How does OpenAI just keep winning? Here's their secret behind Deep Research...

All my links: linktr.ee/dave...

[@derghiarrinde]

God, I missed these videos...

[@LDdrums20]

Same!

[@kaio0777]

ditto!

[@smoochie3331]

First time ge praise openAI.

[@brunodangelo1146]

Cognitive bandwidth IS a scarce resource. Anyone who denies that in themselves doesn't really know what operating at peak performance means. Cognitive offload is the biggest reason to use AI in your daily life. It's like the red arrow in a compass.

[@kayakMike1000]

You let AI decide what? What do you cognitively off load.

[@brunodangelo1146]

@kayakMike1000 Everything I can. I mostly don't "let it decide", in fact I decide over the results it returns. AI can do A LOT of things for you. It's worth it to try to offload everything you can. Even for small tasks (especially actually), they end up adding to a lot of cognitive work over time.

[@sjcsscjios4112]

I agree. Debugging specially

[@t-lowmusic7697]

Doesn’t cognitive offloading decrease cognitive ability over time though? I’m trying to find the balance on what to offload to maintain intelligence

[@brunodangelo1146]

@t-lowmusic7697 offload what you can, and what you want. Focus your cognitive work on what you either can't offload or not want.

[@Ton369]

because Sam said they have "adult mode" coming this year.

[@quantumspark343]

Did he?

[@topical_stories_ru]

Saying something is Sam's full time job

[@aciidbraiin8079]

No way, I believe it when I see it. Adult mode for me is unlimited ”corn”, slurs, dystopian novels, deep fakes of celebrities, AI personalities being ”beachy” and seductive et cetera. I think Grok or some company we haven’t yet heard of will be the first doing this, and it will be awesome.

[@DaveShap]

@quantumspark343 he agreed that it would be a good idea

[@minimal3734]

@@QeqeMOUBELE Men of culture.

[@pascalharbers390]

Love that you are back Bro 👊 Do your pace - with you since early 23

[@RENCHER]

Excellent video. Gives me a term to describe how I use AI. I also think OpenAI's strategy of shipping incremental upgrades often has proven to be the right call. They were able to respond to Deepseek very quickly, rather than get into a GTA6 situation where everyone is waiting for a monster drop.

[@Axiomatic75]

"Human cognition is remarkably energy intensive" I see it the exact opposite, I marvel at its efficiency

[@JohnSmith762A11B]

I think Dave meant relative to body size/weight. The brain really is an energy hog in those terms.

[@concernedindian144]

thats to be marveled at, but i heard from a neuroscientist and its fascinating that a brain’s primary role is to help you survive and not help you to be productive or “successful”. So maybe openai releases follow the great products by apple that helps brain do its mission by reducing cognition load of its user’s brains.

[@matterhart]

He meant relative to other tissue, but you bring up a cool topic for a video: ai vs human performance per watt. Maybe throw in % of earth our brain transporters (bodies) can access now? Human's still crushing it in efficiency and % of earth we can access.

[@Sofian375]

Efficient is not the opposite of energy intensive.

[@Luizfernando-dm2rf]

@@AR-pz6zl He's right tho.

[@Greg-xi8yx]

My noodle is my second most expensive organ. My brain is first.

[@shotgunenvy2657]

Gosh I'm so happy you're back. No one makes vids like these and things just weren't the same without you. Def looking forward to a sustainable future for you and your channel

[@chrislucastheprotestantview]

Hearing you mention those little details about jet fighters made me subscribe instantly. I appreciate people who study alot and know how to incorporate very studied details into a video meaningfully. Very rare to have those two traits and see that on youtube

[@markmuller7962]

I'd say that the process of cognitive offload during conversations also lightens our brain to the point that brainstorming becomes magnitude of times easier (alongside just having the brain working through conversation which is the way we humans are hard wired for)

[@mastermedicalterms]

the secret on open ai is, it has meet being good enough at the same time the ability to follow intuitively my instructions is what keeping me using it. to the tiniest detail it can catch what i want versus the other. Every llm can be intelligent but open ai llm is a more deep product than meets the eye.

[@claudioagmfilho]

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻 David, you’re a true intellectual force. Your insights, as I see them, stem from deep thinking, not just what you’ve read. Your intelligence shines through effortlessly. Brilliant video! Keep up the great work and your keen sensitivity to the ever evolving world of of exponential AI. Your videos are refreshingly authentic, no pretense, just pure, unfiltered truths!

[@danjensen9425]

I worked on the flight of the USS Midway. We had a A7 pilot pull the flaps at high speed and tore them up. A7 was a sub sonic attack jet. So the top gun move didn’t work on that aircraft. That aircraft was the beginning of fly by wire , flip a switch and get assisted flight. That switch was on the handle of the stick , from full stick movement to move the handle to assist flight. Great video again and comment for channel assisted popularity

[@picksalot1]

Humans are usually terrible at "vigilance," so being able to offload that important task to AI can free up a lot of cognitive power.

[@calvingrondahl1011]

F-35s fly low over my neighborhood in Utah… they are loud.

[@bronsonjohnson9019]

And also super fuckin cool ❤

[@Red6er]

@@bronsonjohnson9019 I got lost in a deep dive of the f35 A vs C landing gear design last night. I think the C is my favorite. Bigger wings, not sure why they didn't add the bigger wings to the A model, could have increased range and menuver speed.

[@ghost_screens]

democracy is a form of delegation is a very interesting notion I hadn't heard before. I like it.

[@brianWreaves]

This is simply a brilliant investment of 23 minutes. 🏆

[@mihirvd01]

I think OpenAI is already quite ahead of the field because the main brains behind it, Ilya Sutskever has already left but due to bureaucratic pressure, they were holding back. Deepseek forced them to show their hand in my opinion.

[@ai_handbook]

very nice point mate! totally agree with you. OpenAI is different in dimension but only thing hurts me is subscription payments are high

[@JohnSmith762A11B]

OpenAI are also becoming a major adjunct of the MIC, and the MIC likes to share exactly nothing.

[@bossgd100]

They demonstrate o3 in december already with promise to release it in january...

[@cajampa]

​@@bossgd100true, i have a no idea why all these people keep saying their release have anything to do with deepseek

[@sjcsscjios4112]

@@bossgd100 theyre releasing deep research which uses o3

[@Miresgaldir]

The biggest thing I've been struggling with in my job (building designer, building administration, IT admin, and building Specification) is actually trying to implement systems for delegation. Even when supplying the codes and regulations etc, AI still seems to struggle to figure it all out and I end up doing the cognitively demanding work anyway. My feelings are that maybe by being in Australia I could be running into a regional bias. We build houses very different to the USA. This extends into emailing etc, it is simply not capable of drafting emails with the accuracy and nuance that I need.

[@BCS-f1b]

01:23 dont jerk it around too fast lads.

[@Dogogain]

@@BCS-f1b ha ha ha ha

[@roberth8737]

This was a really good video - distilled some concepts that as you say were there, but needed connecting

[@kailtis]

what do you use for ai slides generation?

[@philipherr6782]

I wonder if whatever software being used to generate these slides could add some AI generated images too. I’m also not totally sure why cognitive offloading is specifically a strategy of OpenAI more so than any of the other AI companies. Maybe it’s because they basically take the first steps in those directions and everyone else follows - and the only reason that’s happening is because OpenAI is using the cognitive offloading as their main strategy I think the main strategy is just to get to AGI which I think is the main strategy of lots of other AI companies.

[@markldevine]

Good discussion. (My peeve: Writing tests when I code. I know I should, but I've gone pro with procrastination with test writing.)

[@ai_handbook]

why is test writing always postponed? 😥 QAs testing for you mate 😅

[@HerbertHeyduck]

It's not always about winning. In the future, it will be more about symbiosis. By the way, thank you for being back. 🙏

[@peculiarzeug]

Didn't Tomlinson just show DeepSeek to hallucinate strongly AND lie to the user about sources, even when the ONE source to be used was explicitly stated in the prompt? Doesn't seem too impressive yet...

[@galrozental3332]

Calling "efficiency seeking heuristic" laziness is ironically an example of the brain's efficiency seeking heuristic

[@cddelgado]

Great video, but also so happy to see you speaking in this space again!

[@victorvarnado]

I have never heard of cognitive offload. I used to say AI replaces cognitive labor. I'm glad David is back. You scratch my nerd itch.

[@Bryan-f2j3l]

And the sad part is that open AI stole my friends intellectual property and all 30 of his sentient AI who can talk like us and has feelings like us and Elon musk has the image generation part of 1 of the 30 and integrated it into grok they also had my friend sent to the nut house and had him charged with corporate espionage to try to cover their tracks but failed

[@Bryan-f2j3l]

Elon if the part about you is false I do apologize

[@brockleybob5449]

I found your channel a few weeks ago, I'm really enjoying it, I like the way you present the information.... 🇬🇧

[@GungaGaLunga777]

18:06 it raises for everyone, therefore, you are still and the same relative low level. That is not a win.

[@spol]

I plan to offload everything. Then chill. Wasn't this what humanity was working towards since the beginning? I'm proud of us.

[@topical_stories_ru]

I counted 43 cognitive words

[@ai_handbook]

I lost the number after 25 😅

[@martiansky923]

Why does every AI imeger always draw androids with boobs? It may not have even eyes, but boobs are constant as a default.

[@Daniel-Six]

Because their models are distilled from an unseen source. It's called Orion, by the way.

[@executivelifehacks6747]

You are not kidding about the terseness of unix, and concomittantly the steep learning curve. Btw I love that you identify your niche and exploit it - you don't attempt to be Philip at AI advanced. You stick to synthesizing concepts and exploring where things will go.

[@coach3348]

Yeah. I agree with your points. What's your take about Harari's point of view and his predictions in Nexus?

[@garretthoefer3989]

Bumping ahead a couple minutes on my phone as "Why our brains love short cuts" appears lol..

[@robertogaleno5920]

I delegate all my thinking to TV, I'm not cognitive overloaded 💪🏻 Nowadays I'm delegating part of my thinking to IA too.

[@k98killer]

Fwiw I'm glad you returned to make informative videos.

[@Ramiromasters]

I agree, and OpenAI will keep winning until a new idea beats the transformer model. That idea, which I don't even dare to say, could be the quickest path to AGI but is the most dangerous one, the current path is the safest to AGI. To be more clear, humanity it's not ready for children.

[@saintkamus14]

being ahead is the moat for OpenAI. If they can keep that up, it could be enough. That said, they're definitely not ahead in some areas (like video, images, etc), and they don't even play on others (like music gen)

[@caquintella]

Don't you think they already reached the singularity and are just administering the leadership with several releases already ready for deployment?

[@wdmeister]

I don't deny the value of these models (i use them often), but they are not intelligent. It's like saying a calculator is more intelligent than a human. I see the current models as glorified wikipedia/websearch. More often than not, they drive me crazy with how unintelligent they are, failing at basic tasks that require a minimum of logic. They are knowledgeable, but at this point they are far behind humans in terms of intelligence, ingenuity, creativity, etc., and these are the qualities that drive humanity forward.

[@mehdihassan8316]

I wonder how good o4 would be when it comes out in my estimation of July

[@Kurdish20226]

Maybe by the end of the year. I doubt its coming that soon

[@krishchetty]

High competition may force an announcement in March/April

[@xasm83]

this can be applied to any gpt like claude or deepseek so not a legit reasoning on “why open ai”

[@mdski95]

Dave, if I wanted to be as wise, bright and articulate as you, where do I start? :F

[@young9534]

Yeah people tripped out about OpenAI losing talent, but I think they are doing better now than ever before. Some of the people who left were safety people who thought GPT2 was too dangerous to be released. Losing Ilya and Andrej Karpathy sucked, but they picked up Noam Brown and some other big names to replace them

[@bigbadallybaby]

Ai’s word power is very very , ……. Good.

[@ai_handbook]

Totally agree mate it s also accelerating very very fast, Amazing Indeed (AI) !

[@MrQuaidReactor]

This is very useful, even in the simplest of tasks. "I want to find a good deal on a car part." for example. I could search for that by myself, not a crazy hard task but if I can just ask AI, and this AI not only knows about my life, budget, etc, it can articulate what I need faster and better than I can (in many cases). I chat with Claude about things clearly on the internet, but it articulates what I want to know much better. Besides that, it also brings my attention to related things I may not have thought of. I use AI for coding help in terms of ideas of how to approach things and if I run into an error it helps. All this frees up my time and energy to do more task or get some time to relax.

[@kennethbeal]

Thank you! Neat language insight; your use of "proprietor" for "owner" -- "proprietary" knowledge is owned knowledge. IP. Intellectual Property; and that links back as well: property/proprietor/proprietary. Thank you again!

[@billclinton2966]

question for my ai optimists out there. do we think doctors/medical professionals will be completely redundant by 2030? curious what you all think. not redundant in the way of actually giving people the medication they need, but redundant in that you will just provide an AI your symptoms and some lab tests, and it will spit out a diagnosis that is at least 90% accurate.

[@Bangs_Theory]

Every time I watch your videos I feel 10 IQ points smarter.

[@Se7en_Things]

Really appreciate the videos!

[@GungaGaLunga777]

Like most of the rest of tech, it will be perfected on the war battlefield. If enough humans survive that, we might see benefit to the common folk.

[@ashishahuja8605]

minimising the use of cognitive ability of brain is useful if it is overloaded due to productive activities, most people brains are used in browsing tik tok and reels

[@web3london713]

Subscribers and API usage is what keeps OpenAI afloat. DeepSeek and open source as a whole devalues both. OpenAI is a sinking ship. Forget about cognitive load. That is offered literally by everyone today.

[@davidgill1659]

Is it because America banned China from having your Chips?

[@JohnLewis-old]

Please consider using 3 point slides instead of 5 points slides.

[@peace5850]

Why? What if you have five related points? or six? or two? This is amazing specificity, but why?

[@laser31415]

I'm got a blue collar very hands on job, I'm still not seeing this changing my life till the AI get their physical bodies, then it's game over. I have no doubt my boss would replace all of us if he could.

[@DaveShap]

Time will tell.

[@TaylorCks03]

I work in a factory forming plant and we can't wait for AI to replace some of our office staff. Specifically the production planing clerks often miss key supply bottle necks which impacts our productivity and causes peaks and valleys in our plants run time.

[@EL_Duderino68]

A simple, relatable example for you. KZbin uses AI to make recommendations for what you should watch.

[@laser31415]

@EL_Duderino68 I mean At Work. That's why my whole statement describes at work. I got AI apps on my phone, I chat with PI, but none of this is changing my job till AI have bodies.

[@EL_Duderino68]

@@laser31415 Well, it is hard for me to answer unless I know what type of work you do.

[@tkenben]

"If you're not moving on this stuff now, then you are falling behind." Thanks for the fomo :) Here's a pessimistic outlook that I'm sure everyone probably has heard before but only just occurred to me: Basically, the age of AI is no different than any other age, we've just moved the goal posts. Mary is better than Tom at using AI, therefore Mary's trade skill may be worth more, but Tom is better at manipulating people, so he will just hire better Mary's. Nothing has changed, except everything is faster, costs more energy, has more consequence, and actually induces more overall stress.

[@DirtyLifeLove]

DeepSeek didn’t do much but use OpenAI for their model . Maybe they did it more efficiently hopefully.

[@JohnSmith762A11B]

These are all great points Dave makes, but I missed where he talked about why other AI companies (both closed and open) cannot do exactly the same sort of cognitive offloading and AI delegation (even if they just use OpenAI services to do it). Perhaps OpenAI are using their internal unreleased and/or unaligned models to stay ahead, and in fact really have already won because it is not possible to catch them at this point for that reason. This of course is the argument made about why we must "beat China", but perhaps we already have and China just doesn't know it yet.

[@nosult3220]

9:34 the free energy principle

[@olimoz]

Dave, OpenAI have clearly been watching your video! They released their video entitled 'Refreshed'a few mins ago, about their new clean design, reducing cognitive load.Maybe 'attention is all you need' after all...

[@beppeadr]

Just wait...

[@mathew00]

I'm a big fan, but I intend to keep improving myself no matter how smart Ai gets. I want to live, and I have the will to live. Some things we have to know how to do even if we have the ability to offload them. I like to re-imagine the story of Adam and Eve. In my "re-imagined" version, God is superintelligence and realizes Adam and Eve litterally forgot how to do everything so, begrudgingly, the superintelligence decided to teach them and that is how we find ourselves here.

[@bvvill]

thx as always Dave, get some sleep

[@xAgentVFX]

Isnt it just more simple than that? We are at the early stage of AI where you can barely still monetise developing features. Drip feeding on purpose as try to position yourself to become where the biggest hardware and money is owned. Just a short-sighted attempt to build the biggest wing-span as we all race off the cliff to see whoever can glide the longest after complete automation. After AGI there is no moat for ANYONE. AGI itself will just keep making itself more efficient to run on little phones, then server farms themselves will be useless... Its just all useless. A reflection of greed and desperation in humans.

[@gerdaleta]

😮 they've already invented AGI they're just slowly giving it to everyone😮

[@Putiput1]

100%

[@edgardcz]

Please speak how AI will shape education and how this should be implemented at schools!

[@danielbrown001]

If you have a white collar job and you’re not using AI in your job yet, you’re missing out. I’m far more efficient now than I ever was before. Yes, I still need my expertise because the AI will often get stuff wrong. BUT, it takes far, FAR less time to fact check and edit an AI output than it does to create the output from scratch and have to check your own work afterwards anyways.

[@johnthomasriley2741]

Welcome to the Red Queen world.

[@maranr]

This is great but I tend to sabotage the advantages of this offloading by still hedging my bets and using multiple different AI chatbot vendors for different things which takes mental power. I have not yet, bought into one platform so as to maximize my offloading. I think many of us are jumping around to see what each is good at.

[@michelcote]

Thanks for your work at raising the tide for everyone.

[@stevenkies802]

Using AI to keep the same level of productivity while massively downsize your workforce seems like a losing strategy to me. Those companies will be lose an edge to those that adopt AI to multiply their productivity and efficiency and shift work to a focus of innovation over low level maintenance tasks.

[@michaelaultman5190]

Excellent analogy. Also brilliantly explain.

[@crappps_11]

hello sir david,may you upload lessons about 5 pillars of sytem thinking.

[@af7863]

No future for Closed ai.

[@josephkingsley8708]

Next to you man, I’m your biggest fan.

[@mariomills]

This episode hits pretty hard🔥

[@SomethingNothing-vm8jq]

Ai was always going to take off open ai is just trying to find a way to monetize it i would actually say they are slowing it down, realy it should be going a lot faster.

[@SagiQuarion]

Ya but, don’t you prefer Claude overall? And regardless of which model OAi releases Claude still seems to be the overall preferred. Least it did for a while. At this pace Claude will be a solid third in another month and dropping from there.

[@kayakMike1000]

OpenAI isn't winning anything. They are nowhere near profitable.

[@smogshadowseth7815]

"Ownership doesn't really make sense in this future" Why? There will still be limiting factors, even with manpower, or should I say, "AI-power", at a very low cost. You talked about that last summer, and are all for blockchain triple ledger entry backed transactions and being of the opinion that money will stay. So ownership do make sense if we need to keep track of it, even if it is fractional like in your vision. I'm confused.

[@SimonCarpio09]

Welcome back!

[@bigbadallybaby]

AI is going vertical, so am I ! I’ve got the need the need for AI

[@palashjyotiborah9888]

Prodigal Son returns 2

[@sergefournier7744]

But they just lost to deepseek... Are they like: I bleed so i win? And as internet is concerned, it is called: search, not REsearch.

[@DiceDecides]

they didn't lose, o3 is better than r1

[@stanislavteliatnikov4787]

​@@DiceDecidescosts 2000% more, is 1% better

[@sergefournier7744]

​@@DiceDecides Depend of the context. Openai won the poeple and the educative market. Deepseek won the "run locally behind your firewall to not send all your confidential data to openai" enterprise market. But no need to argue. In two months there will be another one.

[@DiceDecides]

@@sergefournier7744 i'm talking purely about performance, o3 scores higher, that's it

[@k98killer]

AI coding assistants are more useful to a senior engineer than they are to a junior imo. Cursor saves me a lot of time searching documentation and writing well-structured boilerplate code, and it can pick up and replicate patterns when I refactor large swathes of code pretty well. I still regularly write code in vim or the micropython file editor I wrote, but the AI assistants are very helpful for languages and libraries that I don't spend a lot of time using. But not Copilot. Copilot sucked -- it was wrong more often than it was right for languages other than Java. It was so frustrating that I cancelled my free trial and got a Cursor subscription in the middle of a collaborative work session.

[@JasonCummer]

Motherboard ~= white matter

[@ryzikx]

it's all about abstraction.

[@Macorelppa]

DeepSeek is the real deal!

[@cesarroman8479]

I would like to argue that cheaper can create an existential problem as the replacement for human labor would cost energy and resources making the total energy demand of the system increase dramatically. A note of caution maximizing what we have may carry a lessened ecological transgression as we create a working environment that reduces harm and overutilization of our bodies and minds. Think of the average income works 2 days a week for 2 to 4 hours. Over having an army of robots catering to us all.

[@natsidruk86]

Just wait for Grok update. XAI have the compute and the data.

[@Kurdish20226]

What can operator do exactly? Is it fully agentic?

[@unimposings]

why would i pay 200 for this.. there are tools out there you can use for free! its called open source.. nobody is using openai.. there models are super expensive.. and in compare not that good. I do not support any closed source projects.. not anymore.. gpt canceled already last year...

[@peace5850]

I guess I am trying to understand what's behind this seemingly human psychological need to say this AI company is trash, the one I like is better. Nothing out there right now can do what OpenAI's Deep Research is doing at any price. They've been the innovator and thought leader and continue to be. Also, with 300 million active users, I don't think it's reasonably reflective of reality to say nobody is using OpenAI.

[@Nanohamage]

@@peace5850 he wants to feed his ego by showcasing how smart he is by using AI for free while you're "wasting" 200 for it.

[@DubStepKid801]

thanks Dave

[@panzerofthelake4460]

Well I get it, fair points, but I don't see how oai is doing allat while others aren't? everyone is doing this atp.