Siebe.

To make LC more attractive to pharma we need 1. A surrogate biomarker 2. An international patient registry ➡️ reduces costly recruitment 3. Public precommitment by FDA for accelerated pathways for the first effective drugs Explained: atelfo.github.io/2023/12/23/bio… #UniteToFight2024

@nikitsh @yarbatman @michaeljswalker It is not by definition a war crime. Only if disproportional or not serving a clear military objective. It is a war crime if it's economic warfare. And it's bad policy I agree there

@yarbatman @michaeljswalker Besides being pushed towards the edge of an abyss, bombing power plants is a war crime, a collective punishment and is by definition a war crime, and it is also a failed policy, as is clear from the start.

In a discussion with @michaeljswalker on Friday, I said that Trump was being presented options to strike civilian infrastructure in Iran. The aim would be to cripple Iran’s economy and to make it harder for Iran to persist in the war. I explained that hitting Iran’s critical infrastructure might shorten the war, but it would also intensify it. For that reason, that overall damage done to region and the global economy could end up being far worse during a shorter war. If Trump hits a major power plant, Iran will respond by hitting utilities in the Gulf states and Israel—including power plants and desalination facilities—and will count on the fact that it has to hit fewer targets to have a bigger impact. Iran has nearly 500 power plants, Israel has around 50. The largest Iranian plant, the Damavand Combined Cycle plant, accounts for about 4% of Iran’s total capacity. Israel’s largest plant, Orot Rabin, accounts for 20% of electricity production. The fact that a US president is posting a message like this on social media represents a total breakdown in the systems that are meant to ensure the US fights wars intelligently and judiciously. The commander in chief is not of sound mind, no one in his cabinet is willing to admit it, and we are being pushed towards the edge of an abyss.

@riendebock @RobWust @MarcBonten Njah niet echt nee, of in ieder geval niet eentje die probeert via de standaard weg market approval te verkrijgen. Wel een patent geloof ik? Anyway mijn punt is meer algemeen natuurlijk

@PatientPersists @RobWust @MarcBonten Deze kliniek is toch helemaal geen farmaceutisch bedrijf?

De advocaat van de lidocaïne-kliniek heeft het maar druk. Ook @MarcBonten en Frits Rosendaal kregen een dreigbrief met belachelijke eisen nav inhoudelijke kritiek. Een kolossale verspilling van tijd en energie, terwijl er 0 euro voor nieuw onderzoek op de rijksbegroting staat.

3) Hoge regeldruk voor klinisch onderzoek --> duur 4) Hoge *regulatoire onzekerheid* voor first movers --> te veel risico

4 grote redenen dat farmaceutische bedrijven niet investeren in ME/CVS: 1) We weten weinig door te weinig financiering 2) Farma ziet dat overheden er niet serieus over zijn: onze kennis zal matig toenemen & farma krijgt weinig politieke dekking 1/2

@RobWust @riendebock @MarcBonten Ik ben een grote voorstander van farmaceutische bedrijven (zie Clinical Trial Abundance), maar we mogen ze natuurlijk wel bekritiseren op slecht onderzoek en juridisch dreigen

@riendebock @MarcBonten Rien, ik snap je frustratie, maar bedenk wel dat farmaceutische (lees commerciële) bedrijven de sleutel zullen zijn tot langetermijn oplossingen voor long COVID patiënten. Ik heb geen idee of er bedrijven zullen gaan investeren in dit type onderzoek.

2/ Between 2010 and 2020, an estimated 32,000 published papers used misidentified cell lines. That's not a typo. 32,000. The most common culprit: HeLa cells. They grow so aggressively they can take over neighboring cultures through aerosol contamination in shared incubators.

@pausedME Didn't you also take lamotrigine back in 2022?

No medical advice! At the beginning of February, I started Lamotrigine. Since then, so much has changed for the better. On February 14, I started tracking my steps. Since then: 364,771 steps. For me, that number means far more than distance. #MECFS

@liamsLCjourney @RuxandraTeslo it's a great article sir

@PatientPersists @RuxandraTeslo learninghealthadam.substack.com/p/why-clinical… The prequel to that, written by someone I know!

I can confirm this and think it's a unique thing abt them. I am not an EA, but I think they will have forever changed philantropy for the better.

@RuxandraTeslo I don't know if I have the spoons for substantive help with writing, but I can definitely feedback extensively, and maybe @kroetscha or the people at @1DaySooner help/can write it?

@RuxandraTeslo It would be a pretty standard 'Case for a new cause area' post, discussing Scale, Tractability, and Neglectedness. I don't think it'll really make sense to make a cost-effectiveness estimate. 'Scale' would be something like "unmet medical need" of all diseases' burdens + other

@Clarksterh @RuxandraTeslo @AlexTISYoung What do you think about Right to Try being able to pick up some signals in the noise, especially strong ones?

‘Right to try’ is, at best a weird distraction/obsession by ‘reformers’. Even if drug companies wanted to do it, which, except for some shadier companies, they do not, it wld actually slow things down. Conditional Approval after ph1 is likely a huge problem - slippery slope to approving absolute junk that you cannot then remove from market. See attempts to remove prior drugs after conditional approval (eg when companies dawdled on running Ph4 or failed Ph4 - weaponized ‘patient advocacy’ is a huge problem ). Little of this looks like it is informed by Chinese speed or in-depth knowledge of industry, but is instead an extension of standard Tech/Libertarian (mis)understandings of biotech. cc @sytses

Sid has created a list of 14 proposals to make the biotech industry more Patient First. But out of these, only 1, 2 and 6 have detailed policy proposals behind them. Anyone looking to work in biotech policy should look at credible paths to implementing the others!

@IsaacKing314 This data presentation hurts my brain

These results are very much not what I would have expected

@StuartJRitchie manifold.markets/billyhumblebra…

Hate to say it, guys, but I am getting “LK-99”vibes from this whole thing…

@erturklab I thought this was really useful background: open.substack.com/pub/siliconcon…

What do we actually gain from plastic bottle lids hitting us in the face, and from having to scroll through every website just to find “accept all”? Nobody likes it. Nobody seems to know the real benefit. Nobody feels more efficient, better informed, or better protected. So why do we still have it? To me, this is a small but very clear example of a much bigger European problem: bureaucracy that keeps growing, even when it is obviously not helping. We received a prestigious ERC grant for our academic research and an EIC grant for our biotech company, Deep Piction. But the reporting burden is so heavy that you almost need a full-time person just to manage it. What worries me most is not regulation itself. It is the inability to reverse bad regulation. Even when many people can see that something is not working, the system keeps it alive. That is a serious problem for Europe, especially in technology. In areas like AI and LLMs, progress happens weekly. If our rules and institutions need years to adapt, then we will not compete with the US or China, no matter how many strategies or programs we announce. I am not arguing for no rules. I am arguing for fewer useless rules, less administrative friction, and much faster correction when something clearly does not work. So I am genuinely curious: what do we really gain from this kind of bureaucracy? And more importantly, how do we reverse it quickly when it brings little value but keeps dragging us down?

@ohabryka @AndyMasley @PhilWMagness A lot of people caught their first covid infection AFTER being vaccinated thanks to this strategy, which has saved a lot of lives and reduced disabilities from long covid. "People got it anyways" also assumes proper only get infected once, which is clearly incorrect

Flattening out was really bad in-retrospect though? Like, most people got it anyways, the costs of lockdown were much greater than the disease burden would have been. Like, the whole thing where you ended up with a control system that maximized length of pandemic, without substantially affecting total people affected was really the worst case outcome, IMO.

The Covid lockdowns achieved nothing in stemming the disease, but they left a trail of social wreckage, economic harm, and mental health problems that will continue to play out for several generations. Thanks, Fauci. wsj.com/opinion/march-…

@minzlicht @ImplicVincent The scale in step 4 doesn't make sense/is wrong though! I also gotta say that, as an avid AI user, I notice myself discounting the bar graphs in the image given they were made in a non-deterministic way. Use AI discerningly..

@ImplicVincent The diagram was 100% LLM. I fed it the paper and gave it a one line prompt.

Ask a colleague why they refuse to use AI. They say it uses up all that water. You point out the water use is far smaller than some would have them believe. Then it's the hallucinations. You mention accuracy has improved dramatically. Then, finally: the process is the point. The struggle. The craft. The deeply human act of sitting with uncertainty. They're not reasoning. They're rationalizing their gut intuitions. My amazing student @vicoldemburgo, with Éloïse Côté, Reem Ayad, @yorl, Jason Plaks and I have a new preprint that explores this more thoroughly, called "The Moralization of Artificial Intelligence". We started by asking how moralized AI has become in public discourse. Analyzing 69,890 news headlines from 2018 to 2024, we found that AI was moralized at levels comparable to GMOs and vaccines, technologies whose moral opposition has been studied for decades. It ranked above both. The sharpest spike came within weeks of ChatGPT's launch in late 2022. When we surveyed representative samples of Americans, a majority of AI opponents said their views wouldn't change even if AI proved safe and beneficial. That's consequence insensitivity, the hallmark of moral conviction, not practical calculation. Across art, chatbots, legal tools, and romantic companions, AI moralization loaded onto a single latent factor. A global moral stance, dressed up in whatever practical language is available. The behavioral data make this concrete: a one standard deviation increase in moralization scores predicted a 42% drop in actual AI usage, even when it would have benefited that person personally. The conviction preceded the behavior by up to 573 days. The next time someone gives you three different reasons to oppose AI, each one dissolving under mild scrutiny, you're probably not watching someone think. You're watching someone feel. Preprint avaulable here: osf.io/preprints/psya…

@cstroeckw Lol just above your post I got this one: x.com/i/status/20309…

This one is really wild. Researchers put a fly connectome (basically a fly’s nervous system) into an artificial environment, and it started behaving like a 'real fly.' When I read Lena by qntm in 2018, I thought that the quite disturbing story was probably realistic, but I didn’t expect things to move in this direction so fast.

People who have taken opioids while on LDN, did LDN reduce the efficacy of the opioid?

@GarrisonLovely Thanks. Having read it, well, it's a bit stronger but it's still pretty shaky. They write very little about how they estimated task length. If the estimators weren't blinded to the hypothesis (probably weren't), this introduces a strong bias

@PatientPersists Metr transformed the data. Read the post!

People should be more aware of the secret METR graph, where they plotted 9 other benchmarks the same way and found exponential progress on all of them. Absolute performance and the slope of the graph (i.e. doubling time) varied, but they were all exponential. Ofc, anything you can grade in this way is something AI can hill-climb on. Experts I've interviewed have disagreed on how different things that we don't have benchmarks for would look. Some think it'd be similar, just starting from smaller absolute values. Others think the slopes for what I'll call "unbenchmarkable" tasks would be a lot flatter, but still exponential. I think at this point, the preponderance of evidence points toward: if you can start making deep learning work on some domain, it will improve exponentially. This won't be true in every case, but appears to be closer to the default than not. And when AI is really bad at something (e.g. computer use), it may just be doubling from 4 second tasks to 8 seconds. But when you're in the minutes range, you start to see the progress very quickly. I think this explains agents working "all of a sudden" very nicely.