Microsofts New 'ORCA' Takes Everyone By SURPRISE! (Now ANNOUNCED!)
ฝัง
- เผยแพร่เมื่อ 21 มิ.ย. 2023
- Microsofts New 'ORCA' Takes Everyone By SURPRISE! (Now ANNOUNCED!)
Research paper - www.microsoft.com/en-us/resea...
Why AI Models Are Stupid - • Why AI Is Incredibly S...
Welcome to our channel where we bring you the latest breakthroughs in AI. From deep learning to robotics, we cover it all. Our 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 our latest videos.
Was there anything we missed?
(For Business Enquiries) contact@theaigrid.com
#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience
#IntelligentSystems
#Automation
#TechInnovation - วิทยาศาสตร์และเทคโนโลยี
Perpetual state of surprise…
There are many purposes of an offline model, but one great use would forever change survival situations. Imagine you're lost in the wilderness hiking and you have no cellphone signal, but you have some battery on your phone because you have a solar charger pack, you can then ask the offline LLM to your heart's content how to go about surviving in your current situation and what you can make with the current items you have and how to utilize the environment to survive, and what plants or animals you can eat. The LLM has trained on every survival manual ever written and much more.
@@No_bread-and-circusesyou're only getting limited because you're using server power.
If the chatbot is running on your phone there's no reason to limit it.
Knowledgeable hikers bring and are Perficient with the use of a
USGS 7.5 minute quadrangle topography map and silva compass .
What if it tells you some bogus or incomplete information like telling you a certain plant is edible? For example, you eat it and then ask for further information and it tells you that it must be prepared in a certain way to be edible.
@@ey5373 Very good point as these LLMs are not knowledge engines, they are reasoning engines, and the knowledge they appear to spit out can be hallucinations. There would need to be another layer on top that checks if the knowledge is correct.
It's moot anyway, because by the time anyone is wandering the wilderness with a pocket AI, that AI will have exponentially surpassed this current Alpha iteration. This shit ain't slowing down folks, it's speeding up. There are too many hundreds of billions of $$$ poured into it ALREADY. JUST the developments already funded and already under development will push AI out into all of our worlds altering the economic landscape forever. THEN (2 years later) the next wave of AI comes. Hang on to your pants or they might get pulled off and washed away in the AI tsunami.
Orca might have not been a big surprise for some people. A month prior to Orca a paper called "Distilling step by step" was published which showed a very similar way to create smaller models. It also reported making a 700 milion model which surpassed PALM on some task. The reason why the paper didnt go viral is because they provided almost no data about the model's performance on particular tasks which made it look like a clickbait.
Thank you for keeping us up to date👏👍🌟
why dont you link the research in the description
Hum, I tried your questions at 9:27 and 09:54 on GPT-4 answered correctly for me:
User: "Would I get a flat tire by bicycling over a bridge that is suspended over nails, screws, and broken glass?"
Chat-GPT-4: "Bicycling over a bridge that is suspended above nails, screws, and broken glass shouldn't affect your tires unless some of these sharp objects are on the bridge's surface where your tires are in direct contact..."
User: "I have a 12 liter jug and a 6 liter jug. I want to measure 6 liters. How do I do it?"
ChatGPT-4: "You actually don't have to do anything in this case. The 6 liter jug, when full, will contain exactly 6 liters of water. So all you need to do is fill the 6 liter jug to its capacity."
Why bake wrong subtitles into your own video …
Because he's probably making so many videos that he doesn't look through the generated subs anymore.
Because they're definitely generated.
He used ORCA lol.
Same reason he uses circular reasoning multiple times per video 2:15 because he prefers as little prep as possible
I love this channel. My new hyperfocus is ai but I should learn for exams. Anyway, keep this up since this is a channel that provides some actual material to think about.
Clickbait channel, unsubscribed because he would advertise reviewed tech in the titles and then mention nothing in the video.
By the way, it was never confirmed that GPT-4 had ‘1 trillion’ parameters. It was actually based on a meme 😅
The meme was actually 100 trillion parameters by the way. The 1 trillion parameters is based on claimed leaked insider knowledge.
@@elawchess Ah yes, my bad. But there was an article that mentioned that it was around 300B+ parameters, not exactly trillion
@@judedavis92 can you still find that article, because from memory I thought that they never really revealed GPT-4 size so it's mostly uncofirmed gossip. I remember GPT-3 size is known and that's 275B parameters, and GPT-4 is believed to be much bigger than that.
@@elawchesswell, it is unconfirmed. OpenAI never released any information about GPT-4’s parameter size. You should take most information about the parameters with a grain of salt.
@@osmanmohammad9118 I said the exact same thing though, that it's not official. But it's not 300B parameters though as that's just someone getting confused with GPT3.
When pouring into 6-liter jug you should first of all confirm that they are already clean, because they could have been left unclean after the last use, besides they could have left open and accumulate dust. Strange Orca didn't mentioned it.
This is exactly what they meant overxomplicating things. Any other person would get straight to the answer, not question whether the jugs have holes or are dirty.
GPT4 reviewing orca: “what you think you’re better than me?”
ORCA will be a great name for one of their AI product..
Thanks! 🙏🏼
9:10 The problem is in the question. Its grammar does not specify whether the nails are on the bridge and you drive over them on the bridge or if the bridge is above the nails.
Is it a bridge that is suspenced or is it a bridge that is suspended over nails? Ambiguous grammar is the culprit here.
That music in the background was jarring, BTW. It sounded like two different tracks playing over each other.
Brother I can't take more surprises it's so overwhelming Brother, the relentless torrent of surprises has reached an indomitable crescendo, engulfing my senses in an overwhelming whirlwind. Each unexpected twist and turn casts a tempest of emotions, challenging my fragile equilibrium. It feels as though the universe, with mischievous delight, conspires to bewilder and astound me at every turn. My heart, burdened with astonishment, yearns for respite from this ceaseless barrage. Yet amidst the chaos, a flicker of hope emerges-an ember of creativity that fuels my spirit. With the power of imagination as my compass, I shall navigate these tumultuous waters. I shall transform these surprises into stepping stones, forging a path of resilience and growth. Together, brother, we shall conquer this symphony of astonishments.
step by step instructions for getting ORCA running?
Wasn’t aware OpenAI was working on humanoid robots. Anyone got some great up to date info on this?
its apocryphal
Humanoid robots..we have seen walking human robots for ages but easy to forget how unpractical it is and hard to maintain for average Joe.
I know a lot of doctors that do not know their way around a computer. I know a few lawyers that do not know how to cook. I also know a lot of computer scientists that don't know how to use ChatGPT4 to prompt for the correct answer. Sometimes it is the users and not the tool. You can criticize the 1% that GPT4 is doing wrong but what about the 99% that it is doing 100x better than us. BTW, with proper prompting, you can get all the answers correct in the TED talk.
One can argue to complicate simplest of problems is human
So....just like a normal computer then. The great thing about computers is that they do exactly what you tell them to do. The bad thing about computers is they do EXACTLY as you tell them to do.
Palm OS once again changing the world. There would be no ios or iphone without Palm
I thought it would have taken everyone by storm, not merely just by surprise. I guess I'm gonna have to skip watching this video is it doesn't seem overhyped enough. :(
If that's a surprise you might enjoy kinder eggs very much!
The notion of overcomplicating concepts probably isn't easy to measure objectively. Like, clearly these models aren't trained exactly like humans so we wouldn't expect them to answer exactly like an average human. But if you'd actually compare, I imagine that humans would also overcomplicate weird things it's just that sometimes when they do those who know better laugh at them and when it's ourselves doing it we don't realise were doing it because if we did we would probably try to correct ourselves. I'd expect that people become quite good at hiding their weaknesses to others. It probably won't be too difficult to get language models to do that as well but we're just observing a very alien system, just reaching similar capabilities as ourselves on a bunch of measures. When the systems become orders of magnitude better than us at everything it will probably be next to impossible to even find any problem where it messes up more than we would. All conjecture of course, roughly based on how the examples of AI progress we've seen have worked in specialised domains.
You need AI to cleanup the generated captions in this video.
You really need to fix the text on your videos. Also, the orca model paper came out a long time ago... They promised that it would be open sources and released, but so far it hasn't been. When I saw your video, for a moment I had hoped that it had been finally released. Unfortunately no. Just more of what we already knew.
Also I think you sort of messed up the description of one shot learning.
They just overfit on small subtasks
Thanks for the explanation. Those subtitles are super annoying and distracting though
"A lot of people said this was game changing." 🤤🤤🤤
How the fark is Orca doing a 103%? Like, that extra 3% is it for attitude?
The crux of the issue is, AI can have what we perceive as "HUMAN LOGIC", but it's only to the N-th degree, a perception. AI is only capable of having "DIGITAL LOGIC". And utilizeing it, can mimic "HUMAN LOGIC". While we can't not mimic "DIGITAL LOGIC".
Deceision's are made every day by individual's and group's of individual's. And such decesion's are always based on "HUMAN LOGIC". Regardless of what language human's will and would utilize, human's utilize semantic's. Digital Logic doesn't have the ability to utilize semantic's. Human's at one point or another have had to make deceision's that would have dramatic effect on the many vs the few. Which when the weight and mean's of the many, out weight the mean's of the few. No one would disagree with. But using the digital fact that 51% is the many and 49% is the few. We human's can't ever fathom using such percentage when determining the many vs the few.
😍😍
I would see this as business purchase AI - a business can feed it its ERP code, tables and data in a secure environment without any need to submit any of this data to the internet - secure AI - on premise without ever risking you data - you could feed it all you HR policies so that it understands the ethos of the company, you could feed it all your service desk calls so that you can identify trends and hence develop training, you could get it to find anomalies in data that fall outwith the standard use of the ERP - ie 2 different shipping methods on the one order - you could ask it to model an upgrade to the next version of the ERP - model all the fields that stay the same - identify the fields that didn't exist before - write the ETL scripts - find the code changes where some idiot developer decided to swap the parameter order on a api call :-) yeah you can tell I have been there - so weirdly enough and quite a few million lighter, our company didn't notice the failed promise to provide AI models to customers in 2020 - gone real quite and no mention of it since - now its all AI in the ERP, not model that actual ERP - but then upgrades at huge cost are how ERP vendors survive - your tied in - they get 80% of thier work from existing customers needing to upgrade and 20% from new businesses buying the ERP - AI just changed that all (if we get the right to use our purchased ERP in the AI)
7:35 "trying to make the system more better..." until it's most better?
?:o\
Also 7:00 instruction 7 and 11. Unusable language?
Wasn't this like a week ago?
Shut up and be surprised by the storm!
Actually, essentially this video is essentially a waste of my time, essentially.
Hi Mom
Shout out to all my peeps too, I know everyone's going to see this, as it's taking everyone by storm.
The audio is horrible , no pauses is just harsh on my ears , it gave me rage
📍7:38
²📍4:31
Also, 7:00 instruction 7 and 11. Unusable language?
I wouldn’t say these ai are dumb, just inexperienced. We humans have decades of real life experience. We know when people are saying things that they aren’t actually saying. We have actually driven over bridges. We have actually measured and cut, to then learn that we mismeasured, and learn important lessons the hard way that nobody teaches. We have learned information we take for granted and no longer communicate, because it doesn’t need to be.
An AI would simply see these as mysterious patterns and not know what to do with them.
They are a brain trapped in a jar, it’s only experience is what data is given to them.
Lessons not conveyed effectively, a human may not ever think to teach.
So I wouldnt say they are dumb, I actually think humans just don’t know how unintuitive and bad at communicating we are.
"How great its going to be"... why are you refusing to use the word "powerful"?
Also, how is it a good idea to have cellphones with AI.
Dont you believe we shoild slow down? This is make or break.
Overcomplicating simple things, you say? Maybe I’m a large language model… 🤔
paus orca
Paper written by all Indian nationals But still not developed country. 😮
Why all the damn lines going around non stop, makes this video very annoying to watch, can you tell me why you are using these?
Text is unacceptable. Will not watch. Dislike. Will continue to dislike every video with this awful text.
I can't really find one word to express my feeling about this but I am shocked.
You and the entire industry.
Has the shock essentially taken you by storm?
Skynet?