pfisherman

Follow up in this one… I asked an LLM to give me the funniest way to combine the concepts of a dinosaur and monetary policy of the Ottoman Empire. This was the answer.

Imagine a “Dinoflationosaurus”: a giant dinosaur who has the job of overseeing the monetary policy of the Ottoman Empire. However, this dinosaur is hopelessly behind the times, using outdated gold coins that are buried in random locations, like a prehistoric central bank.

Instead of regulating currency or adjusting interest rates, the Dinoflationosaurus spends its days stomping around, either hoarding or releasing massive piles of treasure based on whether it sees its shadow, causing huge economic fluctuations. Merchants and citizens scramble to predict where the dinosaur will dig next, turning the entire economy into a game of dinosaur-sized hide-and-seek with inflation spikes tied to the beast’s mood swings.

The Ottoman economists, dressed in traditional robes, nervously try to explain to the sultan that no one knows when the giant lizard will “stimulate the economy” by smashing a treasury vault open.

Try to code something like that up in prolog!

That is exactly the point I was making in my comment above. This type of unambiguous problem is best solved using formal languages - something more like quantitative reasoning. But stuff like prolog or classical automated reasoning approaches are quite brittle. They break down quickly when you start to introduce ambiguity and noise. Statistical approaches like hidden markov models that people used in these instances were the precursor to the LLMs we have today.

But I was going down a rabbit hole there. My main point is that trying to use LLMs to solve logic puzzles - that can easily be solved in prolog - is a waste of time and a failure of the imagination. The applications that should be explored and would be most fruitful are those where there is ambiguity and contradiction.

How can one / should one combine the concepts of a dinosaur and monetary policy of the Ottoman Empire? What differentiates verbal reasoning from logic?

I don’t know that either of those can be solved well with formal languages or logic.

LLMs and NLP are to verbal reasoning what the calculator is to quantitative reasoning.

Language and by extension verbal reasoning is full of ambiguity and semantic slipperiness. For example, what degree of semantic similarity distinguishes synonymous from synonym-ish concepts? When do we partition concepts into homonyms?

I think part of the problem with how people evaluate LLMs is that the expectations that people have. Natural language != ontology. The expectation should be more Chomsky and less Boole. Asking it to solve math problems written in paragraph form is a waste of time. Use a calculator for that! Solving riddles? Code it up in prolog!

Instead you should be thinking of what operations you can do on concepts, meaning, and abstract ideas! That is what these things do.

The big companies beefing up their portfolios is more of a defensive measure from what I have been told. The threat of a countersuit makes IP litigation among peers a game of mutually assured destruction. Funny quote from a VC friend “patents are a sport of kings.”

But defensive portfolios are not a concern for patent trolls. They can’t be countered for infringement because they don’t make anything!

But you are not wrong that they can be like gnats that suck the blood from startups. In one car I heard of some a patent troll engaged in behavior bordering on criminal extortion - threatening mom and pops businesses for using printers and fax machines. All I am saying is that they play a role in balancing and maintaining the health of the IP ecosystem.

By going out of business and the troll acquiring your IP portfolio in a fire sale?

Well I think it depends on the industry. Patents tend to be less relevant in software. But in other capital intensive industries where there is some manufacture of physical matter - like biotech or hardware - IP is a big deal.

Heard an interesting counterpoint to this from a patent attorney. In the IP ecosystem patent trolls serve as a sort of check on the big companies - the apex predators - to stop them from willfully infringing on your patents and then bankrupting you in litigation.

While you as a startup may not have the resources to go after them in court; your IP assets in the hands of a competent and aggressive patent troll could be a very big problem for Big Co.

So in that sense they are also kind of like a parasite that infects the apex predators who eat tainted meat.

Had the chance to watch some non programmers use copilot for data science (using pandas) and it was an eye opening experience. I came away with the feeling that the tool landed in a sort of “uncanny valley” of productivity. If you can’t write the code without copilot then you won’t be able to debug the errors it makes. And if you know enough to spot and debug the errors, then copilot just gets in the way.

Questions for those in the know about chip design. How are they measuring the quality of a chip design? Does the metric that Google is reporting make sense? Or is it just something to make themselves look good?

Without knowing much, my guess is that “quality” of a chip design is multifaceted and heavily dependent on the use case. That is the ideal chip for a data center would look very different from those for a mobile phone camera or automobile.

So again what does “better” mean in the context of this particular problem / task.

I think you may have a bit of hindsight bias. The internet was not immediately recognized as useful. One of my first jobs was at an e-commerce startup that folded in the dot com bust because (1) investors did not believe that our target customers would favor online shopping over ordering from physical catalogs, or (2) foresee how online sales might impact business operations or the viability / scalability of business models - such as drop shipping.

I think that you are not appreciating how much of a quantum leap this is in technology. This is like a calculator for verbal reasoning. Humanity has little idea of what is even possible with something like that.

Also, I am old enough to remember - and have experienced in my professional career - a time when the mainstream opinion was that the internet was a tiny niche and that things like online shopping or dating would never gain widespread adoption.

There is a big gap between a master’s and a PhD, and then another between a PhD and a seasoned pro. To do something like a bioweapon, you would need a reasonably sized team of pros w/ a lot of capital intensive infrastructure. It would be virtually impossible to do in secret.

It is a bit more nuanced than that. An incorrect unstated assumption here is that schizophrenia or depression is a singular disease rather than a group of observable symptoms where any two cases may or may not have the same underlying cause.

There are certain types of schizophrenia that are highly heritable and others that are more sporadic. Highly pathogenic variants tend to be very rare because those regions of the genome are under evolutionary constraint.

GWAS are used is to figure out which genes may be involved in the pathology. So we are essentially looking at rare cases of inherited disease to figure out which genes might play a role in more common non inherited versions.

Downtown has never been where the bulk of the ppl in SF live or hang out. It is for the “bridge and tunnel crowd” —- mostly offices and businesses; and not meant to be habitable.

What you are doing is just good business. You are not implicitly or explicitly colluding with other property owners to set you prices. You are simply doing market research based on publicly available data and making an independent decision.

Contrast this with a cartel that includes a significant number of landlords in your city (i.e. way more than just a couple of your buddies) that (1) shares non public info such as current occupancy rates, current lease terms and durations, etc., (2) sets prices as a bloc, and (3) enforces compliance on pricing targets.

It’s qualitatively different from how you described your situation.

Eh, I used chat-gpt to make up a story (with elements specified by the little). Was quite good for keeping them entertained in the car for the 10 mins mommy was in the store picking up a prescription.

Bedtime stories seems unnecessary. Like there is no shortage of good books out there. Plus, I would feel a sense of loss if I didn’t get to do pirate voices when reading treasure island.

Tutoring … lol. Those kids must be better behaved than mine if they can be trusted to do that sort of thing autonomously.

Which high tech child rearing trends do you think are harmful? Just curious.

The government funds a lot of early stage preclinical research. These are the inexpensive stages of the pipeline.

As soon as you move into humans you can add another 2 or 3 zeros to your burn rate.

It is not politically feasible for the public sector to fund later stages. The numbers are just too big. Just think of the campaign ads that would run around $200M late stage failure funded by the government.

The reason why mega giant pharma companies exist is because they make enough money and are capitalized enough to withstand multiple $100M failures without going belly up.

I don’t know where the original poster is from, but I am kind of scratching my head at their comment. There is basically nowhere in SF I would have the slightest bit of fear walking around with the exceptions of bayview / hunters point or sunnydale.

Counterpoint: contentment is a prerequisite for healthy sustained effort. It’s easy to get up every day and chisel away at something inch by inch - sometimes with little to no perceptible progress - when you are fulfilled elsewhere in your life.

Nice story about perseverance. But I do not know how helpful this is to young researchers.

My advice for young researchers is read more articles - like at least a solid month of reading and journaling full time (40 hrs per week) before you even start to think about what you want to start off with by replicating.

The other related mistake I see young researchers make a lot is not leveraging pre-existing work / results and wasting weeks or months reinventing the wheel.

The issue is the lack of informed consent. This is pretty basic ethical conduct of research stuff.

Textbook case of unethical conduct of research. The key here is lack of informed consent by the study participants.

The APA put out a press release about this study violated their code of ethics.

https://www.apa.org/news/press/releases/2014/06/informed-con...

You should generally be setting attributes in the init method of your class, unless you want it to act like a class attribute - shared by all instances. The exception is some special types of classes such as dataclasses or pydantic objects, which do it automatically but require a post init method if you want to do anything more than set values passed to the constructor.

Tbh, if they were writing something for prod they should have factored out the uuid generation and passed values into a pydantic dataclass w/ validators for all the fields. Just sayin.

USD has the FDA and EU has the EMA. I am not sure how EMA interacts with each individual country.

While regulatory agencies may consider data generated from trials in other jurisdictions, they may ask questions about whether those data and conclusions and relevant and generalizable to their patient populations and health care systems.

There are also ethics to consider. Sure it may be cheaper to run trials in developing countries. But at what point does it become exploitative? Keep in mind that we are talking about human experiments, which is very serious business.

Without comment on the broader subject of drug pricing - especially in the US - this article leaves out enough relevant information that I would consider it to be intellectually dishonest.

Where did they run the trials? Across how many sites? Do they plan to apply for marketing approval in the US or EU? Are they the manufacturer? Did they beat the cost of setting up the supply chain and manufacturing facilities prior to the phase 2/3 trials?

The answers to those first three questions will have a huge impact on the price tag.

These gene names are human readable strings, not concept unique identifiers. We in bioinformatics / comp bio world strongly encourage the use of ncbi or ensmbl gene IDs. In general it is bad practice to use human readable “names” of things as a primary indexing key.

Sounds cool, but how you define success for something like that? I can copy a prokaryotic genome mutated based as some non-zero rate and it would probably be viable. Is that synthetic enough to count? Are they going for a minimal genome?

How about something more useful, lucrative, and easy to define success for like engineering a morphine synthesis pathway into E. coli or something.

Imo, if you are talking about synthetic biology, then their training data is insufficient. Synthetic bio explores a lot of design space that is far outside of anything you would see in nature. There the secret sauce would not be in the generative pretraining, but in the RL. Unfortunately bio experiments are noisy, slow, and expensive so good luck getting enough data before the heat death of the universe.

DNA libraries are read out using sequencing techniques. But I doubt anyone would say they are just measuring DNA. It’s kind of like saying a luciferase based assay is just measuring the intensity of light.