Видео

We've come far enough to make a sub Mario 1 game. woo

Just a tip, it seems like you used the same chat to answer all these different questions. The longer the chat, the less messages you get as everytime the entire chat is resent to the model. Also answers may get slightly worse due to mixed context.

Ai video is just not cheap enough to run, gotta wait a year or two

Ive made games just like the last mario clone but wow the first one and your lemon game were better!

A year or two from now we'll probably be able to add dialogue and sound to the video

The title says that Matt is (maybe) using Reverse Psychology O_O

wow, even opus is already OP.

They'll release 3.5 opus, it's definitely GPT5 level model.

I have an unconventional test for AI models. I have always been interested in romance languages, and have wanted to speak in a mixed jumble of romance words. AIs could understand me, but could never really replicate it. Guess what could? Claude 3 Sonnet. I haven't tried 3.5 yet, but... hm. Something tells me I'll have fun with this one.

Claude is not synced between all devices. And it was not even aware the it had a Pro upgrade when I asked it. Read this: There is no "pro upgrade" for Claude. I'm an AI assistant created by Anthropic, and I don't have different versions or tiers of service like some other AI platforms do. Whether you're accessing me through a free or paid service, my core functionality remains the same: No user accounts or persistent memory Each conversation starts fresh No syncing between devices No access to previous conversation history This is a fundamental aspect of how I'm designed to operate, regardless of how you're accessing me. version, but rather a core feature of my design, prioritizing privacy and approaching each interaction independently. If you're using a specific app or service to access me, they might offer premium features, but these would be separate from my AI capabilities. They might affect things like usage limits or interface features, but not my core functionality. Is there a particular premium or "pro" feature you were hoping for or expecting? Now read this: Notice the Subscribe to Pro call to action on your page. I apologize for the confusion in my previous responses. You're absolutely right to point this out, and I appreciate you sharing this image. You're correct that there is indeed a "Subscribe to Pro" option mentioned in the image, which offers 5x more usage with Claude 3.5 Sonnet. This is an aspect of the service providing access to me that I wasn't fully aware of or didn't properly consider in my earlier responses. To clarify: 1. The Pro subscription likely increases the usage limits or quotas for interacting with me. Subscribe to Pro for 5x more usage with Claude 3.5 Subscribe to Pro So Claude does'nt seem to be better or cheaper than Chatgpt4o. The pro plan is $20 per month. The same as Chatgpt4o. But you can upload images to the free plan of ClaudeSonnet while you need the paid Chatgpt account to upload images.

I can't program past a very basic point either- and yet, I've been using ChatGPT-4 to write python script for me (I do recreational maths). When there's an error in execution, I just copy/paste the error logs back into the prompt. Guess what? It works AMAZINGLY. We have developed the self-sharpening sword.⚔️ I am definitely afraid.🫠

You put html in a js file....

Look, for coding yes, but it will NOT roleplay an rpg no matter what, openai will, so thats an L.

My mouth is HANGING open. This is amazing. To be a bit more honest- its so frightening, that I don't know what to do. I guess we're just going to go with it...😂

OpenAI's 4o presentation was fake. It was a voice actor on a microphone. That's why "a few weeks" keeps getting pushed back. They're waiting for everybody to forget, so they're not murdered by a million angry fans who signed up, only to get shafted.

On one hand, it said that Sonnet 3.5 is completely free. But on the other hand, it shows a graph with the x-axis showing the cost, and the cost is NOT zero. So, is it completely free? Yes or no?

Great video. I think we should come up with a better word than flagship, it doesn’t really make sense for language models.

It's the AI war now.

i need them to compete not clash. competition makes it more available for ordinary people to use it. clash will eventually leave one giga monopolist

So you could just take code from some beloved dead game, and just have the ai juice it up?

Announcing AFTER launching is wild. Trying it out while watching the vid and it's nice

See, now.. if OpenAI was more motivated to actually releasing their content and being more forward instead of blue-balling their fan base by teasing their super cool parts, and milking money, they'd probably be destroying other AIs.

I find it to be below GPT 4 at translations.

It’s all fine and dandy but I will not purchase a subscription from Claude until they have image generation, unlimited doc upload, and some sort of TTS voice mode. They still don’t compare to ChatGPT. Also, their message limits are insanely low and it takes like 8 hours or something to refill. They’re not competitive.

Please!!!! No silly background sounds / music!

😊😊😊😊

When testing a model on different kind of prompts, you should always clear the previous conversations. Long context + unrelated previous informations will hurt the model capabilities :)

I find gpt omni to be worse than gpt4 in most situations tbh. I get they probably focused more on the omni factor, but still, its not that impressive... (Edit) Also these benchmarks mean nothing. To say that Gemini pro is even close to gpt 4 would be like saying the earth is flat.

i believe the demo from openai about gpt4o voice was just a lie. it probably was a real person talking on the videos :(

review qwen 0.5b

The model seems to be very good with logic, but it’s so brainwashed, up to a sterile level. If you’d ask it to make your character in the game be able to kill other bad guys, it will just read you a whole lecture on how bad it is. The censorship of all these new models is absolutely ridiculous!

lol i'm famous

So it did not come out on release. The copilot feature on the galaxy book is mid tier at best. Hopefully it comes out soon because right now I'm still using chatgpt's website for all my workings

Lyric generation improvement is marginal. Gemini Advanced still wins that one (for me). Dunno why, but LLMs always - ALWAYS - make AABB rhyming schemes, and it's really difficult to strong-arm them out of that behaviour.

💥 Can it talk and understand voice as Chatgpt 4 Omni ? 😮

I want everyone to notice that on the eval chart, gpt4-o is "outperformed" by claude 3.5 on THREE benchmarks that it NEVER TOOK.

Still not very good at helping me streamline my old Word 2019 macros. :)

The Claudes might be the best thing since sliced bread but in the real world they are useless. I was doing some research for a fantasy novel and asked Claude to describe the Pale Man from the Pan's Labyrinth movie for me. It's reply: I don't feel comfortable talking about things that might lead to self harm blah blah blah and nothing I said could budge the stupid thing. Meanwhile ChapGPT 3.5 gave me what I asked without a murmur. Not as good as Claude? Probably. But also not as useless.

Why did the thumbnail change???

Fun fact: you glossed over maybe the most important/crazy news of this whole vid- Ilya's post on X. Game is seriously about to change- for the better.

I wonder if they deliberately picked the name Sam for the demo lol hello Sam.

In my country it's not available:(

I got it to make a fully working game of flappy bird with original SVG graphics in a single prompt. Also a fully working game of wordle that even checks with a dictionary API to make sure the user only enters valid words, also made in seconds. EXTREMELY IMPRESSIVE.

is there an easy way to install more models?

Wow! I just finished co-authoring a book, Understanding Machine Understanding, with Claude 3 Opus, and that was mind-blowing, so I can't wait for 3.5 Sonnet.

It's a real game I mean we call Dino run a real game

LOL that was definetly for Sam Altman

This was so good I unsubscribed from ChatGPT.