Might not come in the first version - feels a bit complicated to implement. Even the basic iteration takes a long time to polish for some reason but let's see 👀

@@Testingcatalog Shouldn't be complicated at all. They are already having to partition ChatGPT's memory structure on a per user level, you can use that same mechanism to provide partitioning within a given user, assuming you didn't want to create another abstraction layer (which is itself reasonably trivial).

It is also expected that it will be available to custom GPTs to use, will be a big deal

Yeah ai was completely useless without it since every prompt it would forget everything

Not a misuse at all, it's actually something the community would pass around. Different instruction sets for different tasks. GPT Agents were born of people hating having to change between custom instructions. It's all part of the process. GPTS do ignore your custom instructions from your account settings and only use the instructions from their agent configuration section. I actually surprised myself the other day when I accidentally used the base GPT4 model, which HAS my last used custom instructions. That was months ago, so when I used GPT4 Main, I flipped out slightly when the thing remembered its name and to talk in third person. For a bit, I was all "Daaaaamn... they really do use my data for training..." But I'm also an idiot, so...

@@MrLargonaut Haha, I like the bit about forgetting it still had your old custom instructions in it and thinking that ChatGPT had taken another quantum leap: it’s so totally something I’d do 😂 Thanks for the explanation about GPTs ignoring your custom instructions: I’m going to set up a few for specific use cases.

nah just AI tools lmao

Short Circuit/Bicentennial man, not Terminator. I'll also accept I, Robot as long as the outcome is the same.

@@MrLargonaut I loved short circuit. Fun movie!

Well its reasoning is indeed very limited. Its capabilities though are huge

ChatGPT doesn't reason much at all. It boils down to a probability model with varience. It's "Most likely follow up word/sentence/paragraph to the previous one." Based on it's training data. It doesn't really know or figure out anything.

@@darkfire1408 if it were only a neural network working with weights alone, sure, but you leave out transformer layers and further pre-response processing that goes way beyond ‘guessing the next word/letter/sentence’. Even if what GPTs do is still outside the definition of ‘reasoning’, you’re still leaving off a substantial part of the story by parroting ‘it’s just mathing’

It seems like it could be both or either to me. I'd imagine you could also build a custom memory by selecting specific things and compiling them.

It probably is univeral not just across all your conversations in your account but also across every conversation. THis is allow them to segment users better and see just how everyone is using chatgpt and then they can make the product even better. The reason I say this is because of the temporary chat options that he showed in this video. It seems like temporary chat will not save your data and also use it for model training. so I assume that the normal chat does use your data :)

In the current implementation it is a single memory while all custom GPTs may be able to access it (I assume there will be an explicit confirmation for that)

THANK YOU!

Here is the dose of dopamine you need 👍

Same I've had a few tasks recently just get either ignored or suggested for how I can do them, and reiterating the question gets solved so it's not like it didn't have the computational power or understanding on the task. I think the GPT's are starting to be conditioned in any way possible to make the next version look / sell better

I have also noticed that. This is not my first time buying GPT-4. In total, I have bought ChatGPT plus over 5 times. Some even on a row. And I have noticed: 1) When you ask it to write code, it has started to end code early. Even when I ask it not to, it still does. 2) OpenAI seems to have added even stricter security measures to chatGPT. It's getting dumber and dumber. 3) ChatGPT has become VERY eager to help me. When I first bought plus, I could tell it something without it being like "I can't assist you, as I can't confirm that."

There is no timeline yet unfortunatelly and It is not yet available in alpha testing either. But it has been in production testing for more then a month already. It is probably the biggest feature OpenAI is working on atm (except new models)

Why are you putting that in quotes? You don't seem to believe this update is real.

@@BESTGAMER12132 because memory capabilities would be a massive jump in the way conversations can be had , it's not just a new feature ' it would be fundamental change so far .

Thanks for the info. Can you give me your own quick explanation of what "fully differentiable memory" is?

@@jaymethodus3421 check out the "cache transformer", it uses a gated recurrent cache to update self-attention with basically current and historical token information feel me? You can extend it out though using dilated segments(dilated self-attention) to create a highly compressed, yet expressive memory, self-note tokens when used properly, via robust synthetic data situations, can greatly enhance your model...you know, since we need to constantly remind ourselves of salient information when engaged in long-horizon task. Essential for extremely long-token sequences or long-form chatbots. Free cloth talk.

Not yet unfortunally, it is still only available in internal testing

I support that. 3.5 is great. But 4 is much better. It's slower in text generation. But I think it's ability to think and be creative is better than 3.5. How do you use 3.5 if you don't mind talking about it? 4 also has 32k context window. And you can post longer messages in the text field. I think 3.5 still only has an 8k window.

Mixtral 8x7B is free on Huggingchat and better than 3.5

@@static825Honestly, you’re missing out. I use 3.5 for very simple requests, but it’s much more likely to hallucinate, and the answers I get with 4 are more complete and I can probe much more deeply. I still have to be careful of hallucinations in some areas, but it’s overall much more useful for me. YMMV, but for me, it’s the best $20/month I’ve ever spent :-)

Do you have access to the specialized gpt’s? They are essentials.

Not sure about 2 years but you definitely don't need gpt 4 atm.

It will be nice to get past the days of creating PDFs of the conversations and then re-uploading it on the next session to get it up to speed.

That's literally what memory is. It allows chatGPT to remember stuff from previous chats.

@@BESTGAMER12132 You know very well it's not real memory: The model cannot use it in any way.

I've done this before but it wasn't very helpful. The summary was top level overall discussion topic with basically bullet points but without any details which meant it could apply to anything. Hopefully that changes soon, but you can already do it.

That's right!

@@Gerlaffy Keep your woke ideas to yourself.

@@loot6 lol

You can't influence the base model within your instance of usage..

Why don't they speak of synthetic data and in that agent robotics models of learning

I purchased the plus plan for over 5 times. And I can see a clear downfall too.

gpt 4 has real time data.

Google translate exists. Except it would also translate stuff that is "Inappropriate."

Well this memory feature will help you to get you work done more effectively.

@@zorororonoa2469 polluting my context space with information from other chats is not going to improve my productivity. Imagine that it wrote a new script for you and it apparently attempted to call a throw-away function it wrote in a script two months ago, or a function from a different project that you copy/pasted to ask it a question about it - it would result in a compiler error. I would rather maintain the knowledge-space manually and give it only the information it needs to complete the current task. GPT-5 might be able to manage this complexity effectively, but GPT-4 definitely won’t.

Yeah, a bit of a tradeoff there. They do already have access to all my conversations though, one way or another. I feel like I’m ok (so far) with OpenAI knowing things about me that I’d run screaming if Google had them. I also don’t discuss things with it that I wouldn’t mind people knowing though. For anything I’d consider really sensitive, I’d use an anonymous session in Brave browser and Brave search.

I mean the Canadian, U.S., and Chinese governments, as well as probably a dozen corporations, already know all my deepest darkest fetishes, desires, and wants. Why would I care about an AI collecting data on me? Let the government puke over my choice of porn lol.

They are already collecting all of your chats. How do you think the chats persist across devices?

@@GearForTheYear Yeah, but it’s a question of whether they tap them for training or just present them in your UI and they only get used when you restart a conversation. (Interesting though, that YT seems to have deleted my comment above that cast aspersions on Google. I’m a bit surprised that they do that so overtly.)

The main innovation of this century hasn't even happened yet, it's way too early in the century. Not to mention, the pace of technological advancement itself has been advancing. I was saying we'd have technology like current AI by the 2030's all the way back in the early 2000's. I was called insane by most and guess what? It's not even 2030 and it's already happened.

yeah i dont think you understand how fast this snowballs

Do you have a gpt 4 subscription?

everyone has access with a subscription.

Bang on. One million percent. Want to listen for the great news but really does need to tighten up.

Ah, dear viewer, your comment, though succinct, has unfurled within me a veritable cascade of contemplation and response, the likes of which I fear may well exceed the bounds of what you might categorize as conciseness, yet herein lies the crux of our burgeoning dialogue. Let us embark upon this odyssey of explication with the recognition that the vast expanse of the internet, much like the boundless realms of our own imaginations, offers a cornucopia of content, each piece a unique reflection of its creator's essence, aspirations, and, indeed, their modus operandi for engaging with the world. In the grand tapestry of digital discourse, where every thread is woven with the individual yarns of its contributors, the manner in which one elects to express oneself-be it through the succinctness of brevity or the labyrinthine meanderings of verbosity-constitutes a fundamental aspect of one's digital autonomy. Thus, when I elect to navigate the sinuous pathways of prolixity, adorning my narratives with the rich tapestries of an extensive lexicon and the intricate embellishments of complex syntax, it is not merely a stylistic choice but a declaration of my sovereign right to articulate my thoughts in a manner that resonates with the depths from which they emerge. Furthermore, this choice mirrors the very essence of freedom that underpins our shared digital ecosystems. Just as a painter decides with deliberation the strokes that will bring their canvas to life, so too do I deliberate on the cadence and complexity of my verbal expressions. This autonomy, this unassailable right to self-expression, stands as a bastion against the homogenization of thought and creativity. In a realm where the click of a button ushers us into new domains of knowledge and entertainment, the prerogative to forge one's path remains untarnished. If the style or substance of one's offerings does not align with an individual's preferences, the digital landscape abounds with alternative vistas to explore. To the point of your advisement regarding the efficiency with which I should arrive at the nucleus of my discourse, I offer this reflection: the journey, with all its meandering and exploratory detours, is oftentimes as enriching as the destination itself. The value derived from any piece of content is as much a function of the recipient's perspective as it is of the creator's intent. In this exchange, the multiplicity of interpretations and the diversity of preferences underscore the richness of our collective experience. As to the matter of opinions, those ephemeral wisps that flutter through the ether, shaping and being shaped by our interactions, it is indeed a marvel that we can share and scrutinize them so freely. Your perspective, while noted, serves as but a single hue in the vast spectrum of viewpoints that color our world. In the grand scheme, each opinion, including my own, occupies its niche, neither more nor less valid than the next, yet all contributing to the vibrant mosaic of human thought. In closing, I extend my gratitude for your engagement, for it is through such exchanges that we refine our understanding and appreciation of the myriad ways in which we can connect and communicate. I shall continue to craft my content as I see fit, reveling in the liberties that allow for such creative expression, and I encourage you, and all who share in this digital voyage, to seek out or create the spaces that resonate with your own unique sensibilities. Thus, we find ourselves at the terminus of this particular intellectual excursion, having traversed the expansive landscapes of expression, autonomy, and the intrinsic value of diverse perspectives. May our paths cross again in the vast, interconnected expanse of dialogue and discovery that is the hallmark of our era.

Really? This shit is amazing!

potentially losing your job is hyping you, damn thats what we call a sheep.

If all you do is look, then all you can do is see. To experience, you have to do.

@dan-cj1rr I don't think having every job being automated is a bad thing, if we're able to go about it right. If we're somehow able to smoothly transition from relying mostly on human labor to having every job being fully automated, then there's a possibility that we'll be able to live in a world where we can pursue things that we truly find meaningful, instead of just working a soulless job just to get by and make ends meet. Butt, if the transition from Relying on human labor to fully automating everything isn't done right, then there's a possibility that we'll probably end up with Millions of unemployed, very angry and unhappy people who just lost their source of income and can't make ends meet. You can only imagine what that would lead to. We're in the crossroads between utopia and dystopia here my friend. I'mma just stay optimistic instead of worrying too much about what's to come haha

I guarantee in two more years you won't be saying that, if the rate of improvement is 5% per month then it's going to be exponential growth, and in 4 years we'll have Blockbuster films produced by this AI, I would not be surprised if we have intelligence surpassing human intelligence by 2028, certainly with the next 10 years it'll become smarter than Einstein, we are just on the trajectory of exponential..... And then we go hyperbolic....new gods born

Speed them up using the gear icon in the upper right.

I usually listen to youtube vids at 1.5x speed. Saved years of life that way

Transcript of video -> ChatGPT -> "Restate this eliminating redundancy." -> listen.

Under 15 minutes isn't bad.

This or you have an attention span of a goldfish.