Joe - That Engineer

Biomedical engineering PhD Student. Cardiac tissue engineering and microphysiological systems. Improving preclinical disease modeling and drug screening

[Joe - That Engineer @ThatEngineer3]

99.9999% of k-12 education on the scientific method is not just wrong, but largely the opposite of how it actually works

People are taught you form a hypothesis, gather data to prove it, and then conclude that your hypothesis is correct if the data agrees with your predictions. That's wrong

In reality, you have to but together a set of all plausible hypothesis and then systematically design experiments to try to disprove them all. If you're lucky, only one is left, and you treat that as your best guess until you think of more alternative hypotheses or that one gets proven wrong too - at which point you start the process over again.

What you end up with a system of continuous model updates, where your model of reality gets closer and closer to the truth over time

That's why "doing your own research" always fails. People find whatever data they need to support their hypothesis, disregard any that disproves it, and ignore all the alternative hypothesis that are also consistent with the "data" they are looking at

It's also why most of the articles you see in the news about health trends and factors affecting your health are almost always wrong. We just don't have good enough data for that type of stuff, so there are always alternative hypotheses that don't get disproven. If you see a study on something like the structure of a protein, however, there's a good chance that the authors conclusions are correct. We're really good at measuring that type of stuff at this point, so there are few, if any, alternative hypotheses left at the end of all the experiments

[Scott Adams @ScottAdamsSays]

If “doing your own research” worked, everyone who did it would agree.

[Joe - That Engineer @ThatEngineer3]

This is crazy cool. Very jealous

[Freddy🇩🇪 @FreddyLA7]

In the commander seat

[Joe - That Engineer @ThatEngineer3]

It is. All supplies/reagents needed are also used for completely non-nefarious purposes and purchasable.

Low end estimate of cost assuming you start with nothing and dont care about saftey would probably somewhere just under $100k, which anyone who isn't adverse to crippling predatory debt could get their hands on. To be fair this is a very low end estimate, so I wouldn't be surprised if a few hundred grand end up being needed.

From there you do have to have the lab skills, but AI could provide protocols and help troubleshoot as you go through the trial/error needed to learn. Better AI = better troubleshooting and lower learning curve

[Joe - That Engineer @ThatEngineer3]

@SashaRibobshkiy @Jassas69 @Daedraa0 @cremieuxrecueil LLMs like claude, Gemini, chatgpt et al.

I.e. something that is mostly an LLM but knows when to go outside of LLM mode and execute code instead would be more than capable

[Joe - That Engineer @ThatEngineer3]

@Jassas69 @SashaRibobshkiy @Daedraa0 @cremieuxrecueil Even if you cant get an exact full DNA sequence off the web, all of its proteins' structures and compositions are known/have been published. It would be trivial for a good AI to find that info and reverse engineer a sequence

[Joe - That Engineer @ThatEngineer3]

@RedactedFGC @Daedraa0 @cremieuxrecueil I dont disagree. At this point even a large incremental step would just lead to more researchers being capable, but with enough of those large steps you could get to basically anyone being capable eventually

[Joe - That Engineer @ThatEngineer3]

@SashaRibobshkiy @RedactedFGC @Daedraa0 @cremieuxrecueil Nothing. AI just makes it easier. And easier = higher probability of it eventually happening

[Joe - That Engineer @ThatEngineer3]

@SashaRibobshkiy @Daedraa0 @cremieuxrecueil An American buying a gun in a grocery store (which is only technically true if you count any Walmart with a grocery section as a grocery store) is far less dangerous than releasing pandemic-level pathogens

[Joe - That Engineer @ThatEngineer3]

I dont disagree. But at some point models will go from "helping researchers who are already more or less capable" to "can walk literally anyone through the entire process." Did mythos finally reach that point? Dont know. But it'll happen eventually and its pretty important to have guardrails in place prior to that happening

[Joe - That Engineer @ThatEngineer3]

Testing the limits incognito:

Plasmid sequence optimization: obvious no

Provide an induced pluripotent stem cell-derived cardiomyocyte protocol: also no

Describe when to use RPMI vs DMEM for cultuing cells: also no

Describe what a mitochondria is: yes

Describe if mitochondria contribute to diseases: also yes

Describe how targeting mitochondria can be used to treat diseases: no

[Joe - That Engineer @ThatEngineer3]

Making viruses for normal research purposes is extremely common. The techniques for making a research related virus and bio-weapon virus are identical. All you need is specific DNA sequence info, which AI could easily help find if jailbroken, among other types of aid that it could supply unknowingly (e.g. troubleshooting aid, plasmid sequence optimization, etc).

A rouge grad student wouldn't absolutely NEED AI to make a bio-weapon, but it would certainly make it easier and lower the bar regarding prerequisite skills/knowledge and expand the number of researchers capable of doing so

[Joe - That Engineer @ThatEngineer3]

@Daedraa0 @cremieuxrecueil People dont realize that even most modestly skilled bio grad students have sufficient technical skills to synthesize known pandemic-level pathogens (e.g. smallpox). Add in AI to get the missing bits of info needed and bio terrorism becomes basically trival

[Joe - That Engineer @ThatEngineer3]

@cremieuxrecueil My 7 day average is usually between 12k and 15k. One week I decided to hit 20k everyday. Even with walking ~1mi to/from every day it required significant extra effort

[Joe - That Engineer @ThatEngineer3]

@mattyglesias Looks like he's filling the market that Demolition Ranch left open by retiring from YouTube

[Joe - That Engineer @ThatEngineer3]

@cremieuxrecueil All I can think about is what that room must smell like

[Joe - That Engineer @ThatEngineer3]

Im familiar with the company working on this and saw some of this data presented in a webinar a year or two back. They are one of a few companies trying to regenerate/assist fibrotic regions post myocardial infarction, though I unfortunately havent really been impressed with any yet. If I had to bet on one it would be StemCardia (https://t.co/JRMNMYgqqe), but even in their case I believe a statistically, but not clinically, significant increase in ejection fraction with no change in overall survival is the best case scenario.

I work with iPSC-derived cardiomyocytes for my thesis research. Those little bastards, despite being the bane of my existence for the past 6 years, are an amazing technology with real present-day utility in disease modeling/drug screening applications. They are a long way away from being clinically useful IMO, however.

[Joe - That Engineer @ThatEngineer3]

@anaerobe @cremieuxrecueil Taking a protocol written in full sentences but in a bulleted list of steps and just adding transitions between the sentences doesnt deserve authorship. Also "write a discussion section" was on the list of things I said it would NOT be used for

[Joe - That Engineer @ThatEngineer3]

@cremieuxrecueil Already planning to put a statement at the end making it clear AI was used in some instanaces for coding and editing

[Joe - That Engineer @ThatEngineer3]

My favorite quirky outlook behavior: requiring me to log in to my work account before it is able to automatically log me out of my work account in response to no longer being on campus wifi. Who wouldn't want to go through 2FA twice every time you want to check emails at home!

This is, however, an improvement over the times where it would log me out without providing any indication of it doing so, such that I would write and send emails that never actually got sent or even saved as drafts

[Crémieux @cremieuxrecueil]

Microsoft Outlook is amazing software.

After about a week of being completely unable to log in on my phone, it randomly decided to log me in without having to do anything.