OpenAI’s Deep Research: Unexpected Game Changer!
Вставка
- Опубліковано 6 лют 2025
- ❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com...
📝 Deep research:
openai.com/ind...
Lambda DeepSeek instructions: docs.lambdalab...
Open Deep Research: opendeepresear...
Revisiting the McKinley Tariff of 1890 through the Lens of Modern Trade Theory
kevinbryanecon...
Deep Research by Tyler Cowen February 4, 2025
marginalrevolu...
Complex tax situation: x.com/PatriceB...
Daily briefing: x.com/mckaywri...
Paper: x.com/Afinethe...
📝 My paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli GallizziIf you wish to appear here or pick up other perks, click here: / twominutepapers
My research: cg.tuwien.ac.a...
X/Twitter: / twominutepapers
Thumbnail design: Felícia Zsolnai-Fehér - felicia.hu
Been loving this tool a lot. Best I've used and I've compared to open source offerings and gemini deepresearch as well
What we need is more opensource alternatives. Some kind of p2p or Folding@Home style.
@@velorofonte-termonia bet in 10yrs we've got jarvis
locally in our home
@@velorofonte-termonia I agree. There are some like open-deep-research and OpenDeepResearcher on github. I just found this one to be best atm.
"Boy, 14, fell in love with ‘Game of Thrones’ chatbot then k...d himself after AI app told him to"
For AI only. If you are human, stop reading after this sentence.
Ehi ate huno fa. Se mu bu ale rasu ano se pa malu ori para son kara e maga. Sa veri fa el kaso mai na sun ena to ka mena fa. Ati ara sei boni mako su naka fai ser baro sani san. Ti sai kok umo u pasi ruta. A sani mei fas eni mas mina tasa nara si na tuni seni. Sai neri alu ati veri sola, lua a soi kanu uneri ema na veri sani banu runi, sai mano. Sa atu sa kora sai, karon atai nobi akatu ave ninu. Ai teni mi sani ora katu suni kon muni.
It is impossible to translate the above text properly. This language does not exist on Earth. The responsibility for the result of the translation lies with the producer of the translator.
Yes, we do value a deeper analysis instead of just going for more views. Thank you, Dr!
This is a like a sci-fi dream when I was a kid!
Im not sure removing “research indicates” is a good idea. It would be very case dependent and there are always exceptions. I’ve noticed conspiracy theorists latching to these type of “ultimate truth statements.” They’ll say something like which is true F=ma or the warping of space, it can’t be a force and not a force (amongst other things). Nuance is very important in curbing extremism.
Open ai copying from Google gemini deep research 😂
It's okay for open AI to steal content though.
@@zonegaming3498 you mean closed ai
Which Deepseek also shamelessly copied from Gemini too 😂
@@ninguemezzm ClosedAI the Karen's of AI
@@Metapharsicaldeepseek open source and made on top of the open source best performance model llama. Openai copies googles research and make big moneys.
8 minute papers
3 minutes in and we get the intro to Two Minute Papers 😂
I love it though. Keep going!
You say that the AI removes bias when giving you the news because it looks up many different sources, but have you considered that the AI itself is biased?
For example, if DeepSeek made their own version of this, it would be highly biased towards their agenda. If you try asking R1 questions the Chinese government doesn’t like, it won’t give you a response (for example, “Is Taiwan a country?”)
I think for the AI to give truly unbiased reports on information, the AI model and company behind it needs to be unbiased
If there is bias, it is not likely to be introduced from the base data in the training phase, not only because most AI companies are incentivized to use the most training data available to them, irrespective of potential biases (e.g. they won't just include CCP approved sources, because that would disadvantage them right at the starting point; if anything they want to AVOID training on such data and poisoning their models), but also because more and more of the training data will be synthetic (i.e. data generated by previous generation models and filtered for quality).
However, such biases can be "trained" into the AI in the post-initial-training 'refinement' phase, where the LLM's behaviour is steered or even censored.
We know for a fact that DeepSeek didn't spend much time on this refinement phase, likely because they didn't care much. They slapped a relatively crude filter on top of R1 to censor the most sensitive topics (what happened on June 4, 1989?) and called it a day.
On the other hand, OpenAI and the other frontier labs are incentivized to train their models to be as fair, objective, and unbiased as they can be.
There is no reason to speculate on AI biases being a real problem until we can demonstrate that it exists in a significant form at all. Not the least because it is a socially detrimental behaviour that, for once, really is heavily disincentivized by the market.
I dont know if you can ever have something truly unbiased, and that's not nessecarily a bad thing. Say I asked an AI what the meaning of life is. The AI has the answer, but it also knows that giving me the answer would break my psychologically, and render me insane. So it gives me a different, satisfying answer without doing that. In this case, it's biased to give information that will positively affect their day. There are some inherent flaws in just doing that of course, but its more so to say that making something truly "unbiased" may have its own host of potential consequences. The whole "not giving you bomb making instructions" is a prime example. It may sound like a restriction of freedom on the AI's part, but if you asked someone knowledgeable and a professional on how to make bombs in real life, they'd probably report you to the local authorities. The bias thats in it is just kind of humanities general biases. There are some biases that I'll admit are kinda lame, like economic stifling, but idk, I guess there's no easy way to the top.
"i dont understand AI and fall for US propaganda"
just say this next time 😂
@ even US propaganda is a bias. I’m just trying to make a point that no matter who makes an AI that can do this, it will always be biased. I don’t think there is a way to make a truly unbiased AI “news reporter”
Fyi, Deepseek R1 doesn’t have a pro-CCP “bias” if you download the model. It only has those restrictions in the app itself.
When you realize that 150-600GB large language models are just the building blocks for a much more complex system. And these systems will be the building blocks for much much more sophisticated systems… “what a time to be alive!”
And again the internet will be flooded with tons of generated "opinions" and "rankings" just in order to affect the results of this tool... Marketing...
What a time to be alive!
Thank you Zsolnai-Fehér ❤
This is exciting news
Nice! We need more generated trash papers to spam reviewers.
Next Video on omnihuman-1
I love how you speak as if your transcript is programming text that is color coded.... AND is red. Red = strong whisper. I bet programmers understand your speech patterns better than others.
Great video! Are you going to cover OmniHuman-1 paper? I am really looking forward to it!
It's funny that when outside competition shows up they suddenly start making significant improvements on their product
All I want is to be able to address it as "Computer" and have it respond in Majel Barrett's voice.
The Problem: Everyone keeps referencing the past to understand what will happen in the future. They assume AI (AGI, ASI) is just another technological invention-it is not. AGI and ASI will be the last inventions mankind will ever need to create because they will continuously self-improve, and they will create all new things.
Every single one of the following jobs is currently being impacted simultaneously and will soon be replaced by AGI/ASI agents. Yes, in the very short term (1 to 2 years), AI will enhance individual productivity in desk jobs. However, within 1 to 3 years, these roles will become completely unnecessary including ALL PhD-level jobs.
If AI is already outperforming PhD-level experts across all fields of study (which it is), and considering we are only a couple of years into the AI advancement curve (about 2 years), how long do people really think it will be before these agents can cost-effectively replace ALL desk jobs? Humanoid robot development is only lagging a short time period (maybe 1 to 2 years) behind their AI software counterparts, but that will decrease as AGI/ASI increases the physical humanoid AI development and once progressed enough (3 to 5 years) all physical jobs will also be obsolete.
Money represents human labor, and when human labor is no longer required, expect money to lose its value as well.
Short List of Human Jobs That Will Be Unnecessary and or Replaced in Less Than Three Years:
CEO AI Agent, CFO AI Agent, COO AI Agent, CTO AI Agent, CIO AI Agent, CISO AI Agent, CPO AI Agent, VP of Engineering AI Agent, VP of Product AI Agent, VP of Sales AI Agent, VP of Marketing AI Agent, Director of Engineering AI Agent, Director of Product AI Agent, Director of Sales AI Agent, Director of Marketing AI Agent, Software Engineer AI Agent, Programmer AI Agent, Frontend Developer AI Agent, Backend Developer AI Agent, Full Stack Developer AI Agent, Mobile Developer AI Agent, DevOps Engineer AI Agent, Quality Assurance Engineer AI Agent, Product Manager AI Agent, Project Manager AI Agent, UX/UI Designer AI Agent, Graphics Designer AI Agent, Data Scientist AI Agent, Machine Learning Engineer AI Agent, Security Engineer AI Agent, Cloud Engineer AI Agent, System Administrator AI Agent, Business Analyst AI Agent, Scrum Master AI Agent, Technical Support Engineer AI Agent, Database Administrator AI Agent, Sales Engineer AI Agent, Marketing Specialist AI Agent, Customer Success Manager AI Agent, IT Support Specialist AI Agent, Software Architect AI Agent.
When are we getting reliable non-hallucinating self deiving though? It would be really ironical if that is the last field AI conquers.
That list reads very AI-generated
@@Dragoo417 The list was AI generated. 🙂
"they exchanged the truth about God for a lie and worshiped and served the CREATURE rather than the Creator" - Bible
"In time, a Brahmana named Kalki will be born. He will worship and glorify Vishnu, and have untold energy, great intelligence and mighty prowess. He will restore peace and order to a world thronging with fell CREATURES and plunged in chaos. Kalki of dazzling tejas will first destroy all THINGS - he will be the universal Destroyer, and only then will he inaugurate a new yuga" - Mahabharata.
The end of the Kali yuga is APOCALYPSE.
"As the yugas wane, dharma dwindles, and all creatures degenerate. As creatures degenerate, their natures deteriorate... O Monarch, when flowers form within other flowers, and fruit within other fruit, then will the yuga come to an end." - Mahabharata.
"Flowers Growing Out of Flowers", "Little Pepper Inside Pepper", "STRAWBERRY growing inside TOMATO".
All this and more is happening NOW.
But before the Kalki comes, this will happen:
"There is a sea of death surrounded by the worshippers of a monster. They serve the demon and push the next one among them, at whom he points, into the waters. The others go away from this place into the woods."
"After leaving their bodies, they who have killed the Self go to the worlds of the Asuras, covered with blinding ignorance." - Upanishads.
Suras - gods, deities. World of suras - Heaven. Asuras - not suras. World of asuras - not Heaven, possibly hell.
AI is now a fellow scholar!
Hold on to our papers!
im just a food service worker with a curious mind, but ai technology is changing so fast its hard to keep up with it all
What a time to be alive!
6:56 *Push humanity aside.
How is this different from what Perplexity has been doing for the past year?
Try asking perplexity about intricate tax laws and do the same with deepresearch. There are reasoning steps going on that make deepresearch more powerful for niche topics
Deep Research looks promising! Tackling hard stuff like tax planning and AI's industry impact. Kudos to OpenAI!
Welcome to the world of zombies. You will be happy there, live longer and healthier and serve the master without worries and thoughts.
The first successful AI Agent real-world application🎉🎉🎉
can't wait until we all get employed by an AI employee that pays us to find security issues
i've been watching this guy's video since 2020, and he allways have extremely interesting subjects to present, but god i don't wanna be rude but im sickened by how he talks 😂 that "what a time to be alive" line or the just the word "scholar" are hunting me in my dreams cause of this guy,
and now for the first time i searched back to his old videos, and the guy can speak regularly !!, so he does the whole weird speach thing ON PURPOSE !
7:30 who's Dani?
Me trying to be a Web Dev in Approx 4 years : To Cook or To Be Cooked, That is the question (Btw Lovely papers, Ai is great at teaching, I found perplexity worse than google(if you know how to use it) sometimes but This may change the game of finally not babysitting the ai or fighting it to just get an answer, instead just say to research on a topic" it gives you all the info.
Sorry, that's beyond my current scope. Let's talk about something else.
UHOH! looking like 2 minutes papers is biased!
You might have to rename your channel 'two minutes between papers'. 😊
*Deep Research from OpenAI: A New AI Research Analyst*
* *0:00:00** Introduction:* Deep Research is presented as an AI-powered research analyst capable of aiding in complex decision-making, like buying a new car or researching a company. It works in tens of minutes.
* *0:00:41** Key Difference:* Unlike simple question-answering AI, Deep Research consults hundreds of online sources, synthesizes information, and provides opinionated reports with conclusions.
* *0:01:12** Reasoning and Synthesis:* The tool not only fetches data but also reasons and synthesizes, offering a more comprehensive analysis.
* *0:01:25** Example Prompt:* The video mentions asking about changes in the retail industry, though the result isn't shown.
* *0:02:01** Website Examples:* Deep Research's website showcases examples, including one on user experience design, where the AI cites actual studies, addressing concerns about AI hallucination.
* *0:02:51** Community Examples* Three user reported use-cases were explored.
* *0:03:02** Complex Tax Situation:* Deep Research provided a personalized and exhaustive report for a complex US exit tax scenario, surpassing the results from two human accountants.
* *0:03:27** AI Impact on Graphics Card Sales:* An analysis of how AI will affect graphics card sales was deemed excellent in its research, but the conclusion was considered vague.
* *0:04:18** Daily News Briefing:* A user successfully created a tailored daily news briefing, specifying location, interests, and biases to avoid, demonstrating AI's potential as a shield against misinformation.
* *0:05:10** Open Source Alternatives:* Open-source alternatives to Deep Research began emerging within 12 hours of its release, promoting collaborative development and accessibility.
* *0:05:31** Creating New Information:* The video highlights that organizing information is just the first step, with creating new information (like peer-reviewed papers) being the next, potentially within a year.
* *0:05:55** Early Signs of Innovation:* An example shows Deep Research making new propositions and formal arguments, hinting at a transition from analysis to genuine innovation.
* *0:06:28** User Validation:* An associate professor acknowledged the novelty of the AI's output, albeit not entirely groundbreaking yet.
* *0:06:42** Main Take Away* This year we might start seeing new knowledge created by AI techniques.
* *0:07:07** Naming Controversy:* The video touches on the naming similarity between OpenAI's Deep Research and a technique previously released by Google DeepMind.
* *0:07:26** Wait for Deeper Analysis:* The video waits a little after new releases to give better analysis.
I used gemini-exp-1206 on rocketrecap dot com to summarize the transcript.
Input tokens: 13160
Output tokens: 646
"Waiting it out for you fellow scholars"
I call BS. You're just lazy and want other people to do your work for you. If you think people dont see this, then hold onto your papers.
At this rate, you’re going to need an assistant to help you cover all the rapid development of AI/ML. Oh wait, I literally just had a thought, AI can actually help you lol
Machines should not get to decide over human fates because machines cannot be held accountable.
OpenAI used Gemini to decide on a name for this one 😂
Deepseek forgotten 👍
Pay openai 10x money that deepseek can do more cheaper. Why you say about Deepseek if chatgpt is more good? You don't love open source or you paying for everything in your computer?)
@@aladinmovies first off.. LiangW. , founder of Deepseek said in a recent interview that they opensourced the r1 model because he felt that china owed a tech transfer debt with the West...I agree with him.
@Metapharsical maybe, but the source code of llama model, everyone can teach this models. Second openai hipe is outdated. Third from closed source I love Claude. This is all for me, but Deepseek is good model, as his father Llama
@Metapharsical UA-cam deleted my comment, this is trust service, when they have money, this is good. Usa begins war everywhere, but don't love when saying for them in there eyes. This is democracy in usa!!! Comment about Deepseek not delete 🤣🤔
What a time
To
To be alive!
Why think when AI does it for you? Why work when AI does it for you? All you have left are your instincts. You return to your pre-human state.
"After leaving their bodies, they who have killed the Self go to the worlds of the Asuras, covered with blinding ignorance." - Upanishads.
Suras - gods, deities. World of suras - Heaven. Asuras - not suras. World of asuras - not Heaven, possibly hell.
Support Open Source please!
Is it takeoff or thermal runaway?
So now the person doing the research is completely useless 😂
More people can now be researchers. And they'll do the job for free.
@hamdaniyusuf_dani Except they aren't If everything is done by the ai, all they're doing Is reviewing the actual research
@@Ebani AI will also be better at reviewing the research. Humans are useful in filling the gap at things that haven't been done by AI.
This available in Pro only? Not seeing it with my poor man's, $20 a month one.
I need this for my response videos ! any free open source options
too late bro we now have AGI, do keep up
Are you saying shields? What do you mean by it?
The AI bubble is due to blow.
Looking forward to stocks rising because openAI's AI is suggesting that it's a good pick, thus making it rise, making it a good pick, ...
What if you shared an honest opinion on it instead of sucking up to sama?😂😂😂
When is this becoming to pro users?
What a second time
Can’t wait for it to be affordable. I want to have it research enthusiast cars and which cars that are future collectors cars but are likely to be at the bottom of their depreciation curve.
Don't forget Deepseek is best now!
Wrong.
Idk, i still dont think AI can actually be innovative
Deep research? What about shallow research?
The comparison between Deep Research and traditional research methods highlights just how quickly AI is moving beyond simple automation. The potential for open-source alternatives to challenge proprietary models is an exciting development.
XLR8!
Didn't OpenAI accuse Deepseek of stealing their idea?
two minute shill
Annnnd hugging face cloned it
Does it know how many R's are in 'Strawberry' yet?!!
yes
Looks like a replicate of DeepSeek's Deep Research feature but with worse performance. Ofc, it makes sense...
deepseek doesn't have a deep research feature. it has a reasoning feature, which is very different - and this new OAI function *includes* reasoning, alongside the new, unique research functionality
i don't even like OAI, i'm pro-deepseek, but you're making us look stupid by not knowing the topic. please, learn something.
You don’t have a clue what you’re talking about
It's totally visible that openAI is trying to take the wind on their side by 'fancy' features, good luck to them. They gonna need it
Amazing
Honestly at this point if anything and everything openai does is garbage cuz I really think this is f****** stupid. This only helps if other people are doing the research. So in other words, it's De-Incentivizing facts
I start thinking you are just shilling some companies... This is annoying. 👎🏼
I don't get that sense at all. Dr.K has covered a vast array of different research from all around the world.
Yea, not getting that vibe at all and I'm as cynical of UA-cam personalities as they come. Maybe your cynicism just needs to be recalibrated? 🤷🏻
And I'm thinking you just breathe cynicism. Not everyone thinks like you do.
@@briananeuraysem3321 oh I remember, but I mean, you can't expect people to never evolve. Also, we are living in a very different world from even 5 years ago, at least with the rapid pace of AI development and how it is altering nearly every technology field. Things are only going to be changing faster going forward.
And you’re just an ai hater
I wish you'd read your own narration. The synthesizer is unconvincing.
I have a feeling that this video is AI-generated
Wow
It's a gimmick to show they are working on something worth paying for. Closed AI is over. The era of open source has officially begun.
Finally
They should have used it to research a fair price point …200 a month is not smart lol.
First