izzygonzalez

ur welcome

This is absurd.

In the spirit of CCRU, me and a few dozen other people have been having ongoing discussions on related topics under the banner of effective extropianism. I think it’s important to figure out how the landscape of rapidly evolving tech fits into our lives and vice versa. We’re working on a repository of adjacent texts.

If you’re interested, my Twitter handle is in my hn bio.

“This comparison was done by running multiple hypothesis tests, such as the Kolmogorov-Smirnov test. These tests, combined with our visual analysis of the data yielded the result that repositories containing swearwords exhibit a statistically significant higher average code-quality (5.87) compared to our general population (5.41).”

Abstract:

Theory of mind (ToM), or the ability to impute unobservable mental states to others, is central to human social interactions, communication, empathy, self-consciousness, and morality. We administer classic false-belief tasks, widely used to test ToM in humans, to several language models, without any examples or pre-training.

Our results show that models published before 2022 show virtually no ability to solve ToM tasks. Yet, the January 2022 version of GPT-3 (davinci-002) solved 70% of ToM tasks, a performance comparable with that of seven-year-old children. Moreover, its November 2022 version (davinci-003), solved 93% of ToM tasks, a performance comparable with that of nine-year-old children.

These findings suggest that ToM-like ability (thus far considered to be uniquely human) may have spontaneously emerged as a byproduct of language models' improving language skills.

It’s an error. You can get around the limit by scheduling a tweet a minute into the future or using the API to tweet/reply.

This may be true today, but the rate of progress on these tools is hitting a curve. Text and image generation are already basically indiscernible from human creations, and the entire space is now turning its attention to audio in tandem.

I think you’re right that code models are a vital research path toward AGI.

The steps you elucidated are all expressible in natural language, and we see models like Codex Edit making headway there. One of the most fascinating parts of this is that once access to the known baselines are provided to high-level engineers, they then go on to do much more than what the models alone can do.

The main hinderance to enterprise was compliance but the move toward Azure, etc, will dissolve those barriers this year.

I don’t know what your argument is exactly, but it seems to be something like, “Software engineering is really hard so computers can’t do it.”

A variation of that argument props up most common AI skepticism. I don’t think there’s anything out right now that would convince you, but from what I know, everything you pointed out will be solved within the next few years.