“Chemical analysis was performed using a Thermo Scientific Niton XLt-900 XRF Analyzer equipped with a GOLDD X-ray fluorescence detector and real-time digital signal processing at a speed of 80 MHz. The XRF measurements were taken in the laboratory, using a Thermo Scientific SmartStand, ensuring consistent positioning. Since the Niton pXRF is programmed to utilize predefined matrices provided by the manufacturer, the “General Mining” matrix, which includes the relevant metallic and non-metallic elements, was selected for the current research. Calibration was performed using an internal calibration system defined by the manufacturer and validated using a known standard (SRM 2780).”
Translation: “We used a closed-source proprietary method by a manufacturer and validated it on a sample the manufacturer recommended.”
XRF generates spectra which represent atomic abundance (eg K-, L-, and M-line emissions). However, these are not international standard units. Laboratory units translate from spectra to mass fraction through meticulous preparation of standards. Handheld units often use a secret sauce algorithm to do this to keep analysis non destructive. I’m skeptical of the results - but what’s more important is that these results are not fully reproducible. If the company goes belly up, or less dramatically changes/updates the proprietary algorithms used to create mass fraction estimates, the reported results of this study won’t be replicable even with the same samples.
Not billions, but I was in a similar position. What saved me was the GPL license on the open source code and hiring a lawyer that kept my ownership of any software I wrote (though at a reduced hourly for them) and patenting new ideas connected to the project. When it came time for the company to scale up, I couldn’t have been in a better position.
The fact that so much of Çatalhöyük remains unexcavated is one of the reasons I am skeptical of so much speculation on its social structure. A palace could be under the vast majority that has yet to experience a trowel or brush. It, and sites like it, are so early that it is helpful - though extremely difficult - to forget the 9,000 years of future you know will happen to be able judge it on its own terms.
Thanks! First thank you for all the work to develop a math LLM. Those prompts are helpful. I shrank the weights down to f16 to fit on my system from f32 using the procedure at ggerganov/llama.cpp, then tried some of those prompts. From line 34:
> Problem:
> If $\det \mathbf{A} = 2$ and $\det \mathbf{B} = 12,$ then find $\det (\mathbf{A \mathbf{B}).$
> Solution:
> We know that for a matrix \mathbf{M}, the determinant of its inverse is given by $\frac{1}{\det \mathbf{M}}.$ We also know that the determinant of a product of matrices is equal to the product of their respective determinants. Therefore, we have:
\begin{align} \det (\mathbf{A}^{-1}\mathbf{B}) &= \frac{\det(\mathbf{AB})}{\det\mathbf{A}}\\ &= \frac{\det\mathbf{A}}{\det\mathbf{B}} \end{align}
> Therefore, we have:
> $\det(\mathbf{AB}) = 24
The f32 answer:
> We have that $\det (\mathbf{A} \mathbf{B}) = (\det \mathbf{A})(\det \mathbf{B}) = (2)(12) = \boxed{24}.$
> Final Answer: The final answer is $24$. I hope it is correct.
Final answer is the same, which is encouraging for quantization to expand hardware options.
can you please ask it if algebraic numbers are closed under addition? chatgpt4 just told me no, and then i asked it again and it said yes, and then it apologized profusely for saying no.
I assume that this primarily benefits games and not any deep learning right? The most attractive aspect of Mac M1 is the huge memory boost. Might not be great for training due to the inability to distribute across multiple cards, but it makes for a great inference engine for stable diffusion, llama, and other large models.
There are two modern cross-platform GPGPU standards that Apple Silicon can theoretically use or implement - SYCL and Vulkan Compute.
SYCL is Khronos Group's vendor-neutral, high-level programming framework. Application support is limited, but hopefully with Intel's backing, the situation would gradually improve. Meanwhile, Vulkan Compute sidesteps the entire headache with compute shaders. But I'm not familiar with it in terms of application support.
SYCL can be implemented on top of OpenCL and OpenCL's SPIR-V extension. It soon turned out that this route is unfeasible due to prevalent vendor lock-in that's not going to change anytime soon, so it has largely been abandoned by everyone else but Intel and Mesa. Right now SYCL is usually implemented by backends to GPU vendor's respective APIs, like ROCm, HIP or CUDA. Doing the same for Metal would be very challenging.
Mesa already has experimental support of OpenCL w/ SPIR-V on Intel and AMDGPU, so theoretically it can be extended to Apple Silicon. Difficulty of implementing OpenCL's SPIR-V extension should be comparable with Vulkan compute shader (which also uses SPIR-V). However, currently OpenCL on Apple Silicon is entirely unsupported. The last time I checked, it's on the roadmap.
The only problem with cross-platform standards is they are never performance portable unless they're so high level someone already their primitives have already implemented the algorithm X different ways for you already.
For any low level performance programming you need to code to the specific microarchitecture, so the pros of a single programming language/library are limited (you're not getting any code reuse that isn't available in the top level non-hardware C code anyway) and often outweighed by the ability to take advantage of the vendor's dedicated extensions provided by their preferred programming mechanism.
This issue was well modeled by OpenCL, which never really caught on for programming Nvidia GPUs for this reason.
Correct. You need CUDA, or ROCm, MPS (native to macOS) backends for running deep learning. I found it relatively easy to train some Pytorch model on beefy server with CUDA and running interference on my Macbook Air.
The compute shader portion is a good step but it's still not going to provide the interfaces most of these deep learning tools expect.
That said eiln wrote an ANE (Apple Neural Engine) driver which enables using the dedicated hardware for this instead of the GPU. It is set to be merged into linux-asahi in the future.
> They found that the model relied on the style of language used in papers to judge if a study was replicable.
I think the failure here is to train the model on published results. I’ve worked with scientists who could write and infer a marvelous amount of information from shit data. And I’ve worked with scientists who poorly described ingenious methods with quality data. The current academic system incentivizes sensationalization of incremental advances which confirm previously published work. I’m not in the least surprised at the manuscript level that replication would fail.
The proper way to do this would be to log experimental parameters in a systematic reporting method specific to each field. With standardized presentation of parameters I suspect replicability would improve. But this would present an near impossible degree of coordination between different research groups. But it would be feasible for the NIH or NSF to demand such standardized logging as a condition of grant awards of a certain size.
Have you posted the drafts to Archiv? If so please share. If not, please do so. I don’t like the idea that a couple reviewers block knowledge transmission permanently.
Thanks. Although sadly I think that might have been what happened. After the rejections, I moved into a startup and lost track of where the papers are. :(
On replication, it is a worthwhile goal but the career incentives need to be there. I think replicating studies should be a part of the curriculum in most programs - a step toward getting a PhD in lieu of one of the papers.
Fear of the frontier.. that's why instead of people getting excited to look for new rtsp superconductor candidates, we get a lot of talk downplaying the only known one. Strong link vs weak link reminds me of how some cultures frown on stimulants while other cultures frown on relaxants.
This is vastly more preferable than our current approach of raising children as robots.