You understand that feeling you get while you learn one thing that was clearly written by AI?
It is taking place increasingly and it pains me to see actual human communication usurped, even when it means fewer typos. It is clearly the place we’re headed and untold billions will proceed to be made but it surely’s haphazard, inauthentic and sometimes confidently flawed.
if this picture is any indication, those self same issues are coming to the actual world.
An OpenAI worker this week posted two pictures of the ‘OpenAI Library’ presumably on the San Francisco HQ: the one above and this one.
If you happen to’ve achieved any picture prompting with AI, it is clear what that is. It was undoubtedly imagined by ChatGPT after which some skilled craftsman have been introduced in to show the pc’s imaginative and prescient into actual life.
In idea, I like this type of factor. It is a killer app for enhancing your personal residence and a devastating flip of occasions for inside designers.
However seeing it in actual life is bothersome. The aesthetic is a multitude of some sort of machine’s sense of concord, not a human’s regular use of area. It is a immediate of ‘balanced timelessness’ with the sort of comfortable lighting is simply too cinematic and typical of AI pictures.
However what frightens me is that it would not work. It is as if Sam Altman produced the picture, gave it to the employees and mentioned ‘do it precisely like this’ with out anybody pushing again. For one, the lighting would not work for a library. There may be modest pure mild from the window however at evening it will not be vibrant sufficient to learn something in most of that room, notably these comfy-looking black sofas. Secondly, the books appear to be props. It would not appear to be an actual library the place the cabinets are full and other people truly learn the books and as an alternative appears like they have been chosen for his or her sizes and colours of their bindings.
And critically, discover the vegetation on the highest cabinets. Now that aesthetic might look good in a photograph however in actual life, vegetation want mild. There are not any vegetation on earth that may survive in these spots. Furthermore, the one that posted the picture mentioned they have been “largely cacti but it surely’s true they did look a lil wilted final evening”. So whereas a cactus would possibly want much less climbing up that ladder to water, it might want hours of direct daylight day by day. Even the snake plant on the ground (a notoriously indestructible home plant) most likely will not make it.
Why did nobody have a look at these cacti and say ‘Wait, will not these die?”
This can be a preview of the world to come back. It appears fairly however the info in it’s hole. The picture is gorgeous, but it surely will not maintain the life it was designed for. Briefly, it is a massive gross sales job, not one thing that fixes a real-world drawback.
The machine will inform us what’s greatest and we’ve got to dwell with the implications. It amazes and scares me that the individuals at OpenAI are such slaves to those machines that they constructed it with out seeing the apparent flaws.
The one who posted this picture works in ‘human information’ at OpenAI. These are the individuals who must be most conscious of its pitfalls, like its tendency to be “confidently flawed” or to optimize for look over perform, but they replicate these actual pitfalls in the actual world. What does it say about their skill to manage or information it?
This is not what alignment appears like.
