@dekhn it is true (I also work in the field. I'm a software engineer who got a wet-lab PhD in biochemistry and work at a biotech doing oncology drug discovery)
They really didn't solve it. AF works great for proteins that have a homologous protein with a crystal structure. It is absolutely useless for proteins with no published structure to use as a template - e.g. many of the undrugged cancer targets in existence.
I wouldn't say it's an engineering problem. Biology and pharmacology are very complex with lots of curveballs, and each experiment is often different and not done enough to warrant full engineering-scale optimization (although this is sometimes the case!).
There is a big range in both automation capabilities and prices.
We have a couple automation systems that are semi-custom - the robot can handle operation of highly specific, non-standard instruments that 99.9% of labs aren't running. Systems have to handle very accurate pipetting of small volumes (microliters), moving plates to different stations, heating, shaking, tracking barcodes, dispensing and racking fresh pipette tips, etc. Different protocols/experiments and workflows can require vastly different setups.
We don't need an army of high school sophomores, unless they are in the lab pipetting. The expensive part of drug discovery is not the ideation phase, it is the time and labor spent running experiments and synthesizing analogues.
That a UPR inhibitor would inhibit viability of AML cell lines is not exactly a novel scientific hypothesis. They took a previously published inhibitor known to be active in other cell lines and tried it in a new one. It's a cool, undergrad-level experiment. I would be impressed if a sophomore in high school proposed it, but not a sophomore in college.
I believe it. I have seen people circle parking lots multiple times or sit parked in the middle of the road waiting for a spot to open up rather than just drive another 5 minutes farther away and then walk.
I lived in a neighborhood ~5 years ago where I didn't have a dedicated parking space. I generally had to park a 10-15 minute walk away from where I lived. Many people would rather just circle for 15 minutes instead.
The WASM spec is so well defined presumably because Andreas Rossberg is the editor - and he did a bunch of PL research on extensions to Standard ML, which is famous for it's specification!
Based on the images, it is inconceivable that these are from the same run (see the dramatically different levels of TRF-S in each gel. One column/lane = one sample). This isn't something that would be included because of a reviewer - loading controls are required to meaningfully interpret the results (e.g. the data is useless without such a control).
Close - this is a SDS-PAGE gel, and you run it using proteins. The bands in the first two rows are from a western blot (gel is transferred to a membrane), where you use antibodies against those specific proteins to detect them. The Pon S row is Ponceau S, a dye that non-specifically detects all proteins - so it's used as a loading control, to make sure that the same amount of total protein is loaded in each lane of the gel.
Once you have that trained, you'll be able to publish it and dramatically improve the current SoTA!
I wish it was as easy as this - while there are known toxicophores/no-go functional groups in medchem, there is going to be a big dearth of data on non-acute (e.g. not hERG, hepatotoxicity, etc) toxicity, which is really what the question is here: what are the marginal risks/rewards from eating existing food X (since we know it's probably not acutely toxic).
I don't disagree with you, but we don't have heavy scrutiny on the existing and natural chemicals that are in the food chain from all of the plants that we eat.
I love all kinds of world cuisine, but I did not find the food in France to be better or cheaper than the food in the US, on average (and I love French cuisine). The pastries and wine though... different story!
It requires billions of dollars and a decade+ to translate academic research into a clinical outcome. Most academic research is not reproducible or robust enough to even be replicated outside of the lab that performed it - and this will only be discovered after spending millions of dollars of private funding.
I booted a hobby OS I had written in C on the family PC (off of a floppy disk!) by the time I was 13. Kids can have a lot of free time to mess around with this stuff - I would love to see what I could do now with a whole summer off.
Something like a radix trie can give you the in-order traversal and range query aspect, while still keeping some of the nice properties of HAMTs. In practice though (for my domain, scientific floating point data), I have found that b-trees and simple sorted arrays with binary search (sadly) outperform radix tries and cleverer solutions.
The post you are replying to is talking about high throughput assays for drug development. This is something actually run in a lab, not a model. As another person working at a biotech, I can assure you that screens are not just run as busy work.
I don't know if it will be useful to you, but I worked up dependency-free parsers/type checkers in Rust for the "languages" defined in most of the chapters (at least the ones I cared about) https://github.com/lazear/types-and-programming-languages
"Types and Programming Languages" by Benjamin Pierce is an excellent textbook (and approachable, even to people without formal CS education) that covers type inference and introduces a lot of the nomenclature/formalism gradually.
Definitely depends on the branching factor! There are other encodings like Eytzinger that can provide further improves over "standard" binary search too [1].
Lack of sum types and not much support for a functional programming style. I am totally uninterested in any modern programming languages (e.g. post 1980s) that only allows me to express AND (product types, e.g. records, tuples) but not OR (sum types)
Estimates of GPT-4 parameter counts are ~1.7 trillion, which is approximately 20-fold greater than the ~85 billion human neurons we have. To me this suggests that naively building a 1:1 (or even 20:1) representation of simple neurons is insufficient for AGI.
Who knows! I'm sure it depends on how accurately you want to simulate a flatworm brain.
I think current AI research has shown that simply representing a brain as a neural network (e.g. fully connected, simple neurons) is not sufficient for AGI.