NEAR AI Office Hours w TICLE and SphereOne

Опубликовано: 01 Январь 1970
на канале: NEAR Protocol

457

Illia & Alex discuss the latest AI research topics and unpack some NEAR.AI focus areas.

NEAR AI: https://near.ai/
Illia Polosukhin https://x.com/ilblackdragon
Alex Skidanov https://x.com/AlexSkidanov

Guests:
TICLE
https://x.com/TicleOfficial
https://www.ticle.io
Trustless low-code developer tool that autonomously plans, build, and deploys cloud-native applications by fetching & integrating crowdsourced APIs.

SphereOne
https://x.com/sphereone_
https://sphereone.xyz/
Building an AI orchestration and agent building layer to create agents focused on payments, pool re-balancing, AI solvers, automation and wrapper for different protocols, can onramp, bridge, swap, transfer.

Reactions to LLAMA 3, 400B - https://ai.meta.com/blog/meta-llama-3/
One of the north stars is teaching models how to think well, exposing them to competitive programming
Works well with Fireworks https://fireworks.ai/ and TICLE
Function calling code? not finetuned for, but no problem

What is more important? the models? or what you build with it?
An interesting take: using larger, performant models to train smaller models
The secret to how GPT-2 continues to improve in quality

Deepminds’ AlphaProof on International Mathematics Olympiad (IMO), analyzing their silver medal
AI achieves silver-medal standard solving International Mathematical Olympiad problems - Google DeepMind https://deepmind.google/discover/blog...

How they did it, Lean + AlphaZero reinforcement learning https://lean-lang.org/about/
They took 3 days, humans needed 4 hours

A detour into Lean and model training
About — Lean
The need for synthetic data in using Lean for model training
In this case, we live in one of two universes
From a business perspective, how viable is this? Lean in real-world applications
Generalizing the achievement of DeepMind, what it means for AI development

The complexity of adding new code to existing codebases

The challenge of developing an actual AI use cases in/with Web3
The pressure is on for Web3 to prove it can play a major role in the AI game
How TICLE is taking on this challenge: working alongside, not against big AI models
How TICLE tackles the challenge of APIs breaking down when breaking changes: leveraging AI engineer that fetches kubernets clusters + human devs

TICLE Deep Dive
Ticle
Low-code developer tool that uses AI to automatically generate backends for cloud-native applications
Use community to fetch and crowdsource APIs
Designing Backend-As-A-Service system within the past, transitioning into microservice architecture
Building a collaborative as opposed to a penalty-driven developer community
Determining incentives for API developers: hit rate (recommendations) and usage

SphereOne Deep Dive
Building a one-stop shop orchestration layer for Web3: A SkyNet for Web3 https://sphereone.notion.site/Surface...
Multi-agent system that creates agents
Handling agent lifetimes and incentives/payments for agents
Creating 800 agents in 7 days, done!
Dealing with nefarious actors

What is the difference between an agent and an API
Agents: increased tolerance and fault redundancy, flexible
APIs: highly specialized and optimized for specific tasks, interoperable
The difference is semantic (Agents) vs syntactic (APIs)
Merging agents and APIs https://websim.ai/

Winding down
Exciting possibilities: LLM OS (Andrej Karpathy) https://x.com/karpathy/status/1723140...
AI in 2017
How APIs should be ideally, going forward
How do we make agents more like APIs
Multi-agent systems https://ieeexplore.ieee.org/stamp/sta...

Join NEAR's community:
Website: https://near.org/
NEAR AI: https://near.ai/
Twitter: / nearprotocol
Blog: https://near.org/blog/
Medium: / nearprotocol
GitHub: https://github.com/near
Dev Docs: https://docs.near.org/
#NEAR #nearprotocol #NEARAI