This GPT-5 NEWS Could Change EVERYTHING...

Показать описание

00:00 - Introduction: GPT-5 Rumors
00:23 - Rumor Explanation: Internal Model
01:08 - Opus 3.5: Missing Model
03:16 - Distillation Process: Smaller Models
08:19 - Model Performance: Improved Efficiency
11:17 - Cost Challenges: Expense Issues
13:49 - Distillation Benefits: Enhanced Performance
16:11 - GPT-5 Speculation: AGI Potential
20:29 - AI Future: Internal Focus
23:00 - Closing Thoughts: Future Implications

Links From Todays Video:

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

Music Used

LEMMiNO - Cipher
CC BY-SA 4.0
LEMMiNO - Encounters

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

TheAIGRID

Рекомендации по теме

Комментарии

My favourite use of AI is taking the transcripts from your videos to get a summary :

OpenAI may have already built GPT-5 but keeps it internal, using it to generate synthetic training data and improve smaller public models through a process called distillation
Anthropic followed a similar pattern with Claude Opus 3.5, using it internally to improve Claude Sonic 3.6 rather than releasing it publicly due to high inference costs
The future of AI development may shift away from releasing increasingly powerful models publicly, instead using advanced internal models to distill capabilities into smaller, more efficient public versions
Major AI labs are facing diminishing returns from traditional training methods and have exhausted high-quality data sources, making distillation a crucial technique for future improvements
OpenAI may be transitioning away from needing public user data or focusing on commercial releases, instead prioritizing internal development toward AGI/ASI using their most advanced models

BrokenOpalVideos

I mean, while people criticize how long the videos are, it's honestly not that bad. This is my go-to channel when I fall asleep. You let it play, and then you go to sleep.

CommentGuard

I am still waiting for an agentic ai model that have a long term memory and can plan and run tasks without any interference. I assume this can be done even with 64B models. We don't need super intelligence for most daily tasks just need long term thinking and execution. A local ai trained to use most popular PC programs/tools would be great as well.

ccdj

Pro English tip: adverbs of probability have DIFFERENT meanings. For example, "definitely" means "there's a 100% chance this will occur, " whereas "possibly" means "the chance is less than 50%." In other words, "definitely will possibly do this in the future" means NOTHING. It's like Schroedinger's prediction. It'll happen and not happen simultaneously.

daimonmagus

"OpenAI may be transitioning away from needing public user data or focusing on commercial releases, instead prioritizing internal development toward AGI/ASI using their most advanced models." In other words, OpenAI is disappearing into the military industrial complex, and as the average shlub, you won't even get to see the crown jewels, which are kept behind closed doors for use by the transnational oligarchy that owns this civilization to cement their hold on power forever.

JohnSmithAB

You know 26 minutes is a long time for someone to commit to watching a video. Don't you think you could have done this video in like half the time?

throker

according to this channel, the world has changed 24623476 times in the last 3 days, but weirdly enough, everything is still the same

huehuehuehuehuehuehuehuehuehue

Jesus....this was nonsensical. We need AI to just distill this verbal vomit into a much shorter, cohesive video.

briankgarland

Additional context: Ilya Sutskever's claiming that a straight shot to ArtificiaI Superintelligence is feasible and then raising "only" $1 Billion USD to pursue that. A straight shot suggests no intermediate product (so no revenue) and no benchmarks to hit to unlock future rounds of funding.

entreprenerd

its really improbable how often everything changes around here.

Novascrub

I've already wondered what if AI tells them something insane about the universe and they start hoarding a bunch if important information

timelapsega

GPT-5 has been done for over a year now.

clintfaber

Once r-Star Math is integrated into flagship models I think we'll see them running on smaller traditional data centers rather than these enormous power hungry monsters we're using today. We could probably see very good thinking models that could run on your home PC.

BruceWayne

Altman himself said scaling the models further in size doesn't bring much more improvement and the costs are enormous.

drwhitewash

Have we forgotten about Orion? I thought this was all obvious. Were they not distilling versions from Orion? They pretty much said this a few months back.

rolestream