pfisherman

Yes. I know of two people who had breakthrough infections. I had a mild breakthrough myself. Got the brain fog and fatigue but was thankfully spared the fever, cough, and sore throat.

Definitely not paranoia. Near constant respiratory infections are par for the course with grade school aged children. You typically don’t call or visit the pediatrician unless things are especially bad or unusual.

If you know how these algorithms work, and you can be intentional about what you want, seeding with a few well thought out examples, and curating recommendations; they can be quite useful.

I think atomic power or even better drugs / medicines is actually a good analogy considering the dual use nature of the stuff that they are building. Can improve quality of life if used prudently and responsibly, or cause devastation if not.

Is something like automated personalized content creation (for ads) better targeting? Or is it qualitatively different?

I personally think that the population scale surveillance and behavioral manipulation infrastructure built by meta is unethical and incredibly dangerous.

> They aren’t going to do basic research.

Fixed that for you. Pharma companies do lots research. But the goal of that research is very narrow - to bring new therapies to the market. The Academy has very different goals - train next generation of researchers and make discoveries. IMO, expecting a pharma company to do the work of an academic institution is a recipe for failure. But that is just an opinion.

> The US Economy is pretty vulnerable here. If it turns out that you, in fact, don't need a gazillion GPUs to build SOTA models it destroys a lot of perceived value.

I do not quite follow. GPU compute is mostly spent in inference, as training is a one time cost. And these chain of thought style models work by scaling up inference time compute, no?

So proliferation of these types of models would portend in increase in demand for GPUs?

Adding to that — the mission of academic research centers is to train the next generation of researchers. PIs use this money to fund grad student and postdoc positions in their labs.

Given trends in media (social and otherwise) will have to take a wait and see approach. But sudden cancellations with limited communication is definitely not a strong indicator of a well thought out or executed plan.

Feel sorry for the young PIs and postdocs.

I used to do bench top work too; and was blessed with “the golden hands” in that I could almost always get protocols working. To me this always felt more like intuition than deductive reasoning. And it made me a terrible TA. My advice to students in lab was always something along the lines of “just mess around with it, and see how it works.” Not very helpful for the stressed and struggling student -_-

Digression aside, my point is that I don’t think we know exactly what makes or defines “the golden hands”. And if that is the case, can we optimize for it?

Another point is that scalable fine tuning only works for verifiable stuff. Think a priori knowledge. To me that seems to be at the opposite end of the spectrum from “mess with it and see what happens”.

This is still on retrospective data. The machine learning graveyard is filled with models that worked well on retrospective data, but did not hold up in a live inference setting. Just ask Zillow. The real test is whether they can predict the weather 14 days out in 2025.

I am guessing they did not want to set up the data pipeline to run inference in a live setting. But that is what I would need to see to be a true believer.

Still a cool result and article though.

> The most significant part of Crick’s idea — and the part that Watson ignored in his oversimplification — was that there are three flows of information that cannot occur, due both to lack of experimental evidence and any plausible biochemical mechanisms. These were protein → protein, protein → RNA and, above all, protein → DNA.

These are categorically untrue. Post translational modification of proteins (by proteins) is a thing. Post transcriptional modification of RNAs (by proteins) is a thing. Transposition of DNA sequences (by proteins) is a thing.

The parent of any young child can tell you that they learn through lots of exploration and reinforcement - often to the worry and chagrin of caregivers. Indeed much of our job is to guide exploration away from excessively dangerous “research” activities (ex. locking away cleaning products).

I am not an expert in the particular application domain of that article; but I can see why their argument of pre training might be valid. It is not especially controversial to say that pre training neural nets improves few shot learning performance. And I suspect there is an inflection point for every problem where pre trained neural nets yield better few shot learning performance than less data hungry approaches - such as hand crafted features or strong priors.

That being said, it seems that the question here is whether that inflection point has been reached in this case.

I very much agree with the perspective that LLMs are not suited for “reasoning” in the sense of creative problem solving or application of logic. I think that the real potential in this domain is having them act as a sort of “compiler” layer that bridges the gap between natural language - which is imprecise - and formal languages (sql, prolog, python, lean, etc) that are more suited for solving these types of problems. And then maybe synthesizing the results / outputs of the formal language layer. Basically “agents”.

That being said, I do think that LLMs are capable of “verbal reasoning” operations. I don’t have a good sense of the boundaries that distinguish the logics - verbal, qualitative, quantitative reasoning. What comes to my mind is the verbal sections of standardized tests.

How good are our animal models of ALS? Are they predictive of effectiveness in humans?

Prop 13 had a huge negative effect on quality of public schools in California, which I got to experience first hand.

The difference was quite apparent to me during high school when I compared my older siblings’ yearbooks to my experience of the same school a decade later. They had so many more classes, clubs, sports, programs, and activities available to them than I did.

Agreed. I was on twitter to discover, share, discuss articles and talks in my field of research. A couple months ago I had trouble finding good accounts and content; but now I am starting to find some of the PIs I was following over on twitter.

Biosci twitter -> BioSky?

Why use an LLM as opposed to a more narrow purpose built model? LLMs are not beating smaller, purpose built models on tasks like POS tagging, NER, sentiment analysis, etc. And the inference costs scale quite poorly (unless you are self hosting llama or something).

Have people not always been this way? It seems like people forgotten their Emily Post —that it is generally considered a faux pas to discuss controversial topics like religion or politicd outside of intimate social settings with close friends and family.

Anyway, I don’t want an echo chamber for my beliefs. I just want to be able to discover and discuss scientific articles and watch sports highlight clips without being bombarded by a bunch of bullshit.

The “might work” is doing a lot of heavy lifting in this sentence. How strong must the evidence be for the hypothesis of efficacy to be considered reasonable?

So what exactly did the CEO say? Did he attest to tether’s balance sheet? Did he attest to the dollar amount of tether’s assets managed his institution?

What I am getting is that he simply said that he holds some of tether’s assets. Why and how would that make him privy to the entirety of tether’s balance sheet?

What are the incentives at work here? What is their risk exposure to a collapse of tether? What would be the reputation al risk to their firm if tether was a scam?

I personally would take this with a grain of salt as they have a direct financial interest in Tether looking solid and prestigious and continuing its present operations.

I am old enough to remember when investment banks touting their mortgage backed securities business before the collapse in 2008.

Also they talk a lot about data centers and cloud compute, but do not mention chips for on decide inference.

Given where mobile sits in the hierarchy of interfaces, that is where I would be placing my bets if I were a VC.

Have not read the article in depth yet, but I am very familiar with the Troyanskaya lab. They are good and their articles are always worth a read.

They are using a geodesic finite mixture model, which I am not familiar with, but seems to be an extension of mixture models to non Euclidean metric?

Pleiotropy can be tricky though, as traits / diagnoses / observations can have causal relationships amongst themselves. For example, a variant that impacts obesity may be statistically associated with heart attacks; but the relationship between the variant and heart attacks is not directly causal in the sense of something like a variant that alters function of an ion channel expressed in cardiac smooth muscle.

I can see the big chip makers making out like bandits - a la Cisco and other infra providers with the rise of the internet.

They are facing competition from companies making hardware geared toward that inference that I think will push their margins down over time.

On the other end of the competitive landscape, what moat do those companies have? What is to stop OpenAI from pulling a Facebook and Sherlocking the most profitable products built on their platform?

Something like Apple developing a chip than can do LLM inference on device would completely upend everything.

> This definition of success is founded on power and control.

I don’t get how this follows from the quote you posted?

My interpretation is that successful people create durable, self sustaining institutions that deliver deeply meaningful benefits at scale.

I think that this interpretation is aligned with your nobler definitions. But your view of the purpose of government and religion may be more cynical than mine :)

Who are you betting on then? Anthropic? Google? Someone else? I mean Microsoft was not the friendliest company. But they were good enough at serving their customers needs to survive and prosper.

The etymological origin of “credit” comes from Latin for believe or trust. Credibility is everything in business, and you can put a dollar cost on it.

While banditry can work out in the short term; it pretty much always ends up the same way. There aren’t a lot of old gangsters walking around.

Biosci Twitter used to be amazing. Great resource to find cool new articles if you followed the right people.

Now it is absolutely useless for that purpose. Such a shame. Death of a golden age.

Not sure we know exactly how valproic acid works. HDACs are like master regulators. Histone acylation is kind of like the opposite of DNA methylation. It is a marker of active transcription. So messing around with these can lead to big epigenetic changes (i.e. state of the genome) in impacted cells / tissues.