Phoenix Beaudry

just a little kid trying to understand a big world.

[Gradients @gradients_ai]

Generic LLMs answer medical questions confidently. They're also often wrong.

We fine-tuned one on PubMed literature with gradients sdk. Now it agrees with the peer-reviewed answer.

3 lines of Python:

pip install gradientsio

[Phoenix Beaudry @PhoenixBeaudry]

@Metathea11 @KojimaErgoSum @tszzl Everything ought to be explained as simple as possible - but no simpler.

[const @const_reborn]

https://t.co/U7wIMIE0MC

minimalistic style meets dense knowledge

[Phoenix Beaudry @PhoenixBeaudry]

@goafterwork @Techweek_ first time in SF hope there is more spots available for the {after} party :)

[He-Man, Principal MLE @RandolphCarterZ]

Alignment should probably begin and end with "don't kill everyone", let it learn the rest

Most would not appreciate an immortal super-entity aligned to 1720s human values

Would 2200 appreciate one aligned to 2020s human values?

[Phoenix Beaudry @PhoenixBeaudry]

@RandolphCarterZ @made_by_chris However the kind of super-optimizer I am thinking of is ignorant of it's own reward function so maybe capping these systems is sufficient.

[Phoenix Beaudry @PhoenixBeaudry]

@RandolphCarterZ @made_by_chris You don't think it's possible that super-intelligence and super-optimizers are different evolutionary paths and we might accidentally find ourselves on the wrong one?

[Phoenix Beaudry @PhoenixBeaudry]

@RandolphCarterZ Is there a reason that is a ridiculous thing to think might happen?

[水江未来 @MIRAI_MIZUE]

Do you know what is going on inside the pupa?

[Siqi Chen @blader]

new research from OpenAI used gpt4 to label all 307,200 neurons in gpt2, labeling each with plain english descriptions of the role each neuron plays in the model.

this opens up a new direction in explainability and alignment in AI, helping make models more explainable and potentially easier to align.

https://t.co/k8FkDAygDT

[AI Notkilleveryoneism Memes ⏸️ @AISafetyMemes]

@OpenAI the debate rates

[Matthew Barnett @MatthewJBar]

(I still think AI risk is a huge deal, by the way. I suspect some people think I'm a total AI risk skeptic, but I'm not. I just don't give human extinction from AI >50% chance.)

[Matthew Barnett @MatthewJBar]

In this post @robbensinger neatly lays out the argument for a very high likelihood of AI doom. I probably disagree with premise 2b the most. It depends on what counts as "full value loading" but I've not been persuaded that it's "~impossibly difficult".
https://t.co/eERwb6KwpE

[Phoenix Beaudry @PhoenixBeaudry]

This Tweet was automatically sent just by asking Calo to send a tweet :) stay tuned for the beta.

[Sam Altman @sama]

here is an alternative path for society: ignore the culture war. ignore the attention war. make safe agi. make fusion. make people smarter and healthier. make 20 other things of that magnitude.

start radical growth, inclusivity, and optimism.

expand throughout the universe.

[Sam Altman @sama]

the creative power of the world is breathtaking to watch.

each time our tools get a little better and a little bit more accessible, our creations improve more than we imagine possible.

[Phoenix Beaudry @PhoenixBeaudry]

@leecronin Where do LLMs fall on the Assembly theory complexity scale? Where do systems of multiple recurrent LLMs that interact with each other?

[🌱 John Ash 🌳 @speakerjohnash]

This is Art