daralthus

omp+0x-alpha beat both cc+fable and codex-sol in creating/refactoring a big eval setup. the former just knows where things should belong and completed the task all the way while the other two failed on both metrics.

quite the opposite. it feels refreshing that ppl with talent and money can do inspiring things with it.

ppl pay for this?

how are you handling schema updates? do you migrate jobs or handle worker deployments in a specific way?

that's because it's an acquisition (http://usegalileo.ai/)

This is broadly accurate, but if anyone feels like freaking out and quickly needs an antidote to the "high" class of sociopath grifters, perhaps could find some solitude in Wim Wenders' Perfect Days for a few hours.

While I mostly agree with the grim conclusion, I think there's an important aspect ignored that in every domain the very first and very best brands could leverage as good design and purpose: art and fun.

Remember winamp skins?

"The purpose of art is to impart the sensation of things as they are perceived and not as they are known. The technique of art is to make objects 'unfamiliar', to make forms difficult, to increase the difficulty and length of perception because the process of perception is an aesthetic end in itself and must be prolonged." - Shklovsky on defamiliarization

What pg is really writing about is enshittification, consumerism.

ppl are so sensitive. it's not impolite to be direct. why would you be wasting each other's time by "dear, sincerely etc" every single time.

Honestly this is the best way to learn about inverse problems and it's what made convolution click for me a few years ago.

:(){ :|:& };:

sub-agents that access (and manipulate) the SAME context (a file system or variables in the REPL)

File system access is a must tho, that's where half the power of coding agents come from: efficiently managing context files.

I 100% agree in the importance of the data model, and the examples show that it often makes more sense to start from the user's perspective of it rather than your db schema.

Interestingly AI Agents are all about disrupting the hard bounds of existing data and interaction models and it turns out the lowest common denominator is often the winner. Eg.: file system > database, grep > embeddings, markdown > pdf, generative ui > portals, computer use > api-s etc.

There simply is no need for all that abstraction / interface / infrastructure to eg. answer questions about documents or to keep track of todo lists, workflows or sending messages etc. when you have glue that can translate between the data models.

this would be useful outside of coding. could you release a benchmark so we can have more models tuned for this?

I would

I thought this was pretty well known (at least in the JAX/XLA world). I've hit this many times and got batch variance explained to me before: https://github.com/google-deepmind/penzai/issues/82 and https://github.com/jax-ml/jax/issues/20047#issuecomment-1975...

If that helps, I get burnt out on just red bull already...

A simple tool to continuously vectorize your PostgreSQL tables with the flexibility of your own embedding pipelines.

https://github.com/BarnacleLabs/RAGmatic

> No real human would go four links deep into a maze of AI-generated nonsense.

sure... sure nobody.

yes that's the durable objects ~ durable agents model that cloudflare is building

To be fair, it's more likely they just looked you up on FB.

inference speed of the models is probably the bottleneck

just pet bottle recycling by itself is a multi-billion dollar industry globally [1]

[1](https://en.wikipedia.org/wiki/PET_bottle_recycling)

or you could just compile pulumi js to webassembly?

What if u embed `askLLM("5 thins this Diff could fix" + chunk)` instead of `chunk`? That should be closer in the latent space.

I have seen numbers come differently in JAX just depending on the batch size, simply because the compiler optimizes to a different sequence of operations on the hardware.

So this is like aws step functions and cloudflare workflows and temporal? What are the differences, how is deployment handled?

I don't think the processing happens in postgres at all, it's just handling the transactions and queues.

cool now do delay lines!

Is this just Azure AI Studio?

My experience is that Azure OpenAI is waay behind on API and model parity compared to the same thing from OpenAI.