timster6442

Thanks for the response, I think these are some great points. I originally studied electrical computer engineering as that was the most interesting to me but by happenstance went on to work in a neurobiology lab and felt that I had the ability to contribute a different perspective/skill-set to the lab/field. Through research and witnessing human diseases, I now feel more strongly about contributing to understanding of human disease mechanisms.

Not to be combative, coming from genuine interest, I'd like to ask you why we should care? I study biology, and can think of a plethora of topics that we can study about human physiology that we don't understand and if we did would more directly lead to treatment of human diseases. That is not to say we need invest all resources towards this frontier but in the optimal scenario we must weight resource allocation to the potential usefulness of the discovery.

I believe their argument is weak but, I think it is a fair assumption that the average ride share driver is safer than the average driver. That is to say the average driver is unsafe due to the fact the average driver is the set of all people which includes the set of drunk drivers whereas we can generally assume that set of rideshare drivers don't, especially while on the clock.

I agree that these cases are ridiculous to blame on the waymo entirely, but to give logic to their reasoning here you're unsure if on average the 1.3 fatalities per 100 million miles (FPMM) may also be on average ridiculously not the drivers fault as well. Perhaps if you were to throw out all ridiculous cases you'd have an average of CDL's having a 0.3 FPMM vs Waymo's having 0 FPMM.

The nuance you describe at the end is the better interpretation: fatalities per mile driven is a useless metric since it's such a rare occurance and incredibly circumstantial that it cannot be used as a meaningful comparison metric for safety.

While it is a cool technical advance to see, its use case is probably niche. One of the biggest things people say about photography is lighting. In general, if you can get more photons onto your imaging plane, you will have a better picture. This is why phone cameras, despite advertising 100s of MPs, still fall short. For the same field of view and capture conditions, fewer photons are collected because phone sensor sizes are much smaller. This is a physical limitation that cannot be overcome.

Now given that they already have a lower photon budget, you now want to add aperture control, which when stopped down physically reduces the amount of light from the scene that reaches the sensor. Thus you need to compensate for these lost photons somehow, so gain will probably have to increase quite a bit, but then you get noisier, grainier photos. This then leads to the suspicion that you will need some sort of AI denoising or computational processing to recover some of that lost image quality. Though of course you can comp in other way such as exposure time (they can make big wins here with their stabilization) and tons of lighting.

Also a phone already starts with very deep depth of field because again smaller sensor size thus smaller focal length etc. Most images at their lower f stop is already in focus. So the ability to reduce aperture or simulate shallow DOF is trying to give users more of the creative control that larger sensor cameras naturally have (need to manipulate because of their larger sensor size) and doesn't make much sense to me.

Am neurobiologist and I don't buy the "full" argument, its much more likely a brain aging thing.

The brain, before your born, in GW25 (gestational week 25) has finished growing all the neurons you're basically [we can talk about this later perhaps] going to have for the rest of your life. At GW25 current estimates say you have ~86 billion neurons. Now the timing for this next part is a bit unsure but your body doesn't need anywhere close to 86 billion neurons, so at some point as early as early adulthood you start to lose 85,000 neurons per day pretty steadily until you die. Now neurons are not the end all be all because connections are potentially what really matter but that is to say that the neurons that you have after GW25 are the neurons you basically have. Now going onto synaptic connections where things start to matter more. Now synapses form and then get pruned all the time its natural. But the rate of formation and the rate of pruning is not the same at all times of life. Now from a raw number of synapse scale we see a tipping point at around 16-26 years old (debated hence the big range) where the number of synapses start to go down, indicating that the rate of pruning is now outpacing the rate of formation. [It does seem however that the dysfunction of the rate of pruning i.e. not enough ends up with consequences like schizo or asd {autism}]. There is another factoid that the rate of decline seems to stay relatively stable until you hit ~60 and then synapse related decline becomes much more noticeable and we start to think of synapse loss as exponential.

Now about new neurons after GW25 is a whole topic in of itself... heavily debated but I won't get into that history, the most accepted viewpoint is that it does exist in the hippocampus [other regions as well] (really cool work with c14 carbon dating from atomic bomb {2010} and perhaps less cool new sequencing methods give evidence {2025/2026}). Note estimates of how many are born are comparatively low-ish, 500-1000 neurons per day.

That begs another question however which is why and what do they do? Final interesting part is that when you stop the mouse hippocampus neurons from dividing, [unethical to do in humans :( ] distinct representations of experiences start to look similar and overlap. I also know if you ablate neurogenesis in mouse nasal cortex I think mice lose the ability to form new sensory sensations all together.

Disclaimer, the evidence for function of new born neurons is actually pretty low only a few studies have tried this so its no where close to accepted and thus far far from textbook standard so take it as you will.

Firstly the underlying worry here is about privacy which hinges on the fact that AI companies are stealing ideas in the first place. Stealing from your customers is an incredibly bad business model and I think if they were to steal IP (intellectual property) from researchers mathematicians or computer scientists would be first.

Now why I think biology is safer: 1) Producing novel biology still has to be done in a lab. It requires laboratories, equipment, experimental protocols, trained personnel, regulatory and safety infrastructure, and often substantial institutional organization all of which there is no indication they're heading for. Also I disagree that lab hardware is near a "soon state" where labs can be full autonomous, (liquid handlers are really good at niche tasks but lack any type of experimental general ability [not AI-bounded], especially for in vivo work where its footprint is non-existent). Even the most automated Labs I know where robots do 80% of experimental work, they still have grad students to carry out that last 20% and to oversee.

2) Even if AI could do the pipeline it's not worth it for AI LLM companies to dedicate capital to it currently. A lot of biology research itself doesn't produce a sellable product, in fact most of it never does. It seems currently and for at least the next couple years at least, AI capital is best spent growing compute to research better models, train better models, and sell inference.

I'm in academia (biology but highly computational) and I would say opinions on AI are quite polarized. Some professors in the department equate not using AI as lost productivity. Contrarily some professors abhor the idea of even using AI at all. For us (biologists) it's less of an issue because we have no fear of openai or A/ publishing a biology paper. Though even people I known in physics, data science, or computer science still heavily use AI.

Our university has agreements that stipulate that our institutional accounts cannot be used to train AI models and certain research groups have differential model access.

Further from academic journal sense there is mixed feelings. I once was able to meet with a senior journal editor (general non-medical high IF journal > 50) who claimed that if they think something is written by AI they wouldn't consider it. Yet another high IF journal said it was completely fine if something was written by AI. About a month ago I reviewed a paper by yet a different high IF journal and in big bold red letters it said I was not allowed to feed any part of the paper through AI (even if it was locally ran) but you could ask it to rephrase text that you wrote.

Am biologist and keeping genes that are "advantageous" is the most common trap that people fall into about evolution. Evolution isint actually about keep advantageous traits the real idea is about differences in reproductive success, not about evolution systematically preserving only traits that are beneficial to the individual.

In fact Darwin's theory of evolution is kind of outdated as well. We now understand genetic drift and things like the neutral theory of molecular evolution from moto kimura and the nearly neutral theory of evolution from tomoko ohta may give some sort of explanation to polygenic nature of adhd.

He's a legit researcher though I haven't seen his work on youtube. In terms of how accurate brain scans are at diagnosing adhd you can look at recent meta analyses (gold standard of evidence based medicine pyramid).

In a 2024 meta-analysis of MRI/fMRI machine-learning studies for ADHD, the average pooled sensitivity was 74% (the ability to correctly identify people who actually have ADHD) and the average pooled specificity was 75% (the ability to correctly identify people who do not have ADHD), that also means that there is a 26% false-negative rate (miss/Type II error) and a 25% false-positive rate (accidental diagnosis/Type I error) [https://doi.org/10.1016/j.jad.2024.03.111].

Though there is no set standards but under 80% is pretty low for a clinically useful test but can still be used in some conditions such as to supplement diagnosis. To give some numbers the rapid antigen covid tests had 72% sensitivity and 98.9% specificity. (note the lower sensitivity was acceptable at the time but the CDC guideline was for symptomatic people to retest 48 hours. Viral load and other staging factors could all confound.) [https://doi.org/10.1371/journal.pmed.1004011]

From a molecular neuroscience perspective, for both ADHD and ASD (autism) you're unlikely to find any pure biomarkers as clean as a disease like huntingtons disease whose mechanism is clearly understood as extra CAG repeats on the HTT gene (a clean mendelian disorder). ADHD is defined by behavior/cognitive symptoms and most all behavior/cognitive diseases are multifactoral, which ADHD is. WES/WGS and GWAS has been done and for familial ADHD and unlike ASD there is mostly no clear monogenic mutation (a single gene that would explain the phenotype) [note monogenic mutations are usually a very small subset of cases but give scientists a look into what pathway/mechanism is causing a multifactoral disease]. Thus it is not for a lack of looking classically, and deficit is speculated to likely manifest as a neural circuit organization or synaptic issue (though the latter is more closely linked to ASD). This is currently impossible (basically) to measure in alive human brain as it requires nanoscopic imaging, note how challenging the fly map was (incomplete IMO as well).