AI NEWS : Meta CANCELS New AI Model Release, Q-STAR BANNED, LLAMA 3 Release Date, and more

00:00 - Meta withholding multimodal AI models from EU due to regulations
03:34 - Meta planning to release Llama 3 on July 23rd
07:29 - New research paper on "mixture of a million experts" in AI
09:22 - AI-generated video quality comparison over time
11:24 - Potential classification of AI research as state secrets
15:21 - Samsung's new AI sketch-to-draw tool
18:43 - AI market and "bubble" speculation

Links From Todays Video:

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

[TheAiGrid]

If you see comments about some random name/token do not interact or buy any crypto - ill try my best to remove comments

[Corteum]

_"how EU tends to regulate everythnng, while america and everyone else tends tinnovate"_

Yeah but there's trade offs. and the other perspective is that eu doesnt appear to trust american corporations for some reason. i wonder why

[martinp.l.]

Well, since these models are being trained on the personal data of EU citizens without neither their consent or the possibility to opt out, perhaps the EU should declare the models public domain.

[xeloprint]

Nuclear regulations in the US did not prevent nuclear development in other parts of the world

[TiagoTiagoT]

The fines for privacy violation are low enough to still make it profitable outside the EU, that's why Meta bitches about Europe.

[Ulfberht]

Oh no we wont have Meta stealing our data :( sad

[steve_jabz]

The sketch to draw tool is nothing new. That's literally just masked inpainting. We had that on the launch of stable diffusion 2 years ago.
"midjourney can't do it cos it only generates the entire image" really says nothing. Midjourney can't do most things.

[kvinkn]

Living in the EU is so much fun when you want to try new

[KatharineOsborne]

"EU is insane on regulating things" what like food safety? Americans eat chlorinated chicken FFS. You could do with a few more regulations. At the least the EU isn't controlled by corporations.

[martinsherry]

Personally i like the idea of privacy and I like the idea of fair marketplaces that are not dominated by tech behemoths exploiting their size to the detriment of smaller competitors. And I believe that it is something that is worth regulating. Please don’t take Meta/Apple’s spiel about “uncertainties” at face value - they are simply trying to reduce the costs of implementing legislation that would force them to protect an individual’s right to privacy, and to interoperate with other organisations. They used the same playbook in delaying USB-C, and with alt App stores. It troubles me that YTers such as yourself adopt a stance and a soapbox without considering that there might be another perspective worth investigating, rather than just spouting corporate dogma.

[Youtube_Summarizer]

Here are the key points and the important details of this video, created by @Youtube_Summarizer

Key Points：
1. [Key Point 1]: Meta withholds AI models in EU due to regulatory uncertainty.
2. [Key Point 2]: Meta to release LLM-3, a 405 billion parameter model, on July 23.
3. [Key Point 3]: AI-generated misinformation tools raise concerns for content authenticity.
Important Details:
Here's the timeline, created by @Youtube_Summarizer
00:00:01 Meta's Decision to Withhold AI Models in EU
• Meta won't offer future multimodal models in the EU.
• Lack of clarity from EU regulators cited as reason.
00:02:39 Meta's Multimodal Models and Future Plans
• Meta plans to incorporate new multimodal models.
• Models to reason across video, audio, images, and text.
00:05:05 Llama 3 Model Release and Open Source Implications
• Meta to release largest LLM-3 model on July 23.
• Model has 405 billion parameters and is multimodal.
00:07:48 AI Research Advances and Lifelong Learning
• Paper introduces efficient expert retrieval for AI models.
• Enables lifelong learning and scales to millions of experts.
00:11:37 AI Regulation and National Security Concerns
• White House officials discuss AI regulation with tech leaders.
• Concerns raised about AI's potential catastrophic risks.
00:15:59 Misinformation and AI-Generated Content
• Samsung's AI sketch tool raises misinformation concerns.
• AI can implant photorealistic creatures into images.
00:18:26 AI Market and Hype Cycle
• AI market may be overvalued in certain areas.
• Transformative technology despite hype cycle concerns.

We hope this summary helps you better understand the video content! If you have other videos that need to be summarized, please visit @Youtube_Summarizer

[LiquidAIWater]

There is a difference from a financial bubble of a transformative tech and the tech's future. Remember, the Nasdaq crashed 70% from the peak after 2000 but the internet didn't stop growing or changing the world.

[srit]

Releasing an AI model in EU it's not an issue, using EU data to train it without consent of the citizen, is

[legendarystuff]

I tried for a year to get Midjourney to edit real pictures (needed some sales photos for clothes) there is no absolutely no chance, I tried a dozen different techniques, they clearly blocked this option knowingly. If you use the blend picture and you want an original image to just be modified very slightly so you add the same image twice to increase it's strength, they threaten to ban you

[gubzs]

The path to AGI/ASI should be secretive and not just open source for everyone, but the party we are trusting (the US military industrial complex) is not morally qualified to carry that torch.

[LiquidAIWater]

It wasn't the keeping of the nuclear secrets by regulating them that kept people from creating a nuclear weapon, it was the regulation of plutonium and uranium and how difficult it is to refine it.

[TheManinBlack]

Marc Andreesen is an owner of the a16z, a venture fund heavily invested in a number of AI startups, so I wouldnt necessarily trust him when he has such clear direct vested interests.

[railville]

how is the Samsung image thing any different to Photoshop generative fill?

[hypersonicmonkeybrains]

The problem with the national security state getting involved in AI tech is that they can see potential bad things in virtually any AI tech you can imagine, and so could classify the whole lot. From SORA to Eleven Labs voice, GPT-5 to AGI it could all be conceivably used in bad ways that could be argued a national security threat in a senate or whatever, and so we could see all AI tech made secret by say next year.

[Sandra-svow]

Tech companies ask for regulations and get said regulations and complain. Tell me tech companies just wanted to do some hype marketing. 😂