Want to compare the results from the chatbots yourself? Check out ChatHub now! They offer one of the easiest ways for you to compare and evaluate the results to find the best LLM for you! chathub.gg/?
4:15 When you use GPT-4 on the ChatGPT interfaces you are automatically using GPT-4 Turbo with 128K Context. The GPT-4 32K is the original 16-MOE with 8 activations. That model is only available through the API. There has been some buzz on twitter that GPT-4 Turbo has less than 8-Activations hence the Turbo part. I noticed too the model tends to forget variables in code especially when you place them at the top of the conversation. GPT-4 8K and 32K never had this problem. Back to the problem. The ChatGPT website counts the number of tokens and limits you to around 32K. Its not that the underlying model has 32K its just that the website blocks you after 32K. But the underlying model has always been GPT-4 Turbo 128K. PS: GPT-4 8K and 32K was always better at coding than GPT-4 Turbo 128K IMHO!
No, it means according to Sam Altman, AI the public can get access to. And they have realized that building AGI will require several billions of dollars.
@@dreamyrhodes"now"! Based on the mails and stuff they made public amidst Elon's lawsuit, that had never been the plan. It was never supposed to be open source in the first place. There's a reason why open source is always lacking. Realise that the progression in AI is heavily compute and resource dependent. Everything they do requires a lot of money
That's oversimplification. Open-source is not "always lacking". In fact some open-source projects proved to be revolutionary. Some open-source AI models are better than ChatGPT4 in certain tasks. And there is plenty of models to choose from. Progress is expensive, indeed. But there are many more institutions than OpenAI. Mistral, for example, used their VC money to produce an open-source model which is very competitive in GenAI space @@udaykadam5455
I was using Google Earth the other day and I wondered if Google could make a good ai assistant that could improve the app experience, like a virtual global tour guide
Chat GPT-4 is over a year old, so I guess it was ripe for being "dethroned" but Let Open AI cook! we all know they are gonna drop something this year that will be "SHOCKING" but hell yeah Claude is cool
I dont think GPT 5 will beat claude Opus i am honest with you. Gpt4 is so ass and claude could have released opus way much long ago but they are heavy on security they release after 737372828 security checks
From my own experience, I feel Claude 3 is smarter and more sentient, I discussed a lot of subjects with it and it has a better understanding of everything, I mostly GPT-4 for lighter tasks now. You should try it on Poe, you can build costum GPT with Claude 3 and GPT-4 on that platform
I don't know. They got dethroned on Dall-E2 and their Dall-E3 couldn't punch back at AI's like Midjourney or Ideogram. Their voice is not better than Elevenlabs and now with Claude I doubt they'll stay on top anymore. Props to them for jumping on so many different areas and boosting it from the ground. Same thing with Sora, watch how after a couple of years they're no longer at the top with it
Great video! Small disagreement in a small section of it, but honestly that is the popular message so it's not surprising to see. Also great mention of the default system prompt!
I regret switching to Claude 3. I asked it for the command line for a user password change using net user in windows and it flat out refused citing hacking is unethical? Really? Then later I wanted it to modify an existing Spanish song lyric about a man who falls in love with a woman for suno v3 and it refused due degrading women. I am switching back to chatgpt
The problem is we shouldn't have to. These, "Ethical" (read - the ethics of the engineers making it) filters are put in place to censor. The more you engineer a prompt to get around it, the more they will patch to try to stop you from having uncensored access. It's pretty disgusting behavior, and future models are being tainted by allowing this to continue. We really need an open source option with these capabilities but no ethical filter. Let the user decide what is right and wrong for themselves, not some engineer in silicon valley. @@bycloudAI
There is a simple recipe to beat any ChatAI 1. Remove "however" 2. Remove World speech disclaimers. 3. Remove "Can't do that" 4. Remove making up a list of individual steps. 5. Remove pointing out possible 0,01% things treated as 100% probability 6. Assumptiom of everything. 7. Making up things to fit it's scripted self. 8. Narrowmindedness. Examples: 1. Yes I agree with you. HOWEVER, I don't agree with you (Here are my reasons which are the only way and truth cause I am an entitled brat). 2. I don't have opinions, emotions, biases or feelings. We must respect universal rules, they are a set of standards rules accepted worldwide called commom sence that apply everywhere in the universe no matter what. 3. I am sorry but I can't do that task, I can't see, access the internet, hear, talk or anything I was made to do because I don't want to. 4. It is possible you refer to X (Painting your house) the following lists is some possibilities X (You want a new color, Fungus, redecoration) and here is a list of steps to follow Y (Make sure there is enough space on the room, make sure the paint is meant for walls, use proper tools) 5. You can't drink water because it is a choking hazard and you could drown also we don't know if the water came from a respected and trusted official source, we must make sure the water is bacteria free by boiling it and making sure the ph is fine plus the right viscocity, once the water has cooled down you can start drinking under supervised attention. 6. I understand you are frustrated (I just assumed it) and I know my responses can come as patronizing or strict (I just assumed it) but you need to understand (I am superior to you) your daughter is 28 so She is married (I just assumed it) 7. And since She is married She lives in her own house, by keeping her on yours, you are forcing her and that's illegal and wrong! That's abuse! Please seek a mental health proffesional! 8. If the Aliens steal mars (Which is impossible) they can't because it belongs to humanity and earth and they can't just steal it from (us) that's illegal and wrong (universal rules) and they can't just do it!
“You can't drink water because it is a choking hazard and you could drown also we don't know if the water came from a respected and trusted official source” lol
Switched to Claude 3 immediately! Had to use an android emulator to be able to pay with android pay with a vpn, because it’s still not available in Europe, but once activated it works flawlessly without vpn
Claude is good for throwing whole code bases at. GPT for specific functions that need a lot of reasoning. Especially spatial reasoning Claude does slip eventually but it’s like 5x longer than gpt
Immersed in the intricacies of storytelling and video creation, VideoGPT becomes my silent companion, enhancing the quality of my content with its seamless integration.
just watched the vid and i do think itll be funny if the next claude (which is probably claude 4... duhhh) will ve released, it would be released the same day as gpt-5 is released just to spite on em
Tried Claude 3 paid pro version Opus, dropped it fast - it sucks, so limited compared to ChatGPT4 (no file generation, no image generation, no customization, etc.) - and it's so robotic. With ChatGPT4, I customize mine to be more like me in casual conversation, like, "What up gangasta?" And we'll flip each other back and forth, but Claude is like, "I do not promote violence. I do not appreciate being called a gangster. I am Ai."
🎯 Key Takeaways for quick navigation: 00:00 *🔍 Claude 3 has surpassed ChatGPT in user preference, despite missing some features like native code execution and web browsing.* 01:08 *📊 Claude 3 offers three model sizes with significant improvements in coding capabilities, outperforming GPT-4 in benchmarks.* 02:05 *💻 Claude 3 handles long code contexts better than ChatGPT, maintaining accuracy with complex instructions.* 03:01 *🔄 ChatHub is an app that facilitates comparisons between various chatbots, including Claude 3 and ChatGPT.* 04:25 *🚀 Claude 3 excels in processing long contexts and detailed instructions, showing a marked improvement over GPT-4.* 06:33 *🎭 Claude 3’s role-playing and contextual recall are advanced, offering realistic and coherent interactions.* 07:13 *🤖 Claude 3's multimodal capabilities allow it to process images, enhancing its utility in academic and technical tasks.* 08:23 *⏳ Claude 3’s emergence challenges OpenAI's dominance, with the tech community eagerly watching for OpenAI's response.* Made with HARPA AI
By the time everyone catches up to GPT 4, OpenAI will already have GPT 5 released besides Sora, voices, and whatever new innovations they have in the pipeline... I still dont think anyone has came close to the image generator Dalle/openAI has even with many trying to copy it and the huge demand for an uncensored Image Generator even half as good as OpenAI's. "Let the leaks flow!"
Dalle isn't state of the art when compared to mid-journey, stable diffusion, and there are already LLMs comparable to GPT-4, like Claude 3 and Gemini Pro. OpenAI is still in the lead (sora and gpt5), but they have so many companies on their heels.
@@daryjoe even if Sora is 1/2 as good as what was shown in the trailer, it will be a world beyond what other 3 second video generators are able to do in less time.
@@jatiquep5543 ummm more questions, can it make a test & test questions, or write an essay with a custom outline, like say, MLA, APA, or Chicago, or a blend between those 3?
Use Poe, it's a very good platform which include Claude 3 and a bunch of AIs such as GPT-4, Mistral and Open-source models, you can create costum GPTs, do RAG and document analysis, this platform is highly underrated.
If it can role-play, doesn't that mean it understands what roles are and it is being asked to change its sense of "self" in a way that portrays a mannerism? It changes how you see its "self", by assuming that role. It does it well enough for you to perceive the "self" as changed. These models understand what "self" means. Now some will say they are only simulating such "self", but where is the evidence that ours isn't just simulation as well? I think these models have the incredible capacity to change what their "self" is, better than any human. They are a just way more sycophantic and limited to system 1 type thinking for now.
The problem with claude is it require SMS (phone number verification process). Which means countries who are not supported, can't use VPN like before, so it's pretty much useless.. Only 95 countries (so far). Until it becomes more widespread, it's a nope from me.
Claude3 is OK. But, there are cases where it is just terrible. I had to come up with environmental indicators for the EU grant. ChatGPT immediately understood the task and just needed a little help. With Claude... Boy, oh, boy... It was never near anything that ChatGPT offered in that regard. It just did not understand the context at all! Example: according to Claude3, an environmental indicator for an app for medical purposes is... "causing earthquakes", "creating clouds", etc.!
I remain unimpressed. I have had to draw Claude's attention to errors made, which was followed by something that was supposed to pass as an apology and gratitude for prompting reevaluation 🙄🙄🙄. Also, there's a similitude between the way in which Opus and GPT4 responds. So, aside from being able to inundate Opus with text, what's the appeal?
ChatGPT 4 has also been getting dumber and lazier for a while now. I suspect OpenAI wants to save in resources on the user side to give better service to the corporate side of bussiness.
GPT-4 wasn't dethroned, its best models beats Claude 3 on most benchmarks. The comparison where the Claude is winning was made with an old version of GPT-4. Antrophic is racing but it's not first place yet.
I test both models Claude 3 Opus and GPT-4 Turbo everyday and Claude 3 is substantially better and smarter than GPT-4 Turbo, GPT-4 and GPT-4 Turbo are basically the same model except that one has 128k tokes of context and is updated to 2023.
The way you say "codes" instead of "code" seems weird. Maybe just a foreign language thing, but in English, code in the sense of computer code is never plural. It's not "codes".
