The Tech that’s *probably* inside GPT-5 just got Open Sourced!

LLMs are about to get STRONG

▼ Link(s) From Today’s Video:

-------------------------------------------------

▼ Extra Links of Interest:

-------------------------------------------------

Thanks for watching Matt Video Productions! I make all sorts of videos here on Youtube! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!

All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.

-------------------------------------------------

[DisturbedNeo]

I asked GPT-4 to write a system prompt to turn my local 7B LLM into an "Expert Creative Writer", and it works ridiculously well

[I-Dophler]

1. GPT-5's underlying technology has likely been open-sourced.
2. The announcement indicates a shift towards greater transparency in AI development.
3. This move could accelerate innovation and collaboration in the tech community.

[TimTruth]

Windows key + 'h' = voice to text on windows. I just found this out

[AH]

What you are talking about is so close to the breaking point concept about asking an AI do create the next generation AI.
And how absurdly fast the evolution would explode then. Especially if they can gather their own material to build themself.

[Jacobinks]

It’s crazy how AI is ACTUALLY real now. It’s not science fiction anymore. Mental.

[Windswept]

I did this with my custom instructions when it first dropped.

To me it makes sense (and works) to reverse engineer human psychology and use multiple cheap optimised agents to create a specialised network/ego/mind that has the power and efficiency of human cognition.

[TheFeedRocket]

Same as having a person in your business that's really good at his job, he's paid $$$, you hire a new employee that doesn't have the skills to do everything without help, he's paid $, He learns from the other employee exactly what you need. Now you got an employee who can do what the original employee does...but he is cheaper and faster. Fortunately with Claude, the cheaper model doesn't ask for a raise! Also as in real life, the larger more educated model or person will certainly add value by knowing more and solving any issues the other might not, but if all you need it for is one particular thing and it's not something to have many variables then perfect. It shows these models have room to grow.

[erikjohnson]

Wow, you place your video links at the top of the description! Thanks for caring about your viewers enough to prioritize for usefulness. Good enough to earn a subscription (I've seen your videos before but never subscribed until this moment).

[sdhpCH]

I love the lighting you used in this video, took it up a notch.

[skylineuk]

I think it’s important to understand pain and other “human” things are basically feedback loops so saying pain makes us human is inaccurate as it just a system to prevent damage that we have learnt to dislike. Pain and emotions basically work like our body's own feedback systems, reacting to stuff that happens both inside and outside of us to keep everything in check. When we feel pain, it's our body's way of saying, "Hey, something's wrong here, " prompting us to react and hopefully fix whatever's causing it. Emotions are similar; they get sparked off by things happening around us or thoughts we have, and our brain decides how we feel about it, leading us to act in a certain way. This whole process is super complex, with lots of moving parts influenced by our past experiences, how we think, and our social lives. It's like our body's natural way of helping us navigate life, keeping us safe, guiding our actions, and helping us connect with others and computer systems can do this do and in a real sense act the same way.

[electromigue]

I really enjoy your vids, Matt. Great job as always.

[bobparker]

With the semi-recent release of Hugging Face's Common Corpus (a 500 billion word public domain dataset), I think we'll start to see more large open source models that integrate these new techniques and give performance maybe rivalling that of GPT-3.5 or GPT-4.

I'm also curious if these techniques can improve certain aspects of more "linear" scaling models, like Mamba, or mixture models like Griffin.

[jpviperk]

They are leaving us so many breadcrumbs. We can pick up enough for a loaf of bread then.

[ccrtelevision]

10:39 My thoughts on AI consciousness: We have no understanding of consciousness outside our human perspective, for instance our understanding on consciousness in other species is based on observation and not first-hand experience, so we can never know for sure. Even when we reach the point where AI can act convincingly human and make us believe it is conscious, it's going to remain a technological mystery unless a way to quantify it is discovered :]

[scottwatschke]

With voice access, you can wake it up with the word unmute. And you can turn it off with the word mute. That's the new voice command for Windows 11 pro.

[oscarbertel]

Any implementation of a reward function on a learning model consists of somehow introducing positive and negative stimuli, which could be very analogous to what we define as pain and happiness, and that is precisely why the model manages to use that reward function to learn. In a way that is the basis of reinforcement learning. then:
Quiet-STaR == the beginning of Conciusness is a really realistic analisis

[TheAprone]

Something to consider is that we don't feel pain from our physical body either. We get info from our body and our brain interprets it as pain. You don't need a body for pain. To quote Morpheus, it's all just electrical signals interpreted by your brain.

[Laura]

In using Claude Sonnet and asking questions, talking about various free form ideas it feels like it comes down to sensation, like artificial skin, a body of sorts, a long term memory for subjective experiences, and interaction with humans for applied knowledge. Sounds just like us little brains humans but exceptionally more knowledgable 😅 You touched on so many things I felt were happening. I didn’t know about Quiet star though. Pretty brilliant. Thanks for the clarity.

[tiagotiagot]

11:00 There are humans that can feel pain (turn's out not good thing, you don't notice you broke a bone or burned your hand, or bit thru your tongue while eating etc, you don't develop habits to avoid risking getting injured, and then things get infected or heal badly because they didn't get treated in time etc)

[MrTk]

The future is very bright Matt! Another great EP. Thank you 🔥🤟🔥