Like most efforts to unmaks satoshi, the whole piece is a long exercise in confirmation bias. He pours over posts to find specific shared writing tics, then feeds those specific tics into an LLM to 'eliminate' other suspects? All because more unbiased approaches carried out by the academic were inconclusive.
The US could retain a lot of that talent if it put the same level of funding into science that China is, and remained welcoming to foreign nationals. The US has been brain-draining the rest of the world for decades with enormous benefits to us. We then led in most fields and the flywheel kept spinning. Now we are cutting research spending and closing the door, while China continues to increase its science funding year over year. The sclaes are tipping and talent will be drawn to the leading edge, wherever that is.
"Using new data which tracks US-trained STEM PhDs through 2024, we show that despite foreign nationals comprising nearly 50% of trainees, only 10% leave the US within five years of graduating, and only 25% within 15 years."
That sounds like net benefit for the US. Foreign nationals come, the US sells them (overpriced) education, they do relatively low-paid but high-value PhD research, and then most of them stay and continue to contribute to US research endeavors and the economy. This is such an enviable position, and this administration wants to close the doors? This is the secret sauce. This is what has made america great.
Again, different question. We know, fundamentally, how TMS causes stimulation/suppression of neural activity, and it does not require magnetoreception. Look at it this way: we don't fully understand how SSRI's cure depression, but we do know their primary target and that their mechanism of action is mediated through that primary target.
I think you're conflating one question with another. The "why" in question is why altering neural activity in that way results in clinical effects. It is not the "why" TMS alters neural activity.
You know the mechanism of TMS is not mysterious. It requires no magnetoreception or "stochastic resonance". It is simply inducing electrical currents to modulate neural activity. Its effects are consistent with the known laws of physics, known properties of neurons, and decades of neuroscience research.
Using it in a specialized subfield of neuroscience, Gemini 3 w/ thinking is a huge leap forward in terms of knowledge and intelligence (with minimal hallucinations). I take it that the majority of people on here are software engineers. If you're evaluating it on writing boilerplate code, you probably have to squint to see differences between the (excellent) raw model performances. whereas in more niche edge cases there is more daylight between them.
Exactly my experience as well. Started out loving it but it almost moves too fast - building in functionality that i might want eventually but isn't yet appropriate for where the project is in terms of testing, or is just in completely the wrong place in the architecture. I try to give very direct and specific prompts but it still has the tendency to overreach. Of course it's likely that with more use i will learn better how to rein it in.
Neat idea! However one issue with everyone using a digital tracker is that you can't then easily see what other player resource stockpiles and production levels are. Perhaps at the very top of the screen you could have a compact summary view that people can leave their screen at when not making adjustments, allowing other players to see the stats at a glance.
In this version, in figure 6a and b the new log scale of the IV curve looks quite linear and increasing in the range of 150 to 250 mA. I thought that it should be flat if it was a superconductor (no resistance). Can anyone explain how that behavior still supports it being a superconductor?
This looks really cool! It could be useful to be able to whitelist some sites without the 4 degree connection. For example, if i wanted to include large networks like reddit, github, stack overflow,etc. in my search results, 4 degrees may start to bring in a lot of junk/undesirable stuff into the index. Also love the idea of being able to follow or search within curated lists made by other users.
Does SD have to recreate the entire image for it to violate copyright?
As a thought experiment, imagine a variant of something like SD was used for music generation rather than images. It was trained on all music on spotify and it is marketed as a paid tool for producers and artists. If the model reproduces specific sounds from certain songs, e.g. the specific beat from a song, hook, or melody, it would seem pretty straightforward that the generated content was derivative, even though only a feature of it was precisely reproduced. I could be wrong but as far as i am aware you need to get permission to use samples. Even if the content is not published those sounds are being sold by the company as inspiration, and therefore that should violate copyright. The training data is paramount because if you trained the model on stuff you generated yourself or on stuff with appropriate CC license, the resulting work would not violate copyright, or you could at least argue independent creation.
In the feature space of images and art, SD is doing something very similar, so i can see the argument that it violates copyright even without reproducing the whole training data.
Overall, i think we will ultimately need to decide how we want these technologies used, what restrictions should be on the training data, etc, and then create new laws specifically for the new technology, rather than trying to shoehorn it into existing copyright law.
I agree, it is miraculous. the fact that dna is ultimately storing all the info for specifying neural circuits that robustly support such complex innate behaviors (often with very little post development tuning / learning) , that to me is mind-blowing. And butterfly behavior is one thing, but what about innate detection of predators in some visual systems? How do you encode a snake detector in dna?
I guess it depends on how accurately you're thinking about those functions being approximated. Neurons have a natural nonlinearity to their input-output (transfer) function, most obvious of which is the action potential threshold. Biological neurons have a saturating nonlinearity because there is an upper limit on their firing rate, but in certain regimes the nonlinearity of a single neuron could easily look qualitatively similar to relu or a (non-negative) tanh.
Also the nonlinearity only needs to be differentiable because ANNs are trained with gradient descent. With other more biologically plausible learning mechanisms, this might matter even less (or have other constraints / requirements)
I am also not en expert in this area, but I agree with this assessment. This looks like classic anomaly hunting rather than careful science, but I would love to be proved wrong.
I would agree that resolution is usually considered a linear unit (eg X cm/pixel). The improvement still looks great, but I also find the 9x a bit misleading.
Interesting approach! I'm curious about your sample image comparison. You mentioned in response to another comment that it was taken by a drone and the image on the right was downsampled by a factor of 3 to mimic the resolution of currently available satellite images. However, the image on the left has artifacting that looks reminiscent of deep learning-based methods for deblurring, denoising, or super-sampling. Is the image on the left actually a raw drone image? How much of the stated 9x improvement (actually 3x) is hardware/sensor-based vs software-based?
I think using citation count as a metric for scientific importance by definition incorporates a ton of hindsight bias. Less cited papers are sometimes poorly executed research, but often they also just represent dead ends and uninteresting avenues that would've eventually needed to be explored. they don't form the foundation for lots of new research, but collectively they are important for directing things in the right direction. In that sense the citation count undervalues their collective worth. I don't buy the idea that we could get rid of the vast majority of them and the high impact work would be done just the same.