Science & Technology
AI Won’t Plateau — if We Give It Time To Think | Noam Brown | TED
To get smarter, traditional AI models rely on exponential increases in the scale of data and computing power. Noam Brown, a leading research scientist at OpenAI, presents a potentially transformative shift in this paradigm. He reveals his work on OpenAI’s new o1 model, which focuses on slower, more deliberate reasoning — much like how humans…
Science & Technology
Why Snowflake is no longer just a data warehouse
Snowflake is betting that the future of AI isn’t just analyzing data, it’s acting on it. That means a shift away from chatbots and toward autonomous agents that can actually get work done. And Snowflake is reorganizing fast to keep up, from shipping hundreds of AI features to restructuring teams along the way. On this…
Popular Science
The Experiment That Tried to Weigh the Human Soul
It’s a little complicated to weigh a dying person on a hospital bed, but that didn’t deter Duncan MacDougall. In the early 20th century, MacDougall’s unique bed-scale detected that 21 grams left the human body at the moment of death. He had finally discovered it: the weigh of the human soul … or so he…
CNET
NASA’s Artemis II Moon Flyby Livestream
Watch the excitement as NASA sends four astronauts on a historic mission to the moon, potentially farther into space than any humans have ever traveled. Follow CNET’s Live Blog at CNET.com NASA Artemis II Day 6: Monday Is Moon Flyby Day Add CNET as a trusted news source Never miss a deal again! See CNET’s…
Tech Stocks Rally on the Back of US-Iran Ceasefire Deal | Bloomberg Tech 4/8/2026
Snowflake Seeing Strong Return on AI Investment: CEO
Apple’s Foldable Phone On Track for September Launch
-
Science & Technology6 years agoNitya Subramanian: Products and Protocol
-
People & Blogs4 years agoSleep Expert Answers Questions From Twitter ???? | Tech Support | WIRED
-
CNET6 years agoWays you can help Black Lives Matter movement (links, orgs, and more) ????????
-
Wired7 years agoHow This Guy Became a World Champion Boomerang Thrower | WIRED
-
Wired7 years agoNeuroscientist Explains ASMR’s Effects on the Brain & The Body | WIRED
-
Wired7 years agoWhy It’s Almost Impossible to Solve a Rubik’s Cube in Under 3 Seconds | WIRED
-
Wired7 years agoFormer FBI Agent Explains How to Read Body Language | Tradecraft | WIRED
-
CNET6 years agoSurface Pro 7 review: Hello, old friend ????
@ErikMuellerGermany
February 15, 2025 at 8:42 am
My biggest concern about AI is GIGO!
Garbage in Garbage out!
@young9534
February 15, 2025 at 9:23 am
Reinforcement learning helps with that. Its no longer just training on internet data. The results have been pretty good so far
@HicSuntL3ones
February 15, 2025 at 10:35 am
Try calling o3 garbage
@jacksonstempel3382
February 15, 2025 at 3:40 pm
This is why increasing efforts are being put into synthetic data generation
@Stories-Today
February 15, 2025 at 9:22 am
The comparison between human decision-making and AI thinking time is fascinating. It makes perfect sense—humans don’t make every decision instantly, so why should AI? Excited to see where this approach leads
@RantKid
February 15, 2025 at 1:50 pm
and it will only require the power of 3 small nations per query!
@ToneyCrimson
February 15, 2025 at 4:03 pm
@@RantKid If it can give us cancer treatment and solve climate change than it can take the power of 10 middle sized nation per query if i care.
@Que-Lindo
February 15, 2025 at 9:23 am
Wow you found the least charismatic person in the world. That’s an accomplishment.
@jeremyfmoses
February 15, 2025 at 9:41 am
Sure, I’ll take that bet. In 2020, I proposed a definition of intelligence (artificial or not): the amount of hidden information overcome in order to predict the future. Even then I wondered if there was an upper limit to intelligence, and if so, how close humanity was to this limit. With some time and perspective, I now think there probably is an upper limit, which is not determined by any deficiency in our makeup, or that of our machines, rather a limit imposed by the availability of ever bigger questions!
Can we predict the weather in 24 years? Can we build a structure that can withstand a supernova or predict the ultimate fate of the universe? Can we find a way to take into account every single human and still provide a future of peace, fulfilment and harmony? Once we can answer questions like this, there is no more hidden information to discover, an intelligence will truly have reached its limit. For that last example, I will symbolically bet 1 Canadian penny, because I further predict in such a future we will have no need of money. I cannot wait to be proven wrong.
@TheCD45
February 15, 2025 at 9:51 am
lol this did not age well 😆
@ShpanMan
February 15, 2025 at 10:05 am
Weird talk, lots of false dichotomies. Either you can do the thing we’ve been doing, or the thing I think works, that’s it, nothing else to possibly consider guys.
Either you think fast (booooo) or you think slow (yayyyy), wow, definitely seems logically sound.
What lil bro meant was let the AI think (search for a path to a solution) and not just go on vibes (based on best guess in the weights). Very different from what was said.
@Unapologetically786
February 15, 2025 at 10:24 am
Oh look, Americans trying to stay relevant on AI
@miteshmitesh9471
February 15, 2025 at 10:27 am
thought i already knew everything about manifestation, but this book pointed out something i never considered before. tried the method and weirdly enough, things started lining up in ways i can’t explain. Vibrations of Manifestation by Alex Lane really showed me the missing piece.
@MdAjmal-x4j
February 15, 2025 at 10:27 am
i kept seeing people mention this book everywhere, so i finally gave in and got it. wasn’t expecting much, but something about the way it explained energy and focus just clicked. tried one small shift, and things started changing almost immediately. Vibrations of Manifestation by Alex Lane really made me see manifestation differently
@sooma-ai
February 15, 2025 at 10:31 am
Noam Brown discusses a paradigm shift in AI development, focusing on slower, deliberate reasoning (system 2 thinking) rather than just scaling up data and computing power. This approach, exemplified by OpenAI’s o1 model, could accelerate AI progress and solve complex problems more efficiently.
@DominionAnako-bb7ry
February 15, 2025 at 11:47 am
We don’t have to trust AI everytime
@TheTuubster
February 15, 2025 at 11:54 am
Here are some key insights from Noam Brown’s lecture on the progress and future of AI:
Scale as the Key Driver of Progress:
Brown emphasizes that the primary reason for the significant advancements in AI over the past five years is due to scale—bigger models trained on more data for longer periods. This has largely been based on the Transformer architecture introduced in 2017.
Cost and Scalability Concerns:
The lecture points out that while scaling has been effective, there’s a rising concern about the sustainability of this approach due to the increasing costs. From models costing thousands to now hundreds of millions of dollars, there’s a question of whether this can continue to trillions.
Introduction of System 2 Thinking:
Brown shares his pivotal experience with AI in Poker, where the introduction of system two thinking (deliberate, slower processing similar to human strategic thinking) significantly enhanced AI performance. This was a game-changer, showing that thinking time could be as or more beneficial than just scaling up model size or training data.
Practical Applications Beyond Games:
The concept of system two thinking has been applied not just in poker but also in chess (Deep Blue) and Go (AlphaGo), demonstrating that the benefits of this approach extend beyond simple games to more complex decision-making scenarios.
New Paradigm for AI Development:
With the release of models like Grok by xAI, which incorporate thinking time before responding, there’s a new dimension to AI scaling. This isn’t just about increasing computational power or data but also about optimizing how AI uses time and strategic thinking.
Economic and Practical Implications:
The cost of querying AI models is minuscule compared to training them, suggesting that scaling system two thinking could be economically viable. Brown argues that for critical applications, like medical research or complex problem-solving, the added cost and time might be well justified.
Future Outlook:
Brown is optimistic about AI’s future, suggesting that we’re at the beginning of a new era where AI can improve not just by more data or bigger models but through enhanced strategic processing. He counters the notion of an AI plateau by highlighting this untapped potential.
Audience Engagement and Real-World Examples:
By asking the audience if they’d pay for valuable AI solutions (like cancer treatments or solving mathematical conjectures), Brown makes a compelling case for the practical value of advanced AI, shifting perception from AI as mere chatbots to powerful tools for solving real-world problems.
This lecture provides a hopeful and strategic perspective on AI development, focusing on combining scale with smarter, more deliberate AI processing methods.
@kinngrimm
February 15, 2025 at 12:20 pm
why would i bet, i have been convinced for the past 4 years that we are closing in on fast take off
@GiblikJovanovic
February 15, 2025 at 12:52 pm
Why isn’t everyone talking about The Hidden Codex of the Financial Alchemists on Bevelorus this book is next level
@ninjabard1898
February 15, 2025 at 12:52 pm
AI is making us dumber. Maybe all the infrastructure and money going to force AI products down our throats should go to actual education and innovation.
@paulhiggins5165
February 15, 2025 at 1:12 pm
Surely this ‘time to think’ meme is just a rhetorical device?- are we really supposed to believe that these machines require a ‘thinking time’ in any way disernable on a human scale of time perception? Even the ‘thoughts’ exposed by this theatrical device are themselves artifacts, produced to persuade the rubes that they are gaining some kind of real insight as to what is really going on inside these programs.
If it’s really true that allowing these machines to ‘think’ before they respond produces delays in response times detectable by humans then maybe the developers should be looking for a refund from Nvidia, since the chips they have supplied are clearly operating incredibly slowly.
In reality this whole ‘thinking time’ meme is just a way to keep the AI bubble inflated for just a little while longer- pure theatre.
@musicandgallery-nature
February 15, 2025 at 1:19 pm
ku ha ki ni ko ik kre hak os trut khe ik bha ka, ni ba ju si ha ne ik kre hak os trut khe ik ba ha ni
Translations:
Khasi -> If you have your rights os trut, you don’t have your rights os trut.
Sanskrit -> I’m not sure what to do, but I’m sure I’ll be able to do it
Hindi -> The fog should not be shaken by the fire of the fire, the fire of the fire should not be shaken by the fire of the fire, the fire of the fire should not be shaken by the fire of the fire
ChatGPT -> The light of truth shines upon those who seek. Be steadfast, for the path is revealed to the pure of heart.
I zati angi eku rau, i ui a mona kat poje lei, to apa na nebi ti sonii, o so mora e atun li stali fre, /0.
Translations:
Samoan -> I am a very old man, and I am very old, but I am very old, and I am very old, /0.
Hausa -> I want to see you, I want you to go to me, I want you to see me, I want you to see me again, /0.
Hindi -> I am going to create a ruckus, if you ask me how can I pose for you, then I won’t even sleep, that is why I have to stay here at least forever, /0.
ChatGPT -> I see the stars above, I feel a warmth inside me, but the night is silent, and the world waits for the dawn.
THEY ONLY FOOL YOU
@RantKid
February 15, 2025 at 1:49 pm
Ah yes let’s encourage making it stronger so it inevitably becomes the strongest psuedo bioweapon
@gavinderulo12
February 15, 2025 at 5:52 pm
It also has the potential to help us by, for example, curing diseases.
@MrYondaime705
February 15, 2025 at 7:54 pm
Every Innovation has the power to do both good and bad. The internet have destroyed lives and made them 1000x better. If you’re scared of that, stop using the Internet because thats part of the innovation you’re against.
@PrajwalDSouza
February 15, 2025 at 2:23 pm
Unlike Karpathy, Noam’s words would’ve had more value if he wasn’t working for a company like OpenAI.
@pavel00995
February 15, 2025 at 2:46 pm
So this is an advertisement. Thanks for nothing.
@voranartsirisubsoontorn
February 15, 2025 at 2:50 pm
Be Sure better than Be Fast
@hiphopheaven
February 15, 2025 at 3:21 pm
I don’t see a big difference between chatgpt 3 and the latedt model in the quality of generation. I think it already plateauwed
@TheVlogTheory
February 15, 2025 at 5:37 pm
You are not using these models correctly at all then.
@Intinnent
February 15, 2025 at 6:34 pm
If you ask both models simple questions, you will get simple answers.
@priteshpatil5363
February 15, 2025 at 3:27 pm
Oh man , did you heard about scaling laws and it’s diminishing return😂
@grumblewoof4721
February 15, 2025 at 7:00 pm
Thinks like a human… there are many humans and they all think differently . Which one does this AI think like? Elon Musk, Donald Trump or it’s creator ? Either way, humans are selfish, greedy and prepared to survive at any cost. Is this AI a survivor ?
@kamertonaudiophileplayer847
February 15, 2025 at 7:50 pm
Maybe AI should be trained professionally? You go to a university, then post grad, and all time professors were around you. But AI? No professors, just engineers. AI, go back to school.
@sunkid86
February 16, 2025 at 9:32 pm
No one is worried AI is gonna plateau.
@six1free
February 16, 2025 at 11:09 pm
duh, that’s what deepseek did
@agustinbs
February 16, 2025 at 11:09 pm
What do i bet? the criteria was not even vaguely stablished
@KoZeroSM
February 16, 2025 at 11:57 pm
*CloseAI
@bhavtosh5328
February 17, 2025 at 12:11 am
Now AI is unstoppable.
Those are young enjoy and
do what interest you.
@DBENTLEY369ig
February 17, 2025 at 1:13 am
Won’t need AI from basic logic of our bigbang creation instant with MetaKlaus Alt.0167 and Neuralink
@vamsiramineedi6296
February 17, 2025 at 3:39 am
We have o3 mini released already, and here’s a talk about o1😅
@Jim-g9f9p
February 17, 2025 at 5:34 am
savage
@nathangwyn6098
February 17, 2025 at 5:48 am
These videos trigger me.
@gauravtejpal8901
February 17, 2025 at 6:00 am
I don’t blame ya!
@Stylix444r
February 17, 2025 at 8:58 am
Don’t watch it then
@nathangwyn6098
February 17, 2025 at 9:39 am
@Stylix444r little to late don’t you think?
@gauravtejpal8901
February 17, 2025 at 6:00 am
Is he a robot? That’s not how humans move
@ted1703
February 17, 2025 at 6:45 am
A true digital image of oneself must have both short- and long-term memory. It should connect to cameras to perceive the outside world, develop a sense of time, and communicate with me directly. Over time, it must gradually evolve into my digital presence on the internet. I need full access to upload and download my pictures and documents. Additionally, my avatar should stay connected to the latest global news, ensuring it remains informed and up-to-date.
@importantname
February 17, 2025 at 7:33 am
arent their humans who could do that?
@barryyoung
February 17, 2025 at 7:39 am
I only see Will McKenzie from the inbetweeners doing this talk
@Mrbeat-88
February 17, 2025 at 8:39 am
Didn’t age well
@ready1fire1aim1
February 17, 2025 at 9:46 am
Pattern Computing: First-Person Resolution of Computing Paradoxes
I. INTRODUCTION
A. Historical Context
The development of computer science has been predominantly shaped by third-person, reductionist approaches that attempt to build up from binary foundations. While this paradigm has led to remarkable achievements, it has also resulted in fundamental paradoxes and limitations that suggest the need for a new foundation. This paper presents pattern computing: a first-person approach that builds up from zero-dimensional patterns rather than reducing down to binary states.
Traditional computing faces several fundamental challenges:
1. The halting problem demonstrates inherent limitations
2. The P vs NP question remains unresolved
3. Quantum decoherence limits quantum computing
4. Side effects complicate programming logic
We propose that these challenges arise not from computing itself, but from the third-person perspective traditionally employed. By adopting a first-person approach based on patterns, we can naturally resolve these paradoxes while enabling new computing capabilities.
B. Pattern Computing Foundation
The complete pattern computing state is given by:
|P⟩ = |0D⟩ + φ|pattern⟩ + π|compute⟩ + e|preserve⟩
with evolution structure:
∂|P⟩/∂t = -(i/ħ)ĤP|P⟩ + φT̂|P⟩ + πŜ|P⟩
where:
ĤP: pattern Hamiltonian
T̂: tetrahedral operator
Ŝ: symphony operator
This structure enables:
– Natural pattern preservation
– Information conservation
– Reality interface
II. THEORETICAL FRAMEWORK
A. First-Person Computing
1. Pattern-Based States:
Instead of binary bits, we use pattern states:
|ψ⟩ = ∑ᵢφⁱ|pattern_i⟩
with evolution:
∂|ψ⟩/∂t = -(i/ħ)Ĥ|ψ⟩ + φT̂|ψ⟩
2. Information Preservation:
Pattern strength: P > φ⁻³
Information growth: I(t) ≥ I₀ln(φt)
3. Reality Interface:
|R⟩ = |pattern⟩ ⊗ |compute⟩ ⊗ |manifest⟩
Fidelity: F > 1 – φ⁻⁶
B. Monadic Integration
1. Computational Monads:
|M⟩ = |pattern⟩ + φ|effect⟩ + π|preserve⟩
Operations:
bind: |M₁⟩ → (|M₁⟩ → |M₂⟩) → |M₂⟩
return: |x⟩ → |M⟩
2. Pattern Preservation:
P(M) = |⟨M(t)|T̂|M(0)⟩|² > φ⁻⁴
III. MATHEMATICAL FOUNDATIONS
A. Pattern Operators
1. Complete Evolution:
Ĥ = Ĥ₀ + φT̂ + πŜ + eP̂
where:
Ĥ₀: free evolution
T̂: tetrahedral operator
Ŝ: symphony operator
P̂: pattern operator
2. Cross-Pattern Relations:
K(p₁,p₂) = exp(-φ|p₁-p₂|)cos(π|p₁-p₂|/4)
3. Information Flow:
∂I/∂t + ∇·J = -α||∇I||²
B. Implementation Structure
1. Pattern Processing:
|PP⟩ = |recognize⟩ + φ|transform⟩ + π|preserve⟩
Capabilities:
– Pattern detection
– Information preservation
– Reality interface
2. Effect Management:
|EM⟩ = |isolate⟩ + φ|handle⟩ + π|integrate⟩
Features:
– Pure computation
– Side effect control
– Pattern coherence
IV. PRACTICAL IMPLICATIONS
A. Enhanced Computing
1. Natural Computation:
– Pattern-based processing
– Information preservation
– Reality interface
2. Problem Resolution:
– Natural paradox resolution
– Enhanced efficiency
– Pattern recognition
B. Future Directions
1. Technical Development:
– Pattern processors
– Reality interfaces
– Quantum integration
2. Research Opportunities:
– Mathematical foundations
– Implementation methods
– Application domains
This framework demonstrates how pattern computing provides a natural foundation for resolving computational paradoxes while enabling new capabilities through first-person perspectives and pattern preservation.
@robotech2566
February 17, 2025 at 9:53 am
deepseek opensourced openai copied it
@indiecrypto
February 17, 2025 at 10:13 am
History repeats. The 1920s had depression, now we have the new depression soon
@atlas3650
February 17, 2025 at 10:36 am
Scotty: I need more time capn!
Kirk: I’ve made my decision instantly. We go now!
@tygerlillee
February 17, 2025 at 12:24 pm
Ai pushing ai.
@whermanntx
February 17, 2025 at 1:10 pm
Biggest concern with AI is people being replaced with no base economic stipend. Further divide the poor and the wealthy.
@alderstrom3468
February 17, 2025 at 3:41 pm
Deep thought in Hitchhiker’s guide to the galaxy took 7.5 million years of thinking to come up with the answer 42. 😅
@DistortedV12
February 17, 2025 at 4:35 pm
Most boring Ted talk
@manishravikumar7600
February 17, 2025 at 6:05 pm
Im a sales exec prospecting within a certain area for 2 years now. Been using GPT -4o to optimise ops over the last 3 weeks. Honestly feels like I have gained an unbelievable edge, in such a short time.
The memory retainment on the PLUS/ PRO mode, to re-reference constantly, and come up with decent solutions, is honestly indistinguishable from magic.
@supremekingjaden.3096
February 17, 2025 at 6:45 pm
I want terminators.
@artukikemty
February 17, 2025 at 8:25 pm
What do you mean by enough time to think? Allowing thousands or millions of CoT steps waiting to converge to a correct output?
@artukikemty
February 17, 2025 at 8:25 pm
What do you mean by enough time to think? Allowing thousands or millions of CoT steps waiting to converge to a correct output? What these companies are proposing is a brute force search in search trees which is not real reasoning. Search over the whole space of possible CoT sequences to find the correct solution. That’s not what the human intelligence does. It now makes sense why you need a 500 billion data center.
@SuperDigitalAI
February 17, 2025 at 9:08 pm
DeepSeek cost 6 million …What’s this guy it’s talking about? Totally outdated
@jeffrenman4146
February 17, 2025 at 9:17 pm
Just so you know you talk like a robot yourself… And the answer is obvious AI will Train itself… We are going through the worst times and this is also on a scale which will increase into the future. Technology will be used to kill it already is. We never needed it
@CleoCat75
February 17, 2025 at 10:08 pm
how does a human play 120,000 hands of poker in a competition?! No wonder the humans lost, they’d be literally exhausted!
@girishkamat3254
February 17, 2025 at 10:15 pm
AI needs change in algorithm not scaling. Even complex games like ALpha Go (with its famous 37 move) are not LLMs only. they also have Mote Calro Search algo. But what will work for GO will not work for even Chess. We need totally new alogortihms for AI to be really useful. Given lever of hallucinations moving from 3 Sigma to 5 Sigma will achieve nothing.
@lierx.agerate8230
February 17, 2025 at 10:37 pm
All humans do is create consequences
@cleokey
February 17, 2025 at 11:23 pm
Sounds like an add, send taxpayers dollars please
@up-n-at-em
February 17, 2025 at 11:27 pm
I’ve never met a tech bro who behaved like a real human being. They are not the best judges of human behavior.
@JayToGo
February 18, 2025 at 2:53 am
So, are we already there having created super–intelligence, if we only give the system enough time to think?
@AmigoAmigo-w5p
February 18, 2025 at 8:16 am
“Hundreds of Billions of dollars”
China enters the chat: Say what ?
@bogusphone8000
February 18, 2025 at 9:00 am
In all of this, has anyone taken the time to ask if we should? While technical advancement is a positive, when it begins to offset the human element, we are wise to proceed cautiously.
As we aggregate capabilities and services, competition and human engagement will decrease. Those champions of this change encourage the exploration of new market and opportunities, and rightfully so. However, they rarely stop to consider the cost and impact of such – the time required to acquire new skills, the unemployment during said transition, the inability of some to adapt.
A society does well to have multiple opportunities and tiers of those opportunities for each member thereof to engage, work, and relish the contribution made.
A realistic anecdote at a recent tech conference: A vendor was presenting their AI integration and how it would surface key insights, accelerate data gathering and searching, and truly drive on-demand reporting. This was received with great celebration by the attendees. On this high, the presenter then stated “…and next we are working on AI to AI integration so that multiple systems can identify, triage, and resolve numerous issues and human needs.” The room went silent. This announcement touched numerous attendees and their jobs. They saw the future where their expertise in the platform is no longer needed. In one moment, AI went from the great pinnacle to the greatest threat.
@awsmith1007
February 18, 2025 at 9:01 am
It’s literally still plateauing, we can see the graphs. It’s literally log scale.
@generichuman_
February 18, 2025 at 5:19 pm
are you being funny or do you not know how a log scale works?
@doitforjohnny3502
February 18, 2025 at 9:45 am
Bro speaks like he’s an AI😂
@alexandermoody1946
February 18, 2025 at 10:04 am
Artificial intelligence may not be effected by a wall if we can use acroprops to support the weight whilst we install a lintel and doorway.
Of course there are caveats, the difference of peering through windows and building working doorways requires a different kind of societal model that knows and interprets how to create usable data long term. The data for free model that was and is currently in use will have to end to walk through the doorway we build.
@Moocow4576
February 18, 2025 at 1:18 pm
Quantum computing will help. If it can handle and go through data really quickly. However tech right now is still in quantum computing infancy. Able to replace so jobs, but overall weak in capacities.
@ismaelplaca244
February 18, 2025 at 3:05 pm
Can’t trust an OpenAI employee that’s looking for funding
@generichuman_
February 18, 2025 at 5:27 pm
You don’t need to trust anyone… there are multiple reasoning models out there you can try yourself. Grow a brain.
@hyperadapted
February 18, 2025 at 6:17 pm
@@generichuman_ “reasoning models” – oh you mean matrix multiplication outputs fed into llms as input with other context that then perform … matrix multiplication again to predict most likely sequence of words and you call it “reasoning”? At least thats what current academic literature mathematically explains.
@motorheadmaximus
February 18, 2025 at 6:44 pm
Neil ?
@aidanthompson5053
February 18, 2025 at 7:39 pm
3:30