neural nets.

We're looking to sell papercode. Papercode is an complete end to end ML preparation platform built for students, featuring AI-powered mock interviews, coding practice, and tools to help candidates prepare for technical hiring in ML riles A few numbers: • 3,000+ users • ~$100 USD MRR • Fully built and actively used We've loved building it, but our focus has shifted to other projects. Rather than letting it stagnate, we'd rather see someone take it further. If you're interested in acquiring Papercode, my DMs are open, or email me at cneuralnets@gmail.com

drop france vs spain preds

@RadhaDasi__ Which one?

@cneuralnetwork Even I am trying to get my brother give olympiads ,but I think there is no definite resources and things that i know ,how to approach and guide him.Any help?

sof exams used to be so easy as well complete medal farming

a farmer in a rural part of india gets to use claude for his farming and claude helps him - very nice thing just to get called racist stuff from random foreigners arabic shows the same plot as well, but let's go - let's be racist against indians woo hoo i will get engagement!! BE NICE.

what are these real estate prices 48k for one room 😭

@V900Vishal it's the same everywhere

NSUT students, can someone please confirm this once?

dawg how tf is indian football doing so good 😭😭 hats off @THE_RanjitBajaj sir

I got selected in the assignment round of a startup recently. I was not interested to move ahead ~ so I did this I got gpt 5.6 sol max to solve the assignment and I sent it to the HR team of that startup (with disclosure that it's done by AI) the assignment was so good they didn't believe that it was completely made by AI end to end and told I must have given inputs :P it's joe-over

a sovereignty gap in science some opportunities appear first as markets. others as pressure. "sovereignty gap" (h/t @abhayvenkatesh1) shows up when society requires an essential function that its established institutions cannot perform without exceeding their mandates. the need is justified but no real existing actor is structurally equipped to meet it. scientific r&d is revealing such a gap consider this: the alphafold db contains more than 200M predicted protein structures. the protein data bank (global archive of experimentally found macromolecular structures built over more than five decades) contains roughly 256k structures. nearly 800:1 difference. frontier AI systems can populate possible worlds much faster than science can "verify" the real ones. we have reasoning at silicon speed. the physical labs still move at the speed of instruments, calibration, queues, reagents, grad students, etc - - the physical reality. the verification deficit our scientific institutions were designed for a world in which hypotheses arrived at ~human speed. experimental execution + judgment + verification were always scarce. but the production of credible candidates and the capacity to test them operated on roughly the same human timescale. that balance is breaking. models are now starting to generate candidate molecules, materials, mechanisms, experiments far faster than we can evaluate them. this bottleneck is ultimately challenging us to scale production of chips, new materials, etc. we are also bottlenecked by experimental judgment - basically deciding what deserves to be tested, performing the test reliably, interpreting noisy results, and using each result to choose the next experiment. physical reality is the (often) expensive oracle. without a corresponding expansion in verification capacity, AI can flood science with plausible claims that existing institutions cannot absorb. the machinery for generating possibilities → becoming computational. the machinery for establishing truth → largely artisanal today. good old verification deficit? everyone internalized verifier's law (@_jasonwei) --> ease of training AI to solve a task is proportional to how easily the task's output can be verified. AI accelerates fastest where the verifier can also be made computational. this has been happening for code, path, some part of physics. these verifiers are not perfect but cheap enough to run repeatedly and good enough to guide the next set of rollouts. this is good for scaling: a system proposes an answer, receives a reliable signal, updates the strategy, and tries again. millions of "simulations" can happen without waiting for the physical world. materials science, biology, chemistry, manufacturing all eventually reach a point where the final verifier is physical reality itself. no amount of simulation can substitute for synthesizing the material, observing the biological system, etc. also- the physical world does not often return a clean scalar reward. we end up with partial measurements, delayed rewards/outcomes, conflicting instruments, net new failure modes that the models don't know exactly know how to reason about. a useful scientific verifier therefore has to answer several different questions: -- execution: was the experiment performed as intended and is the measurement trustworthy? -- scientific validity: did the intervention produce the claimed effect with what uncertainty and under which conditions? -- process robustness: does the result survive variation in operators, equipment, raw materials, batches, and environmental conditions? -- manufacturing validity: can it be produced at the required throughput, yield, cost, safety, reliability, and quality? making the verifier computational does not mean replacing physical reality with simulation. it means making physical evidence reliable and machine-readable enough that a system can learn from it. this is where today's scaling laws meet physical reality. compute can continue scaling, but scientific performance will not scale with it unless the feedback loop scales too. the so called "institutional vacuum" universities have been organized around the incentives of academic science: principal investigators, grant funding, publications, and training the next gen researchers. national labs operate within defined missions and long planning cycles. industrial r&d orgs are optimized to advance existing pipelines. frontier AI labs are designed (currently, although this is changing) to scale computation in digital environments. process engineering and manufacturing teams are optimized around throughput, yield, cost, quality, and reliability. they often face the most important physical failure modes but only after a discovery has already moved far downstream. none of these institutions is failing at its stated purpose. the problem is that the new requirement sits between their purposes. the orgs responsible for intelligence don’t often control experimental throughput. the orgs responsible for experiments don't often learn as a single compounding system. the process engineers responsible for scale up and the institutions responsible for validating knowledge were not designed for a world in which machines can generate scientific claims continuously. society increasingly needs the ability to turn machine-generated possibility into reliable physical knowledge and then into reproducible industrial capability at machine relevant scale. from manufacturing sovereignty to discovery sovereignty governments and nation states are already revealing the pressure through their actions: chips, medicines, energy, advanced materials, industrial processes, etc are increasingly understood as foundations of economic and national security. but these policies largely address the capacity to manufacture known technologies. manufacturing sovereignty ultimately depends on discovery sovereignty. a country that can manufacture yesterday's technology but cannot discover tomorrow’s remains dependent. the deeper measure of sovereignty is if it can repeatedly create the knowledge from which new strategic technologies emerge. but discovery sovereignty without scale-up not enough. a country that can produce promising papers and prototypes but cannot turn them into repeatable processes remains dependent on institutions that can. closing the loop is important the decisive scientific systems will connect reasoning, simulation, experimentation, measurement, process engineering, manufacturing feedback, and learning into a continuous loop. they will also become better at choosing what to test because they learn from the outcomes of previous tests. early demonstrations show what happens when even narrow versions of this loop are closed. every experiment in a closed system performs two jobs: 1/ evaluates the current idea and 2/ improves the system that selects the next one. every failure should become reusable insight. physical feedback must be earned. experiments take time. instruments behave imperfectly. etc. an institution that closes this loop early develops the ability to decide which data should exist. better experimental judgment produces more informative experiments. more informative experiments produce better predictive models. better models reduce wasted physical work and expose new regions worth exploring. the entire system improves together. this creates a form of advantage that cannot be reproduced by simply licensing the same model. compute can be purchased quickly. years of aligned experimental feedback, operational knowledge and accumulated failures cannot. the gap between institutions that learn from physical reality continuously and those that consult it episodically will widen. scale up is another experiment scientific discovery is often presented as if the work ends when a new property is demonstrated in a lab. in practice that is usually the beginning of a second and equally difficult search problem. scale changes the system. heat and mass transfer behave differently. mixing becomes uneven. impurities accumulate. at lab scale, a one-in-a-million failure may never appear. at massive manufacturing scale - it appears constantly. this is why process engineering should not be treated as a downstream implementation detail. it is another layer of scientific verification. it tells you whether a phenomenon is good enough to become a product, process, or strategic capability. the feedback should also travel upstream. if a promising material cannot be synthesized consistently, that should change which materials the discovery system proposes. if a reaction requires uneconomic purification, that constraint should shape the next search. if a manufacturing defect appears only at scale, the scientific model should learn from it. the best candidate in the lab is not always the best candidate in the world. a slightly lower-performing material that is stable, cheap, safe, and manufacturable may be far more valuable than a fragile laboratory record. the full loop therefore cannot stop at discovery: reasoning → simulation → experiment → measurement → process development → pilot scale → manufacturing → feedback a system that learns across this entire trajectory is doing more than autonomous science. it is learning how to turn scientific possibility into industrial reality. a new measure of scientific power for most of modern history, scientific capacity has been measured through inputs: research budgets, universities, laboratories, publications, patents, and trained scientists. the next era will demand a different measure: how efficiently can an institution convert compute, autonomous experimentation, process engineering, and manufacturing feedback into verified improvement at scale? how much uncertainty does each experiment remove? how quickly does the full system learn? does performance improve as more computation and physical feedback are added? can it repeatedly transform questions into validated results? how many lab improvements survive pilot-scale production? how quickly can a manufacturing failure update the next scientific decision? can the system optimize performance and manufacturability together rather than discovering one and confronting the other later? scientific power has historically been defined by the knowledge an institution possesses. it may soon be defined by its learning rate. that learning rate will influence who develops better materials, energy systems, medicines, manufacturing processes, physical infra etc. it will decide which nations remain dependent on external knowledge and which can create strategic tech for themselves. AI has made possibilities abundant. the first gen of scientific AI learned from the record of human discovery. the next generation will learn from the consequences of its own experiments. the sovereignty gap in science is the missing institution that can make scaling laws survive physical reality -- from the first experiment to the billionth manufactured unit. we are spending time with a small number of R&D leaders, process engineers, manufacturers, researchers, computational chemists and material scientists who feel this gap firsthand. if you're interested in evaluating whether a closed learning loop can improve real R&D outcomes, we would like to compare notes.

TOM CRUISE AND SANDRA HULLER???????????? count me in

@viv_e_k yes

@cneuralnetwork I think both of them were pretty active on quora

I had two seniors who were crazy cracked one of them was single digit in everything and 2x gold medal in icho - recently found out she is an swe at vc firm in usa rn another of them was a madlad, never studied bio, got interested in bio in class 11/12 - but he didn't give neet, he got all exam double digit rank, joined top IIT cs, gave neet next year air 2, and he was one of the leading helpers during covid times for vaccines all over rthe world

@Zer017o tution

@cneuralnetwork Senior?? School, college or anything else

@uvyie Both of them have very public profiles :)

@cneuralnetwork *looks at larp*

5 kids came #1 in the world unde rthe indian flag and the first issue you find is a regional issue 😭😭 what a sad thought process

jk it was the trailer I'm watching evil dead burn

#nw odysey

😊😊😊

What are some must visit restaurants in Blr?