Agents, right time is **now**

Agents are easy to imagine, difficult to make it work…but the right time is **now**. Andrej Karpathy

https://twitter.com/zhouql1978/status/1677355485639634944

https://towardsdatascience.com/document-oriented-agents-a-journey-with-vector-databases-llms-langchain-fastapi-and-docker-be0efcd229f4

https://twitter.com/CobusGreylingZA/status/1674499332635086848

Videos from agents hackathon..

Andrej talk starts at 1:32

But current agents suck, that’s why 1st place was won by AgentEval (agent debugger)

https://osu-nlp-group.github.io/Mind2Web/

Mind2Web is a dataset for developing and evaluating generalist agents for the web that can follow language instructions to complete complex tasks on any website. Mind2Web contains 2,350 tasks from 137 websites spanning 31 domains that:

Reflect diverse and practical use cases on the web.

Provide challenging yet realistic environments with real-world websites.

Test generalization ability across tasks and environments.

What are the challenges with agents?

Almost everything that humans do, should agents do also. Think (reason), memorize (long/short term), recognize, use tools, write code, use APIs/services, etc.

Frameworks and tools for agents/bots

https://github.com/Xpitfire/symbolicai

https://docs.langchain.com/docs/components/agents/

https://www.superagent.sh/

https://superagi.com/

https://fine-tuner.ai/

LLMs for agents

https://huggingface.co/blog/inference-endpoints-llm

Agents, right time is now

It’s a 1st time in a human history, thanks to latest advancements in AI, ML, LLMs, etc. programs can “think” (yes, not actually, but simulate thinking or reasoning processes) and “act” (use tools).

What can agents do for business?

Almost everything you can imagine.

While the agents are on their way, some projects showcase