jamilton

I know this has been said many times before, but I wonder why this is such a common outcome. Maybe from negative outcomes being underrepresented in the training data? Maybe that plus being something slightly niche and complex?

The screenshot method not working is unsurprising to me, VLLMs visual reasoning is very bad with details because they (as far as I understand) do not really have access to those details, just the image embedding and maybe an OCR'd transcript.

I’m sure miegakure is going to come out any day now.

Tangent but now I’m curious about the bot, is there a write-up anywhere? How did it work? If someone says “hi”, what did the bot respond and what did the human do? I’m picturing ELIZA with templates with blanks a human could fill in with relevant details when necessary.

The article links to the case of Adam Raine, a depressed teenager who confided in ChatGPT for months and committed suicide. The parents blame ChatGPT. Some of the quotes definitely sound like encouraging suicide to me. It’s tough to evaluate the counterfactual though. Article with more detail: https://www.npr.org/sections/shots-health-news/2025/09/19/nx...

https://www.figure.ai/figure says 5 hour battery life. No mention of charge rate.

It's cool, but it's total uncanny valley for me, and I haven't gotten that from real robots before. Something about the movement in particular is odd.

At that point just have a messaging standard that allows in-line small images.

Neat. Reminds me of Planet Smith, a work in progress game with a similar concept (spherical Minecraft), except it uses hexes instead of cubes. Hexes reduce distortion, but add their own complexity. There’s a small number of pentagons on the surface, too, to make the tiling work.

I've seen people make finetunes for character turnarounds, there's probably one for 2D pixel art in particular, or perhaps combining such a finetune with a pixel art finetune would work.

It looks like retrodiffusion.ai in particular has something close.

A stock price decreasing essentially "vaporizes" the wealth of owners, that is not a straightforward transfer. Stocks aren't cash.

The dream, as I see it, is that AGI could 1, automate research/engineering, such that it would be self-improving and advance technology faster and better than would happen without AGI, improving quality of life, and 2, do a significant amount of the labor, especially physical labor via robotics, that people currently do. 2 would be significant enough in scale that it reduces the amount of labor people need to do on average without lowering quality of life. The political/economic details of that are typically handwaved.

The morality of it depends on the details.

Elevenlabs has a feature for a "full cast"-type generation, where different characters will get different voices. It's certainly not automatically sensitive to dialect though.

It's probably possible with current systems to do though. I believe there are TTS systems that can use context/prompting to change emphasis and other speech qualities, though I'm not sure how reliably.

Same here, but it worked when I opened it in Chrome. What a weird error - you would think that playing an embedded mp4 with audio wouldn't differ from browser to browser.

A lot of the music seems to be reuploaded from freemusicarchive.org, hadn't heard of that before but it seems cool.

And there's plenty of reason to not take it seriously from Sam Altman in particular.

The amount of people who choose to not drive to work is significantly impacted by policy.

An abandoned quarry is even better!

I'm a fan of shrimptin. https://shrimptin.tumblr.com/

The hand-holding and the ball toss are both super cute.

That's about your user content, not others'.

Surprisingly funny! AI is often unfunny when it tries to be funny, I wonder why this is better.

The PHP implementation works by converting it to SQL, so I'm sure they're aware.

"confidence" doesn't have to be an emotional state. It's essentially just another word for "probability" here - any model's confidence of X is the probability it yields for X. Isn't this common terminology?

1. It's not actually refuting the point being made which is "it's seems hard to take advantage of LLMs' capabilities", which to me seems like a good point that stands on it's own, regardless of who is saying it.

2. The original post is, to my eyes, not implying anything about training a model to help use other models, so the second part seems irrelevant.

FYI, the default file errors out!

Why? Wouldn't existing tools be about as good as they could be?

>Since nobody is being hurt by corporate usage of open source, there's no problem.

I presume this is the problem - if you dislike Microsoft, then Microsoft benefiting is potentially a problem.

Maybe you could just make a big one.

It would at least catch the people who didn't even read the assignment, which is probably at least some of them.

Note it's "not divisible by three". n = 3 is the only exception.