Відео

I'll add to my ZX81 wobble. The ZX81 was my brothers. But years later we sold it at a car boot sale my mum and sister went to. A guy pay £50 for it, late 90s. I thought this was funny as I said "It didn't even work properly because of the RAM pack wiping the data". Little did I know he knew what he was buying, I didn't know that the wobble was a known issue until the film Micro Men.

Quantic computers?

Could we not just use the LLMs to stick to as accurate as possible contextual access to the collection of actual human-generated semantic information, and leave the dreams of making machines into gods we are supposed to bow down to?

Why do you wear your watch on the right hand?

I think it will go up in steps, that's the natural progress of technology. Innovation, growth, plateau, repeat.

You can have infinite memory by using address maps to same location swabs. The reason why AI are so intelligent.

'My Lord is that legal?' 'I will make it legal'

No it hasnt how silly

So a Rube Goldberg machine isn't going to save humanity? What a surprise.

Despite looking like a 90s hacksploitation movie character, Dr. Bagley is the real deal <3.

So, a prof plants a batman bit in the assignment.. Wouldn't it be a directive for normal humans as well? So, if you write an essay and miss the batman bit, shouldn't you expect a lower grade, as you didn't follow the directions?

maybe, just maybe, it needs time & rl experience (human feedback) like a human

It's pronounced "raii".

Very informative.

Then gpt4o came out😂😂

3:20 I have written recommendation engines, and I think UA-cam is exponentially more difficult than a Netflix recommender, just due to the amount of choices. I have videos on UA-cam that go into Purgatory, and after 9 to 23 views, just stop getting recommended to anyone.

I guess the takeaway is that simulations can only make an AI so smart. Eventually it has to learn from directly interacting with the real world, and that would certainly slow down its seemingly explosive progress.

I'm sorry I can't get over the clean straight lines at 5:35

I mean other people have pointed it out if you use a eye on training garage results you're going to just end up getting more garbage results.

It bothers me very much that the thumbnail uses a lower case 'i' in "Ai"

Amoeba forcing generates random real numbers. ;)

The wobbly camera nearly made me sick.

So you’re saying my job will be replaced in 2 years instead of this summer?

Anyone else notice how terrible google search is? Google is the worst search engine for trying to find stuff now.

People watching in 2024 team up in the replies so my notifications are flooded

You can not get to the moon by climbing taller trees. All the data we go fed into any LLM will not get to AGI, do NOT be fooled.

This video lacks critical meaning. Perhaps the paper addresses it better, but what is the vertical axis??? If it is “percent answers that are correct” then the ONLY possible output curve is the “pessimistic” one. OTOH, if the vertical axis is “average number of correct answers before an incorrect answer” there is room for exponential growth. A pessimistic curve tor the former is meaningless, while a pessimistic curve for the latter perhaps corresponds better to the message of this video. Regardless, the evaluation metric absolutely does matter.

More data = garbage in, garbage out.

Even open ai is facing the same problem.

Someone should draw Tesla's attention to this - or at least Elon's - automated Full Self Driving will always be confounded by those unusual edge cases for which there isn't much data, because that's what makes them unusual

well yeah you'd need more advancement in the software engines themselves, not just more data. that's kind of what i'd expected but i guess we see how it goes

Post Malone got his face tattoo - Ctrl+A, Delete

There are few truly exponential progressions in the real universe, and certainly not in the kinds of complex systems we actually interact with. They're all Logistic Curves, which behave exponentially up to a point, and then start hitting progressively more substantial barriers to growth which exponentially *slow* their further growth beyond an inflection point. The end of Moores Law is an excellent example of this - but it applies to nearly everything.

You're making an argument that transformer architecture has peaked as currently trained. Change model architecture or training scheme and you may get further improvement on the same data.

In theory anything less than exciting is already pessimistic. If this generalized intelligence implicates a performance that outclasses the dataset then as long as performance and data continue in parallel it would be like trying to reach zero by only dividing. There's a seemingly unfalsifiable hypothesis I've been chewing on that our consciousness is transcendent of any dimension that we could host artificial intelligence. The closest allegory I could come up with is it's like the game Minecraft, how we can't use redstone computers in Minecraft to play Minecraft within Minecraft. We could play a less efficient form of Doom, sure, but no computer could possibly run an iteration of Minecraft possible to play an infinitely recursive series of Minecrafts within.

lora plugins

I hope generative "AI" gets sued for all the date they stole with the mask of being nonprofit.. But I also hope it stops getting better. Dont want more dirty money getting pumped into Silicon Valley at the expense of 95% of the working classes data.

What if ai takes the same trajectory as flying cars and jetpacks from the 1960s. Everyone thought we’d be flying around in our cars by 2000 but it’s now 2024 and I’m still on the ground lol.

Quantum computing is proudly passing the hype button on to AI. Nice. My expectations are still cool.

Causal ML FTW

AI is not new. It’s simply machine learning and it’s been used in computer programming for many decades. There can be no peak when something is a complete lie. The term AI is used now by big Tech To increase profits and generate interest by investors. Total scam

There's a lot of images I've tried getting so to replicate and it can't get anywhere close.😂

Another factor that I think plays a huge part in this decline of AI art generators specifically, is the decline of recular artists... As AI replaces and kills off artistic jobs, you reduce the data set even further, as generative AI images require other 'look alike' images to generate something... At some point, the data set available will be so small that AI art programs won't be able to make anything unique anymore... It will just be the same handful of images generated with slightly different features... At that point, the AI developers will have to make a choice; to leave it alone and watch their program get progressively more boring and stagnant, or use other AI generated images as part of the data set... But, AI art using AI art is similar to inbreeding... If it takes another AI generated image as 'acceptable data' it will take and add its imperfections with it, which, overtime, will pollute and corrupt the data set, and produce increasingly inferior images...

And here we are with gpt 4

But it kinda can know the origin if it was coded properly no? The main problem with these twitter bots is that they aren't coded properly. The openAI api has clear roles for this. System, user, and assistant (which are its own earlier responses to be fed back into a new request). System is a much stronger directive than user; and its quite difficult for a user to consistently jailbreak out of a system directive. Its possible, but eventually it will always go back.

I’m something of a computer scientist myself. - Norman Osborn 🐸

Great video

The problem with relying too much in AI in the medical field that I have been worried about is that it'd mean overall less practice for doctors at identifying the problem. If most doctors are lacking in the experience that they should have gain with years of doing the job, then they'll lack experience to identify the more obscure diseases and underlying problems. That isn't something AI can cover, especially when these kinds of things can change without us knowing or deprive us of new discoveries.

Reading this comment section has made very clear people are REALLY mad that what they perceived as "human" and "special" can in fact be boiled down to a series of algorithmic processes that can be repeated to emulate parts of the human condition. Often BETTER than what humans can. It's like living in an Orson Welles book and realizing the dystopias start with the individual.

Biggest issue I see is that the computer is based on numbers and strings, even when given images based on this video. I think humans base things on their interpretation of their environment, mainly 3d objects that move and exist in time and space. A cat isn't just a face, it's the whole animal, we just place more weight on the head and faces of animals. Without a base level change of classification for ai it will always be lacking.