit, not a whole new account and contract with a bank.
Those are much easier problems to solve, and surely already solved by some fintechs, than bringing cryptocurrencies to the minimum legal compliance and meeting performance requirements.
I fail to see the use case where it's useful. I understand how it works and what it might enable but typically I want to cherry pick my API because I trust the source and their pricing (as here we are covering only paid for services).
To start, it's great for micropayments globally. There are examples where you want an API once and not again, and you don't want to create an account or link a credit card.
Cloudflare was one of our earliest partners, and they saw a critical need for it for web scraping by AI.
My personal Website supported WebMonetization (details https://webmonetization.org ) for more than 5 years already so no need to convince me about that, I agree. I also believe one could just as easily have a funding.md with an IBAN and structured communication to make the equivalent.
Anyway that's beside the point, what I still don't get is a use case without or without AI according to the constraints I listed before.
Training data for LLMs immediately springs to mind. They've had a free pass so far but there have been numerous threads on HN talking about server costs ramping up. People are creating zip bombs etc. to combat the LLM companies. Artists are not happy about content being ripped off.
If you consider that AI agents may end up autonomously designing, building and running SaaS-like products, or API microservices, it makes sense that they should be able to pay systems in stable coin. It allows them to operate without the restrictions put in place by traditional financial institutions. That's my futurist opinion.
Automatic payments for AI seems like a completely dangerous wormhole for hacks. In the line of smart contracts security but with the indetermination of AI.
The unlimited tokens thing is a sign that YC are expecting all their startups to integrate AI as a core part of their product. It seems like a natural progression for AI to start purchasing things autonomously. My bet is that YC also think this and are building tech that can do this with stable coin - "AI shopping" regimes.
reply