Two Minute Papers

Пікірлер

@PauloSilva-dz6uh 6 сағат бұрын

@shykids could you please let me know if the voice was generate through the AI too? Cheers. PS wonderful ideia for a video.

@gamergrids 6 сағат бұрын

.alive be to time a What

@rhyscooper3693 6 сағат бұрын

I can't stand that the first sentence in Gemini is always "That's a great question!" or whatever. Just get to the point!

@user-wk4ee4bf8g 7 сағат бұрын

I'm curious about AI developments outside of LLMs that will cause the most physical changes. Which AI will make new stuff, not say new things?

@mystic6121 9 сағат бұрын

And remember, people see this and say AI's can't learn.

@sanyanders 9 сағат бұрын

Petition to return your old natural voice

@smellthel 9 сағат бұрын

In my completely uneducated opinion, smaller and more efficient models are the future. I tried just the 8 billion parameter version of LLaMA 3, and it completely blew my mind. I would be happy with an even smaller but less precise model running on my phone locally. Hopefully this is where Apple goes with Siri, that would be awesome.

@neithanm 9 сағат бұрын

"AI assistants are improving at a stunning pace", how can you conclude that? After pouring billions in R&D, data and compute, there are only marginal improvements in some areas over the GPT4 released more than a year ago. I'm bullish on AI but not delusion.

@InventorInvestor 9 сағат бұрын

How do I thumb this video up even more. AI coding is amazing.

@pace7746 9 сағат бұрын

"We also evaluated humans on MATH, and found that a computer science PhD student who does not especially like mathematics attained approximately 40% on MATH" lmao

@NicosLeben 10 сағат бұрын

I don't get why they are geoblocking their stuff. It's really annoying.

@illustriouschin 10 сағат бұрын

Whatever so called performance scores are totally negated by the scummy Meta company that rakes in billions from the worst scammers around.

@OhioNPC911 11 сағат бұрын

AI AI AI AI what a time to see losers getting excited about automation tools

@mako9673 11 сағат бұрын

I don't trust anything from Facebook, especially the 'free' things. It tells me that they are selling all the user data the collect with it

@DingoAteMeBaby 11 сағат бұрын

"open source"

@eSKAone- 11 сағат бұрын

Competition is the best 💟🌌☮️

@morthim 12 сағат бұрын

i just realized that 5 shot doesn't mean 5 shots. it means you give the AI the prompt, their answer, and a correction along with the prompt to improve it. so 5 shot doesn't mean you say "answer this" and give it 5 opportunities with different perlin noise to calibrate it. holy shit. this is so much less scary all of a sudden.

@glumpfi 12 сағат бұрын

I have tested the small 8B version of Llama3 for quite some time now locally and in my oppinion, of the smaller models, it is the best yet. It has very precise text and context understanding. Unfortunately, my laptop is too weak for the 70B model

@AndrejMejac 12 сағат бұрын

What are your machine specs ?

@glumpfi 10 сағат бұрын

@@AndrejMejac Thinkpad T470, ~2,6GHz, 16 GB ram, some random builtin graphics card

@AndrejMejac 10 сағат бұрын

@@glumpfi Thank you, i was scared by requirements i managed to find, not anymore.

@juancarlospizarromendez3954 12 сағат бұрын

It may be speedy as a reconnaissance of 300 words per second, but it is better a reconnaissance of 10 digital books per second including their illustrations.

@signupp9136 12 сағат бұрын

No trying to be a troll. But Llama 3 has been out for weeks at this point and pretty much everyone had done the “presentation video” for it. At first when I check out the video I thought maybe it was going to cover the 400B model. Strange, seems to be week+ lag now with 3MP compared to other AI/Tech channels.

@TheCynicalNihilist 12 сағат бұрын

I understand the monetary reasons why companies want their own AI. but for the benefit of society as a whole, wouldn't it be better for all companies to just merge with the top dog and form the biggest singular ai model? like lets say OpenAi achieves AGI. why would ANYONE use anything but that? What's the point in lesser models? seems like it would benefit companies and everyone else if everything just ran off the same model...one thats proven to be #1 I mean. These multimodels are cool and all but Ive already boiled it down to co-pilot because it uses Chatgpt and has great features. going forward i dont think majority of people are going to swap back and forth. which means, companies are going to be wasting money and effort when they could help make the best one better.

@Mo_Mauve 11 сағат бұрын

If they did that, then there wouldn't be a good enough open-source AI.

@TwoWayOrbitalStation 7 сағат бұрын

Not to mention they all want to be the one in the lead, the one with the breakthrough. Its all about money in the end, everyone flocks to them and they get cash. Why would they want to share it?

@juanvovoncebuitragogaleano863 13 сағат бұрын

But how do they gather information from the real world? Just the camera on their heads and (im assuming) the sensor on their chests?

@ico-theredstonesurgeon4380 13 сағат бұрын

Sumotori Dreams in real life

@HyuriConstancio 13 сағат бұрын

Llama a llama to be a llama!

@klaushermann6760 13 сағат бұрын

It might be intelligent, but it doesn't know all the references from its citations by heart due to low count in the parameters.

@SENYSENofficial 13 сағат бұрын

I spent hours trying to learn how to do backpropagation using the chatGPT-4 guidelines, I personally don't have any knowledge on how to do this, and chatGPT-4 disappointed me because he simply gave me erratic solutions and didn't explain clearly what exactly to do or how to do it, As a result, I looked for a course and now I'm able to learn, so guys, ChatGPT is a good tool, but I still don't recommend using it to learn something new

@net-twin-de 13 сағат бұрын

That is outstanding and it makes videogame development a way easier and less expensive too.

@thetransferaccount4586 13 сағат бұрын

very mid.. zero new devolopment.. AI is already plateaued

@iceshoqer 13 сағат бұрын

I actually hate Llamas, like, the only animal I have said I hate, and it’s an LLM - great.

@Gotlyfe 14 сағат бұрын

Hard to be excited about anything Facebook does after they chose profit over human wellness.

@Napert 14 сағат бұрын

If 70B requires over 24gb vram, i'm afraid of what will the 400B demand for it to run (or even walk)

@paul1979uk2000 11 сағат бұрын

That will be for businesses not consumers, the 400B model, you're probably going to need at least 256GB of memory, which you could do as system memory, but I would imagine the performance would be really slow through the cpu, but I do wonder how that would work with an APU and 256GB of system memory, if I understand correctly, that should be able to use the onboard gpu on the APU to speed things up, but system ram is slower than vram and the APU would probably be too slow for it to be useable. Still it would be interesting for someone to test that out once the 400B model is out. Ideally, what we really need is gpu's with a lot more memory, which vram on gpu's isn't that expensive, but I doubt AMD or Nvidia will bump up the memory that much for the consumer market, with the next gen high end gpu likely to have 32GB of vram and probably a high watering price tag to go with it, if you could get a more modest gpu but with far more vram, it would bump the price up, but it wouldn't be crazy high and could be useful many A.I. tasks, assuming the gpu can keep up with the A.I. model.

@darkoroberts 14 сағат бұрын

What a time to be AI!

@Pyxis10 14 сағат бұрын

Oh no. This is it folks. The llama has a hat on, which means Carl the Ai has already been made.

@breakitdown4346 14 сағат бұрын

What a time to have died and gone to heaven

@joeschroedernz 15 сағат бұрын

6:00 wow 6 gave me chills. It’s so small now, but it looks like simple empathy. Knowing the mind of the other robot to predict its mind and next decision. Technology is about to go crazy. 🎶

@ponyphonic 15 сағат бұрын

3:16 MMLU is one of those flawed benchmarks. AI Explained found hundreds of errors across all categories.

@Benw8888 15 сағат бұрын

This reminds me of how Microsoft published a report about bad actors (ex: North Korea) trying to use GPT4 for nefarious purposes. Now that Llama3 is open sourced, all those bad actors will be using Llama3.

@ZenTheMC 11 сағат бұрын

True. But there’s far more good actors. They can easily use the same tech to cancel out and severely outnumber the bad actors. Not having open source is just as dangerous cause of corruption and a lot of control and power for a small group of people.

@beansnrice321 15 сағат бұрын

The speaker for Llama looks like Patches from Demon's Souls.

@iks8622 15 сағат бұрын

In the EU, Copilot is the best all in one free AI.

@eleos5 15 сағат бұрын

Looks like both models memorized test results, but the 70b learned more from it.

@ralfkinkel9687 15 сағат бұрын

The models are still pretty big and hard to run locally by default, but you can use quantized versions (lower bit accuracy -> much less VRAM needed). I created a dataset and notebook on Kaggle if you want to try it out, it's free! Just google llama 3 kaggle gguf

@tcla75 15 сағат бұрын

I really hate to say this because I've been subbed to two minute papers for years and it was always the first place I would find out about new and exciting A.I features but now there are so many A.I youtubers out there and all of this was covered a month ago. The days of where there was very few youtubers covering this topic are long gone and waiting a month to release a video about a subject when others have done it within hours makes it old news. I know that he puts a lot more effort into his videos than the other guys but I just don't see the point in putting this much effort into a video that is already out of date. I fear that this channel might go into obscurity unless he can get these videos out within a few days of these announcement's.

@mertcuhac6303 16 сағат бұрын

i think the audio is ai generated

@IstvanTothwbtist 16 сағат бұрын

As a UK resident, I've found that this tool only functions when using a VPN with US servers. I'm thrilled to have discovered it and hope it remains free for a while longer, as it's become an invaluable resource for refining my conversations (Including this one) - a great alternative to ChatGPT!

@Johnathan_D_Hill 16 сағат бұрын

I wonder if this can be used to collapse(compress) data into smaller sizes.

@xa-1224 16 сағат бұрын

You can also get llama3 with ollama

@Gaukh 16 сағат бұрын

What a time to be a LLaMa

@jurgbalt 16 сағат бұрын

EU about to take the biggest self-inflicted L in human history

@RocketLR 16 сағат бұрын

Mmh the 70B model is the only good one. The 7B models are useless without any finetuning for specific tasks

@cumibakar10 16 сағат бұрын

Why do they make it free & open source? Doesn’t it destroy the market for AI? Like always, big tech creates a lot of value and then giving it away for free. Instead of creating a thriving industry and ecosystem with many different companies creating & extracting value of a new tech, other/smaller companies have no chance to compete with free.

Ең жақсы KZbin

Пікірлер