GPT-5 is coming: 3 ways to prepare for a 100x improvement in SOTA LLMs

Are You Ready for a 100x Boost in AI Performance? Get Ready for GPT-5 and whatever comes next!

If you don't have a plan, how will you know you're succeeding? In this video, we create a plan and discuss tactics for preparing for a 100x improvement in SOTA LLMs, state of the art large language models.

🚀 Let's discuss the Large Language Models (LLMs) and the mind-blowing potential of a 100x improvement in AI capabilities. Inspired by insights from Sam Altman, this video unpacks the game-changing advancements in models like GPT-4, Claude Opus, and Gemini Pro. Discover how the future might look with GPT-5 or even beyond!

🔥 In today's tech-driven world, prompt engineering, prompt chains, and context-filled prompts are not just jargon—they are the backbone of high-performance AI workflows. Whether you're a software developer, a tech enthusiast, or an AI pioneer, understanding the power of 'Big Ass Prompts' (BAPs) and the significance of a large context window can position you at the forefront of innovation.

🛠️ We're not just theorizing; we're applying! Watch as we demonstrate practical strategies to leverage your current tools to their maximum potential, preparing you for the upcoming 100x leap in efficiency and effectiveness. From expanding your problem set with ingenious prompt chains to mastering one-shot and few-shot prompts, this video equips you with the tools to excel in the age of advanced AI.

🌟 Hit the like and subscribe for more insights on how you can transform your engagement with AI technology, making your work in programming, knowledge work, and beyond more impactful than ever. Stay ahead of the curve, prepare for the future with us, and turn these insights into actions that catapult your skills and solutions into a new era of AI prowess.

💡 Remember, the key to mastering the future of AI is not just about understanding the technology but being ready to implement and scale it effectively. Join us as we pave the way to a smarter, more efficient world powered by next-generation LLMs like GPT-5 and beyond.

Subscribe now and transform your approach to AI with every video—let's innovate together!

🎥 Featured Media
20VC Sam Altman & Brad Lightcap

📖 Chapters
00:00 The 100x LLM is coming
01:30 A 100x on opus and gpt4 is insane
01:57 Sam Altman's winning startup strategy
03:16 BAPs, Expand your problem set, 100 P/D
03:35 BAPs
06:35 Expand your problem set
08:45 The prompt is the new fundamental unit of programming
10:40 100 P/D
14:00 Recap 3 ways to prepare for 100x SOTA LLM

#promptengineer #gpt5 #ai

[vickmackey]

Any AI that is truly a 100x improvement over current SOTA LLMs won't require all this prompt expertise. It will simply understand what we likely want, and if it has doubts, it will simply ask--you know, like a smart engineer would.

[ankeethsuvarna]

"The PROMPT is the new fundamental unit of programming & knowledge work".

Beautiful!

[planetmuskvlog]

Talk to it like you would a competent human. Create custom GPT’s for all the conversations you feel become repetitive. Task the LLM as you would a competent human. Give it all the context and training necessary and then you have a custom trained agent who already knows what to do. Rinse repeat. When the 100x models come online, you will have a team of agents and tools to use together to solve even larger problems and create more complicated workflows.

[JavierRojas-jq]

"I snort LLMs like it is crack" 💀

[tomaszzielinski]

Stay at the cutting edge of current LLM solutions. Prepare difficult problems which current models can't quite solve yet. Have them ready to run and waiting, and the day GPT 5 drops, just launch them and watch the magic.

[justinwescott]

Your 100x circle is more like a 5000x circle. Looks like you multiplied the radius instead of the volume.

[kubasmide]

I agree, some libraries are removing the abstraction of prompting and it’s not really helping us learn how to get better at prompting.

[kyleabent]

Great video I just added OpenAI Vector Storage API Assistant in my pygame rhythm game project, I plugin my guitar and generate logs of my playthrough and now I can chat about the data

[qiqqaqwerty]

Excellent video, very good advises are given!

Thanks, keep the good fight!

[yiouyou]

What do you think of DSPy, which is closing the loop of the prompt engineering cycle, transforming what is often a manual, handcrafted process into a structured, well-defined machine learning workflow?

[techpiller]

I think this is solid advice! Prompt is the new king in thought work, get the prompt reps in, start preparing for large-context workflows.

However, I think code along models, such as RAG and agent frameworks are be here to stay. But the advice to start getting used to coding with prompts is solid, because AI-assisted programming is the future. We just need the AI coding tools space to fight it out, to get to the king feature set!

[brookster]

New listener instant subscriber you just nailed everything my friend. Love your style articulate knowledgeable and very sensible advice probably the most sensible I’ve heard regarding long-term strategy on how to develop for the AI generation applications in agents

[thunkin-ai]

Great video.

I'm still waiting for Mr Hat to present :)

I had a version of the BAP principle in that "humans will be predominately concerned with providing 'inputs', whilst, we trust the increasing capability of LLMs to produce the 'outputs'"

[Bmutch]

Just came across your channel today. It’s very good. Thank you!

[lancemarchetti]

It doesn't matter which model it is, we still have a problem trying to get these LLMs to stop falling off the bus when it comes to simple string manipulation such as character reversal, capitalization, incrementation etc. They can write the code that carries out these manipulations. But the models themselves cannot carry it out in a simple chat interface. They quickly lose track after the 3rd prompt.

Until we can fix that, AGI is only a dream.

[fkxfkx]

You realize that the 100x will also do prompting better than we will be able to do and will develop automatic programmatic hyperoptimized prompts as well?

[elsavelaz]

I completely agree with this; I’ve been saying the same for the last while

[Leonid.Shamis]

@indydevdan If you don't use tools like LangGraph, LangChain, AutoGen, CrewAI, Phidata, and other AI Libraries, then what do you use? Transformers, etc.? Would you consider making a video to explain how you develop your software with LLMs at this stage without giving control over Prompts to AI Libraries? Thank you.

[RobertElliotPahel-Short]

I’d very much enjoy talking with you 1-1. I’ve been working constantly with llms to develop software and am starting to get to the point where it can generate/run its own functionality towards arbitrary goals

[jonesani]

I dont see the relevance. With an 100x AI you will be able to talk to, as if you talk another human being and the AI will most likely get you even better, than another human would and btw. writing will get entirely replaced by speaking.