BabaIsPissed

This is fucked, you don't use a black box approach in anything high risk without human supervision. Whisper probably could be used to help accelerate a transcriptions done by an expert, maybe some sort of "first pass" that needs to be validated, but even then it might not help speed things up and might impact quality (see coding with copilot). Maybe also use the timestamp information for some filtering of the most egregious hallucinations, or a bespoke fine-tuning setup (assuming it was fine-tuned it the first place)? Just spitballing here, I should probably read the paper to see what the common error cases are.

It's funny, because this is the openAI model I had the least cynicism towards, did they bazinga it up when I wasn't looking?

UFO50 and Diceomancer, both are fantastic! Also some lethal company, plateup and remnant 2 with my friends. Want to try out the new factorio dlc but not sure if I want to sink the time it demands.

Your "new alien intelligence" couldn't even count how many Rs are in strawberry, shut the fuck up.

The funny thing is that he's correct when he says that we are not sufficiently organized to deal with climate change. He probably wouldn't like the solution though.

Honestly, this is expected of tech bros, just look at crypto. Shame on every computer scientist that gave legitimacy to these dipshits for a paycheck, especially the big names of the deep learning old guard huffing that heavy copium.

We consistently find across all our experiments that, across concepts, the frequency of a concept in the pretraining dataset is a strong predictor of the model’s performance on test examples containing that concept. Notably, model performance scales linearly as the concept frequency in pretraining data grows exponentially

This reminds me of an older paper on how LLMs can't even do basic math when examples fall outside the training distribution (note that this was GPT-J and as far as I'm aware no such analysis is possible with GPT4, I wonder why), so this phenomena is not exclusive to multimodal stuff. It's one thing to pre-train a large capacity model on a general task that might benefit downstream tasks, but wanting these models to be general purpose is really, really silly.

I'm of the opinion that we're approaching a crisis in AI, we've hit a barrier on what current approaches are capable of achieving and no amount of data, labelers and tinkering with architectural minutiae or (god forbid) "prompt engineering" can fix that. My hopes are that with the bubble bursting the field will have to reckon with the need for algorithmic and architectural innovation, more robust standards for what constitutes a proper benchmark and reproducibility at the very least, and maybe, just maybe, extend its collective knowledge from other fields of study past 1960's neuroscience and explore the ethical and societal implications of your work more deeply than the oftentimes tiny obligatory ethics section of a paper. That is definetly a overgeneralization, so sorry for any researchers out here <3, I'm just disillusioned with the general state of the field.

You're correct about the C suites though , all they needed to see was one of those stupid graphs that showed line going up, with model capacity on the x axis and performance on the y axis, and their greed did the rest.

There is a disconnect between what computer scientists understands as AI and what the general public understands as AI. This was previously not a problem, nerds give confusing names to stuff all the time, but it became a problem after this latest hype cycle where incurious laypeople are in charge of the messaging (or in a less charitable interpretation, benefit from fear of the singularity™). Doesn't help that scientific communication is dogshit.

got the Samsung buds pro 2 at half price recently and I kind of like them, but they were a bit underwhelming even at that price. I've never spent a lot on audio in general, so they were actually a big improvement, but there was no "wow" factor or anything. Plus having to install bloatware that asks for all permissions under the sun sucks (why the fuck would a settings menu want to know my location???).

I do think you underestimate how nice the noise cancelation can be though. I moved to a big city and my hick ass cannot deal with all the fucking noise. Plus I'm clumsy and end up getting wires caught on everything, which means wire stuff also becomes e-waste fairly quickly.

Implementing fascism as a mechanic only for it to be unsustainable gameplaywise is a good bit tbh

u did the thing we designed the game to push you towards doing don't you feel bad u monster lolololol

To be fair to the game that's only the bait and switch at the very start with Toriel, designed to make the player reload and introduce the save meta-fuckery with Flowey. From then on the only incentive to do violence is getting stuck at a puzzle or completionism (which is at the heart of the meta-narrative).

The commentary on violence by itself is naive though (even the game points it out at one point) and if you don't like the characters or roll your eyes at 4th wall stuff the whole thing falls apart pretty quick.

evaluating LLM

ask the researcher if they are testing form or meaning

they don't understand

pull out illustrated diagram explaining what is form and what is meaning

they laugh and say "the model is demonstrating creativity sir"

looks at the test

it's form

I think people remember it fondly because it started a trend of more narrative focused western animation on TV. Like a Nickelodeon show with actual character development and worldbuilding was pretty unusual. And it seems to have that effect where some episodes/scenes just buried themselves in the grooves of people's brains like early spongebob that's probably partly a testament of quality and partly a product of endless reruns (at least that's mine and some of my friends experience with the show).