Murray Vanwyk

[Murray Vanwyk @murrayvanwyk]

@ClaudeDevs Just reactivated my twitter login only so I could say how crap this is. We've automated a lot of work on our devs machines with claude -p and now we'll have to rework all that. $200 definitely won't last a month!

Looks like I'll be starting my codex subscription back up!

[Murray Vanwyk @murrayvanwyk]

@lydiahallie Just reactivated my twitter login only so I could say how crap this is. We've automated a lot of work on our devs machines with claude -p and now we'll have to rework all that. $200 definitely won't last a month!

Looks like I'll be starting my codex subscription back up!

[Murray Vanwyk @murrayvanwyk]

@tunguz @davorrr_ I've worked in pod offices of 4 people. It's no better. You're often wasting time going to another pod to ask a question and someone is always annoyed that you're talking in their pod.

[Daniel Dunderfelt @ddunderfelt]

@llm_sec In the npm world, I've found that it's more likely to suggest a 7 year old unmaintained package than something that doesn't exist 🤔

[LLM Security @llm_sec]

* People ask LLMs to write code
* LLMs recommend imports that don't actually exist
* Attackers work out what these imports' names are, and create & upload them with malicious payloads
* People using LLM-written code then auto-add malware themselves

https://t.co/Va9w18RpWu

[Sam Witteveen @Sam_Witteveen]

Agree! By far the biggest mistakes I see in people’s bot are bad prompting. A single sentence direction won’t usually get you good results from a human so why would a LLM understand what you want.

[Harrison Chase @hwchase17]

When connecting LLMs to your data, semantic search is a great first pass - but to get even better performance you can turn to more advanced techniques:

✒️Contextual Compression
🤳Self Query
🕐Time weighting

Going live in 30m to talk about this and more!

https://t.co/hiwnieuenA

[Murray Vanwyk @murrayvanwyk]

@langchain + @pydantic for #LLM output parsing = 😍

Never looking back!

[Murray Vanwyk @murrayvanwyk]

@codek1 Interesting! Any tips or templates for connecting gcp and aws network? We're primarly gcp, but there are a couple aws services that would make life a lot easier

[Murray Vanwyk @murrayvanwyk]

@jumbld @fpingham @JonZLuo My thinking: keeping test/prod DBs in sync is a pain. Soln: try to keep a single DB. This depends on your DB and your use case. Possible solns include transactions: when agent or others doing CRUD, horizontal scaling through read replicas for read only agents, etc

[Murray Vanwyk @murrayvanwyk]

@jumbld @fpingham @JonZLuo Not just inserts, but any case where you want to isolate effects amongst users. But I take it you're concerned with other issues

[Andrew Ng @AndrewYNg]

I'd like to have a real conversation about whether AI is a risk for human extinction. Honestly, I don't get how AI poses this risk.

What are your thoughts? And, who do you think has a thoughtful perspective on how AI poses this risk that I should talk to?

[Murray Vanwyk @murrayvanwyk]

At first, building with LLMs felt different than other AI/ML projects. Prompt tuning, no worries about training pipelines, etc.

Now I'm spending most of my time cleaning, structuring, enriching data, and controlling costs.

So, just like any other AI/ML project

Just me?

[Murray Vanwyk @murrayvanwyk]

Thanks @Sam_Witteveen for the great tutorials on youtube! Just watched another one this morning.

[Murray Vanwyk @murrayvanwyk]

@matchaman11 FAISS.. Sooo simple to use with @langchain

[Murray Vanwyk @murrayvanwyk]

Oooo enum! Nice! I can get rid of some custom workarounds now!

[Harrison Chase @hwchase17]

Big shout out to @JacobFValdez for adding two of the most requested output parsers:

✏️Enum
🕐DateTime

Enum Output Parser: https://t.co/p1F9n2MrsD

DateTime Output Parser: https://t.co/gt15C0nAtq

[LangChain @LangChain]

Most projects using agents have used one agent. Or, if theres been many agents, they typically interact in a sequential manner

Excited to share a new guest blog post from @cyrusofeden that pushes the boundary - multiple agents acting in parallel

https://t.co/kZm52604Ld

[LangChain @LangChain]

🦜🔗v0.0.145 is out with:

🍜Supabase VectorStore
💿Discord Document Loader
📁File Based Chat Message History

And a bunch of new tools for your agents to use!

📰Arxiv Tool
🦆Duck Duck Go Search Tool
🗺️Google Places Tool