What a fascinating read, thanks for sharing that link.

That paper is incredible. It’s crazy that they found 41% of all DN connections receive recurrent feedback from a downstream neuron.

Some things that may be commonplace understanding in the neuroscience community but that I found interesting:

- speculation that deep recurrence in the learning center is a mechanism for working memory, and allows for multiple high-level cognitive processes to occur simultaneously

- the description of how a variety of neurons in the learning center categorize stimuli, a different group controls the learned value of inputs (valence?), and then another group integrates the valences of both the learned and innate neuron groups for the given stimulus category

Oh, and that most of the neurons were engaged in multi-modal activity

Look into Nix / NixOS

What tests are you running? Anything outside of the alpaca eval test suite?

Also interested. Would be really great to run this against a self hosted LLM agent.

Does this mean your queries are all one-shot instead of utilizing techniques like LangChain?

Exactly, you can see the prompt in this file [0]. I'm not sure how LangChain arrived at their default agent prompt, but you'll almost certainly want to write your own for performance reasons if you put something into production.

[0] https://github.com/BloopAI/bloop/blob/main/server/bleep/src/...

This is great that you got gpt-4 to explore the codebase using an agent approach. I tried this previously with gpt-3.5-turbo and have been meaning to revisit it since I got gpt-4 access.

I shared some notes on HN awhile back on a variety of experiments I did with gpt-3.5-turbo.

https://news.ycombinator.com/item?id=35441666

‘Big Tech’ will always be there in some form down the road. If you have the risk tolerance and early career flexibility, take the path less traveled.

Black bar?

HN will sometimes render a black bar on the tops of pages when a prominent person has died.

I always assumed that was a rendering error

That would require HN to be modified regularly enough to induce rendering regressions :-)

ControlNet is the new development that will really allow us to guide diffusion model outputs more granularly - first I’m seeing it used against video generation

Thanks for your work in this space, it’s incredible!

Absolutely, its just unbelievable what this one person has enabled. ChatGPT isn't mind blowing as much as ggerganov is.