[ML News] LLaMA2 Released | LLMs for Robots | Multimodality on the Rise

#mlnews #llama2 #openai

Your regular irregular update on the world of Machine Learning.

References:

Links:

If you want to support me, the best thing to do is to share out the content :)

[mephilees]

It's great to have you back, Yannic. The last month was dull without your excellent videos.

[henrischomacker]

Again, again and again: The most profound AI news by someone who really understands all that tech and is able to explain it to us "normal" devs so even we may most of the times really understand the mechanisms. - Many, many thanks for your ongoing efforts.

[Ronnypetson]

Yannic from the future is a good person. Always helping present yannic

[etopowertwon]

* Stability also released StableLM2 (CC BY-SA 4.0)
* Acoustic attack are not new. It's been known since 2005 "After just 20 attempts, the researchers were able to retrieve 90 percent of five-character passwords, 77 percent of eight-character passwords and 69 percent of 10-character passwords." (note, they didn't use shift key). I definitely remember some youtubers already muting their mics when entering passwords.

[Laszer]

I think that after having 700 million monthly users, the availability of Llama 2 model would be the least of my problems.

[PCPTMCROSBY]

Ohh, thank you so much. I have been going through major withdrawals, not seeing material posted up by you. You are perhaps the greatest person to teach these custom subject items. I thank you sincerely. Michael Crosby, candidate for president of the United States.

[MaJetiGizzle]

This is why I make the distinction between open-source software (OSS) and free open-source software (FOSS) in the case of both you can review the main codebase or in this case weights of the model. So llama-v2 is open source, but it’s not free and open source with free being open to unrestricted commercial use.

[stevewall]

This is the first channel that i would actually recommend for ML news, zero pseudo-bizness bullcrap.

Mostly the relevant facts and transparent opinions, thank you.

Also IBM has been great lately.

[DamianReloaded]

I am running Llama 13b in 4bit on my RTX 3060 12GB and it EFFIN works. I can talk to my desktop PC as Tony Stark talked to Jarvis, puns included.
I cannot stress this enough: HOLY S! O_O

[unbekannter_Nutzer]

AFAIK you ran into the confusing difference between Open Source and Free Software, which isn't legally fixed (IANAL), but there is some established convention. Open Source at least means, that you are allowed to read the source.
While Free Software has to obey the 4 principles of free software, as stated by Stallman, which are a) Use without restriction (obviously violated in this case), b) study and modify the software, c) redistribute it and d) redistribute your modificaions, c and d with the same license terms applied to those distributions. Examples are GPLv2/v3, for instance, and many others.

[JavArButt]

What can I say? "Having Lord Voldemort in charge of the safe deployment of magic." Your insights are very clear and understandable

[XOPOIIIO]

It fits on floppy disk, finally I could start using disc A again.

[AdamTwardoch]

There is nothing "strange" about Microsofts dual-strategy (OpenAI + open-source). One of Microsoft's key income sources are large business.

Those have decades of data now, and UNDER NO CIRCUMSTANCES will these business share that data with OpenAI or any such company. So Microsoft supports Llama and other open-source(ish) developments so that MS can go to their business customers and say "here's a nice solution that's fully private, this is how you take an open model, fine-tune it on your own data, and then you deploy it securely on our Azure".

MS will make money on a recurring basis on Azure compute & associated services. That's how it can build stable, recurring revenue streams — with simple "hosting" (which will be well integrated into the rest of their infrastructure), not with "cutting-edge products".

[AB-wfek]

In response to the White Castle segment, Wawa has been using touchscreen kiosks to take orders for years.

It prints a tiny bar code that you take to the register and pay while your food is being prepared.

It's one of the most convenient, fast, and efficient systems I've ever experienced, and you still have a chance to actually thank the person who made your sandwich when you pick it up at the counter.

[NoNameAtAll]

lama2 licence specifically targets against large players? cool

[jcorey]

Good to have you back! It's rare to find people who will talk about AI News and focus on the reality instead of the hype.

[uncletimo]

thanks for listing all the sources of the web in your desc

appreciated

[EnricoGolfettoMasella]

Come on, questioning if it’s open source or not cause of 700 million users clause it’s such a distraction, giving all the value that this model is bringing to millions of users (with less than 700 million active users). Besides that, any company with 700 active million users can make their own llamas, horses, lions, elephants and etc etc. Also, all this effort to cheer for ‘not open source’ is bringing confusion to the of the allowed users. Let’s thank Meta and all companies who are investing and allowing us, small business to have access to such amazing tech for free!

[jayhu]

So as always a great research about this topic. Keep going with your good work.

[KolTregaskes]

I've missed your videos, thank goodness you're back. :-) Hope you had a good break. BTW, Claude has been out in the UK for ages.