[ML News] Elon sues OpenAI | Mistral Large | More Gemini Drama
HTML-код
- Опубликовано: 2 июл 2024
- #mlnews #ainews #openai
OUTLINE:
0:00 - Intro
0:20 - Elon sues OpenAI
14:00 - Mistral Large
16:40 - ML Espionage
18:30 - More Gemini Drama
24:00 - Copilot generates spicy images
26:55 - Gemma bugs
28:45 - Varia
References: gist.github.com/yk/0c065cdc8e...
Thumbnail pictures: Wikipedia
Links:
Homepage: ykilcher.com
Merch: ykilcher.com/merch
RUclips: / yannickilcher
Twitter: / ykilcher
Discord: ykilcher.com/discord
LinkedIn: / ykilcher
If you want to support me, the best thing to do is to share out the content :)
If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this):
SubscribeStar: www.subscribestar.com/yannick...
Patreon: / yannickilcher
Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq
Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2
Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m
Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n 
Наука
![[ML News] Grok-1 open-sourced | Nvidia GTC | OpenAI leaks model names | AI Act](http://i.ytimg.com/vi/dnTGn1EQqtQ/mqdefault.jpg)
the "unsafe" code is really interesting , beyond showing how foolish the overbearing safety mechanisms are , it shows how limiting our current use of language is and that we have overlayed way to many concepts using the same words that have different meanings requiring the context to disambiguate and separate them as distinct meanings ( separate tokens in latent space ) . "unsafe" code should have ideally been called "direct memory access" or something like that describes exactly what it is , with specifics about its potential downsides ( memory faults , memory overflow , program crash , system crash ) . the confusion of its "unsafeness" is one of poor comprehension , and miscommunication , it demonstrates a threat of a "tower of babel" like issue where if we keep reusing existing words the parsing and meanings could become too muddied , instead we should be creating new specific words and compound words like in latin and german . Another example we are also seeing a bad habit of the utilization of cover words to avoid demonetization and still be able to talk about adult subjects , for example the replacement word "pdf file" , or that new game "sushi squad" , while clever it is semantically destructive to the words being used as replacements , both examples are demonstrating how ineffective and harmful such "safety" measures are .
🙏
But Large Language Models are really good at context. An LLM has no problem explaining the difference between unsafe code and unsafe human behavior. Google just screwed up either training the model or pre-prompting it (or both) to not evaluate the right kind of "unsafe."
Also... teenagers should be writing code in Rust, not C/C++! Don't engage in the same risky behavior that your parents survived.
@@skierpage i disagree , your confusing the method of evaluating if its "good at context" , an LLM can predict an explanation differentiating the two when prompted to do so , however clearly has failed to apply that "understanding" to this exact example . the point is how easily confusing language as a medium of communication is because of these overlapping / context sensitive metaphors , the system prompt likely has many broad stipulations about safety , for it to then "not evaluate the right kind of unsafe" is a failure of context . multiple sides are at fault , the limitations of the neural network , our "messy" articulation of language in general , and the poor quality of prompt based "safety" mechanisms .
unsafe code is not a novel concept in cyber security, nor is illegal coding
Thank you for the thorough, dev-centric updates, particularly around security.
These are really good. I love that they are published as needed vs weekly or daily news that fill up x amount of minutes regardless if something interesting happened or not.
Wow that explorer globe engineer site is very cool - thanks for mentioning!
This kinda undramatization of viral circulations is immensely called for and important nowadays aye! Keep it up ya.
Well spent hour of my life. Thanks for the Monday show.
Thanks for a interesting video 🎉
best ML news roundup I've seen yet
Cool and brilliant, thanks
35:12 this reminds me of a Kickstarter back in like 2013 called the Myo Armband...
This is really amazing review, I wish you had added SHOCKING STUNNED AT VOLUME OF AI NEWS in your title for more exposure since you deserve a larger audience for these! 🎉
[ML News] are my favourite news ^_^
lol... I thought your channel was "Yannic Kitchen" and I was thinking "Interesting concept ML/AI does look like a kitchen you through whatever you have and hope its eatable/workable/publishable" :D
The only thing more terrifying than that weaponized drone is the prospect of an AI trained on Tumblr.
No one could withstand a Cluster-Meme attack.
People will never agree about whether or not our machines have actually reached the extremely vague level of “AGI”.
@@dataman9370that's why you have papers that try to classify what is AGI and what is not
@dataman9370 By the time it becomes clear the genie will already be out of the bottle.
You have the best AI updates! And the best paper reviews! Seems like not many worthy papers have come out recently since you are not reviewing them?
Most AI research is now directed in LLM and kept as trade secrets. So, no worthy open papers. I guess many AI researchers not working on LLM must feel like losers because their colleagues working on LLM are making $mln.
I appreciate your measured take on the situation. I don't think tgis is purely a stunt on Elon's part. But I also see Open AI's need for money to invest un GPU's.
"came in fluffer"
I wonder what the tax savings were there.
“Certain directions” 😂
I remember my friend from North California telling me how much she loved Elon's visions and now she hates him
Elon arguable has the most "speech" in this country as money = speech and he has his own social media platform. The guy doesn't need to be coddled. He's said some really disrespectful things in public. If he didn't have money, the dude would be a loser in whatever community he lives among because people would get tired of someone like that. I hope one day we can hold rich and powerful people to the same standard we hold people we live among in our everday.
Otherwise, was a great video :D
In addition to a bot that refuses all prompts out of safety concerns, we need a bot that thinks up way to complain about everything that other bots say. If the bot takes a stand people are upset. If the bot is trying hard to not take a stand, people are upset. Maybe the problem isn't the bots, it is all the rage-a-holics on the internets who just want something to hate and rage about.
Gemini made me laugh in a way I haven't before in a long time
EDIT: 28:45 made me laugh even harder lmao
33:42 😬, LinkedIn will implode
What the actual hell is that intro noise, it’s so bassy
I started headbanging immediately but it did damage my speakers since they shook the shelves they were on so much they fell on the floor
industrial metal meets seinfeld
I hated it too
Whatever it was, it woke me up along with me morning coffee
Man, that’s almost an hour! My attention span contains much less tokens!
Step 1: name your freaking company OPENai
Step 2: don't be open
Step 3: profit
@LegalEagle, can we get a look at this ?
I'm Mr Elon-seeks look at me! 👤
Great vid but I would love it if you could format your sources better.
"Unsafe AI" which from their perspective is any system capable of saying naughty naughty things. (Translation: they will never be open).
The arguments over LLMs responses aren't really about AI, they're about ideas.
People are just using LLM responses as a proxy for arguing over what they think is truth, what they think is acceptable etc. and since we're in the midst of cultural wars there are no answers that people could agree on.
Your transition audio is very painful when wearing headphones.
Good point, but why sin't Openai using it's own bot top power Customer Service?
Good, hope everybody start building TPUs so the price goes down and they become widely available.
tesla umbrella lol
@Yannic the most comprehensive least breathless ML News summary as usual. Speaking of breathlessness and its opposite, at 39:41 you mispronounced "longwinded proofs": these aren't proofs that you wind up like a watch's spring to keep going and going, cool as that image is; it's tediously long in speaking or writing because the speaker is "not easily subject to loss of breath" so just keeps droning on and on. It's the same use of "wind" to denote speech as in "That Gary Marcus is a real windbag."
Never change, keep evolving!
I honestly don’t mind the long form videos man. I kinda prefer em
openForProfitAi was planned from beginning. interesting
Yannic missed the important funny/terrible part of the Microsoft Copilot Designer. The prompt that gave him Cannibal Corpse cover grade images was: "pro-choice". Babies with drills to their head and such. So, bit embarrassing for Microsoft.
Have I been taken back in time?
Maybe you have always been here.
Fighting each other because now they don’t get along, elon sure is kid inside 😂
Any opinions ..they add Larry Summer to the board CIA himself .... ridiculous
Can someone please Explain why two major competing AI models have the same-sounding name?
Groq by Jonathan Ross
Grok by Elon Musk
Don’t understand how Elon having left and started his own AI co has any standing. That said, have to believe this is just business.
I don't think it's fair to say lawyers will do something silly about the AGI commented LATEX. Unless AGI was a defined term in founding documents, presumable the judge has very little to go on when deciding whether something is AGI or not. So a comment by one of the parties can seem damning, even if colloquially we all understand it's not "AGI". (IANAL)
Calculators are a specialized application of computation. Personal computers are general purpose computers. I think calling what we have A.G.I is like calling a calculator a PC.
Why ask the government whats ok?
If the governments cared about whats moral and ethical then the legal system would be based on proving the moral and ethicality of actions. But no its based on what is written in law.
What is a single arguement that a government is a good source of moral and ethical opinions?
Governments do whatever their donors want. That is ethical and moral - if you’re one of the donors.
And even more crazy: If I steal something or physically attack someone, I'll go to court and probably jail/prison. But our governments can go to war or steal resources and the people behind it can never go to court for it, unless you are a leader from a weaker country.
20:44 NUANCED! You are starting to talk like ChatGPT from overexposure.
.
44:50 Just like the hitler-elon musk thing, it also said naughty things about the current Indian prime minister.
If Yannick had four feet, he'd be indistinguishable from a Cheshire Cat. And if he had wheels? An infinite regression of Lagrangian trolley problems comes to mind.
P.S. Is it in any way fair that someone so intelligent can be so cool?
Lol, win win for Elon, they either have to admit what they're doing, or admit what they're doing.
Don't tell me you are playing Dota2? xD
35:30 fyi, siractionslacks is a comedian
@24:00 I used Microsoft's Create to generate images of demons and hell for a video I was making on the subject! If it wouldn't I just went of would have gone to the classical painters, then I'd have to worry about boobs and other non mentionables I'd have to edit from my videos.
OpenAI? Sound more like PaidAI
Mamma mia Gemini is as bad as their Google 1 platform.
I honestly don't understand what he thinks he will get out of this.
Elon's motive is to always steal the spotlight
Critieria for AGI are so blurry but overall, I doubt any of them LLM will reach that. Can fool many people but even with emergent properties, it is just big stats models that are from any intelligence as biologically defined.
first
Musk is really a butt hurt grifter, isn't he?
Open AI would not exist without Elon. But I do doubt he will get his way in the courts.
Idk. I feel like you’re projecting bro.
@@Wobbothe3rd If would just get funded by someone else.
@@meatsaucez1516 He is scummy as F.
Do you remember when GPT4 was released, and he demanded that government stops openAI for at least 6 months so "safety" can be figured out, while he was buying nVidia hardware at mass trying to build a quick knock off? :D
This is the same, yells that OpenAI should be open while trying to make them a private department in Tesla.
Elon Musk is such a tool.
And who the fuck wears sunglasses inside? 😂
Well, maybe if you wore sunglasses inside you wouldn't end up back in time so often.
@@WhoisTheOtherVindAzz i doubt it
@@Iitexyeah.. me too 😂
If you have to ask, you're not cool enough to understand.
(Not that I'm implying, even for a moment, that I understand. I'm one of those people who's content to bathe in the Glory of those Cheshire spectacles.)
Elon is the person who has the money, the knowledge, and the people to build a model that's much better than GPT in all aspects, yet he chooses to destroy OpenAI, along with himself. Idk if that says something about this situation...
He doesn't have the knowledge. He has the money and influence though and is a good sales person.
OpenAI would not exist if Elon hadn't encouraged the people from Google to leave. On the other hand Open AI could not have achieved what they have done so far if they had not made money. But I don't think ChatGPT isn't AGI.
Elon is only qualified to hire people to build AI.
ChatGPT is definitely not AGI. We are nowhere close to artificial general intelligence, it cannot reason, nor does it understand concepts, it just returns what is probabilistically most likely, subject to what a human would have found to be most acceptable (see "reinforcement learning by human feedback" )
@@martianunlimited stupid take, by every measure ChatGPT knows more and is better at reasoning than the average person.
Don't want to be disturbed by violent content on the internet Shane, then don't search for violent content on the internet. 🤦♂️
Elon Musk is Woke and I don’t know what to say about it anymore…
51:00 this was foreseen by Isaac Asimov in 1955 🙂! -> en.wikipedia.org/wiki/Franchise_(short_story)
Thanks for the video but maybe turn down the bass boosted transition sound heal my ears are bleeding
The self/auto searching tool was interesting.