Clip from Lew Later (They Actually Pulled It Off…) –
What’s going on well why you got this in My face here announcing gpt4 is that What’s going on Uh yeah a large multimodal model multi Modal model say that 10 times fast with Our best ever results on capabilities And Alignment uh play the video roll the Clip or is the latest AI system from Open AI the lab that created dolly and Chat gbt gpt4 is a breakthrough in Problem solving capabilities for example You can ask it how you would clean the Inside of a tank filled with piranhas And it’ll give you something useful It can also read analyze or generate up To 25 000 words of text Can write code in all major programming Languages And it understands images as input and Can reason with them in sophisticated Ways that’s crazy eh this one right here Um just seeing this image and the logic Of the boxing glove hitting the plank And then the ball going up and then Explaining that Pretty interesting most importantly After we created gpt4 we spent months Making it safer and more aligned with How you want to use it the methods we’ve Developed to continuously improve gpt4 Will help us as we work towards AI Systems that will Empower us all GPT 4 is Extremely vague
Extremely I see another video extremely Bad give me another video gpd4 takes What you prompt it with and just runs With it from one perspective it’s a tool A thing you can use to get useful tasks Done in language From another perspective it’s a system That can make dreams thoughts ideas Flourish and text in front of you Gpt4 is incredibly Advanced and Sophisticated it can take in and Generate up to 25 000 words of text around eight times You’re done with the word chat like We’re growing up now that we can express We’re not a chatbot ideas about them for Example it can tell us that if the Strings in this image were cut the Balloons would fly away This is the place where you just get Turbocharged by these AIS they’re not Perfect they make mistakes and so you Really need to make sure that you know The work is being done to your level of Expectation but I think that it is Fundamentally about amplifying what Every person is able to do Gpt4 training finished last August and Everything that’s been happening in the Past few months up until we’ve released It has been a giant Sprint make it sea Formal aligned and also more useful we Have put in already a lot of internal Guard reels around things like
Adversarial usage unwanted content and Privacy concerns and when we release the Model we know things are not done we Know we have to learn we know we have to Update we know we have to keep improving All the systems around it to make it Suitable for society To me the most compelling use cases of These technologies will come from Starting with a real human need the Obvious one where these systems have Really incredible potential is in Education gbd4 can teach a huge range of Subjects imagine giving a fifth grader a Personal math tutor with unlimited time And patience it’s a great tool to bring Learning to everyone in a way that is Personalized to their skill level G54 brings the dream of having the most Useful helpful assistant to life it’s Really about adding as much value to Everyday Life as possible The partnership to everyday life is to Shape this technology into something That’s going to be useful for the world The power of AI hopefully is that it can Help us be more productive it’s a low Voice leads to better quality of life The development where’s the mic Would you hide that mic inside his Throat everything came together who Produced tucked in there and while it is Very limited it’s already easy to Imagine what the impact sounds like the
Charlie Brown teacher We think back GPT I would like a voice Experience with a highly capable and Advanced AI system so we really care About this model being useful to Everyone not just the early adopters or People very close to technology so it is Really important to us that as many People as possible participate so that We can learn more about tattoos Headphones those focals those are a Couple bucks on them He’s in those headphones he’s the only Guy actually working did you notice that Yeah everybody else is like chatting We’ll give you the expensive headphones Yeah tapping the table and then there’s The other guy just Just plowing through with his headphones On 1500 headphones I understand that it is a much more vast Than the previous version and Apparently it seems to understand Context well now but that seems that’s My takeaway they’re like what if you cut The balloons and you’re you’re Responding to it’s responding to an Existing piece of data that you you’re Providing or reacting to and then it’s Reading that instead of being strictly Reactive to text Pro to text analysis Now it’s like looking at an image and And also text like with the question
Sure so in this case uh they’re showing Off a picture of the inside of a fridge Which says what are a couple of meals I Can make with this And so obviously gpt4 has to analyze This image and then make suggestions Based on the items visible in the fridge This is the response here are a couple Meal ideas yogurt parfait layer yogurt Strawberries and blueberries in a bowl Or jar topped with honey or nuts for Added texture and flavor I can see that You’ve got the yogurt near the front of The fridge you have strawberries and Blueberries easy Carrot and hummus wrap wow the hummus is Impressive to me is this the hummus no It’s over on the left it’s that short Container yeah To know that that hummus is based on What because that could be a lot of Things like that’s impressive and maybe They they tweaked this response and Hummus does tend to come in a container Like that yes But is it looking at the texture is it Looking at that very specific color it’s Incredible carrot and hummus rap hang on Let me just read the rest of it Spread hummus on a tortilla wrap so it’s Not seeing a tortilla in there but it’s Hoping you have one somewhere else not In the fridge top with shredded carrots The carrots are there other desired
Veggies roll up and enjoy Note the lack of Honey nuts or tortillas In the image so it’s able to fill in the Other suggestions along with what it Happens to be seeing In this example it shows off a meme of Uh I think Chicken in like a shape of The world And uh chat GPT identifies it So it’s also referencing against its Enormous data set yeah and it’s like Overlaying extremely quickly to be what Should I be noticing here Um so it immediately knew this was the World this meme is a joke that combines Two unrelated things pictures of Earth From space and chicken nuggets the text Of the meme suggests that the image Below is a beautiful picture of the Earth from space however the image is Actually of chicken nuggets arranged to Vaguely resemble a map of the world the Humor in this meme comes from the Unexpected juxtaposition of the text and The image the text sets up an Expectation of a majestic image of the Earth but the image is actually Something mundane and silly that’s very Detailed that description That explanation What is unusual about this image well Here we have an image of a of a guy on The back standing on the back of a taxi And he’s ironing a shirt and he’s
Somehow harnessed in there It’s very odd image why would you be Doing this and so so gpt4 responds the Unusual thing about this image is that a Man is ironing clothes on an ironing Board attached to the roof of a moving Taxi I mean Nails it it’s like hey I’m like You I understand a Goofy image but does Chat GPT understand that it’s funny or Weird because in this case it says no Emotion well just kind of stating the Fact I would say it would just take the Definitions of these terms in this case The question was unusual now once you Look at a long list of synonyms or close By terms unusual can be funny so you Would have that in your model somewhere Unusual can be funny unusual can be Scary So you’re right this gets complex But the main thing is it’s capable of Determining unusual unusual is a hard Thing to determine and it’s capable of Identifying unusual in that case yeah Like there’s not a lot of images in its Data set that resemble that how many Images does it have of a man ironing Back on top of or back of Taxi yeah a Limited number There might be only one image And then it probably tries to find the a Description of it or something it’s Better at playing with language open AI
Says that gpt4 is better at tasks that Require creativity or Advanced reasoning It’s a hard claim to evaluate but it Seems right based on some tests we’ve Seen and conducted Uh through the differences with its Predecessors aren’t startling so far Okay so the differences currently are Not startling People may need a little time with this To determine where those differences lie Revised summaries new definitions more Tweaking you know people are going to do They’re going to throw all their trumps At it and