The wits in robotics would say we already have domestic robots - we just call them dishwashers and washing machines. Once something becomes good enough to take the job completely, it gets the name and drops "robotic" - that's why we still have robotic vacuums.
I don't get the spike in drama though. "Don't believe everything you read on the internet" was a common sense maxim long before AI came around. What is so different now? If anything, it just amplifies the maxim, which is arguably a good thing.
My fave theory for the origin of life is that the porous spaces in the material around the vents were full of pockets which functioned as different cells in which sets of molecules would interact. Groups of molecules that cooperated to form a successful self-replicating metabolism that excluded freeloaders would then colonise neighbouring pockets. But free loader molecules that could infiltrate a metabolism and contribute nothing, but use it only for their own self-replication would also thrive. according to this theory the evolution of metabolism precedes the development of the cell membrane, the cell membrane might have begun as incomplete barrier to reduce loss and entry to neighbouring pockets and evolved over time to become a complete barrier that eventually removed the dependency on a vent-rock-pocket.
What I love about this theory is that it sets up the contest between co-operation (cells) and infiltration (viruses) as strategies for life right from the get go.
Immune evasion in cancer is really interesting and particularly important because immune therapy has really emerged as a groundbreaking therapy. But, while immunotherapy excels in blood cancers, it still struggles in solid tumors. Part of the reason for this is that a solid group of cancer cells can much more easily change their microenvironment to exclude immune cells.
So this paper explores one of the ways they do that. Cancer cells turn off a protein that sits on the surface of normal human cells that immune cells regularly bind to. Binding to this protein, CD58, causes changes in the immune cell. What these changes are critically depends on the other signals that the immune cell senses on the surface of the cell. If it senses another (somewhat famous) protein, PD-L1, the immune cell self destructs, as PD-L1’s presence typically means that a cell is normal and healthy. But, cancer cells frequently evolve to over generate PD-L1, thereby creating an micro-environment filled with self destruct signals. Further, the other protein CD58 actually increases the physical force of immune cell binding, so when cancer cells turn off CD58 they actually get more slippery.
This paper explores an interesting connection between the two proteins. There’s a third protein which binds to both of them and increases the amount of time they're active on the cell’s surface. Honestly I haven’t had time to fully digest this bit, but it seems like the regulatory protein preferentially binds to CD58, which is good because it promotes immune cell adhesion and recognition of the tumor cell. But! When CD58 is lost, it then binds to PDL1 and promotes immune cell evasion.
Why nature chose this as a regulatory mechanism is kind of baffling, but exactly the kind of thing that makes cell biology interesting
They're also grown with high phosphate fertilizers which produce a lot of decay products ending up in Po-210. THEN they aren't washed, and THEN they're dried under gas heaters which promote the formation of Tobacco-Specific Nitrosamines, which are incredibly carcinogenic.
One of the many reasons why, even though smoking anything is not great for your health, smoking tobacco is particularly harmful.
It makes me worry about the invisible minions that are pushing the blocks into place. They only have a second, blocks are falling, how can the get it to the right numbers in time? But as soon as the numbers are legible, the blocks for the next number are already falling, and the invisible minions must jump straight to working on those blocks without a moments rest, second after second, nonstop for eternity.
I have mixed feelings about the things in this list. E.g. part of my disability experience involves my mental health and, when you’re talking about me, anyway, I don’t mind the term “mentally ill”.
On the other hand, it sort of irritates me when people who are well describe others as “crazy”.
Regarding your comment on disability being “abnormal and negative”, though, I have to disagree. I would not have survived the last few years if I had characterized my impairments as being abnormal and negative.
There’s an approach out there called the “social model” of disability where impairment is regarded as a neutral (neither good nor bad) experience and the good or bad experience is how the impairment changes interaction with society. It’s worth reading at least the introduction on the Wikipedia page, so I’ll link it.
Edit: To be clear, I’m just thinking about the list of words, not any sort of policy about what one can and can’t say. That’s a whole other can of worms!
This is called "hallucination" and I find it to be the biggest flaw with GPT-3 interfaces like ChatGPT. Basically, the system will start convincingly BSing at any point and as a user, you're never sure if any particular detail it outputs is actually correct. The problem is, these large language models are inherently kind of a black box, so how do you fix this kind of behavior?
I started looking and it turns out that OpenAI is keenly aware of the hallucination problem, and even wrote about how they're correcting for it in this blog post about InstructGPT: https://openai.com/blog/instruction-following/
To summarize: it seems OpenAI is addressing this by taking human-labeled output data from GPT and feeding this into new models, using a reward function to train the future models to behave with less hallucination. This way of working struck me as a mix of software engineering and crop breeding.
They discuss the trade-offs involved with doing this. The "hallucination" effect is probably one of the features that make GPT so creative.
I like that you can prompt "longer and with with more tangents".
You would probably enjoy Bureaucrats and How to Annoy Them[1] by R.T. Fishall, a nom de plume of Patrick Moore (of xylophone fame [2], maybe some astronomy too).
Can confirm. Three years ago I spent 2 months in a furnished two bedroom apartment two stops away from Shinjuku - paid $1,500 / month. Had three different subway stations within ~7 min walk each. Great location, very inexpensive.
The company I used is Fontana - English speaking, they set up gas/electric/water and all I had to do was pay for them at the convenience store nearby (7 Eleven). The only requirement is a 2-month minimum stay. Will do again.
Current sequence models don't have the right structures to represent math. Even if they use floating point internally, they can't really float the point because the nonlinearity in the model has a certain scale.
A system that processes language can take advantage of the human desire for closure
Commodifying software was a mistake that resulted in all sorts of perverse incentives. We first saw Microsoft succumb to it, then Amazon, and are now watching Facebook, Apple and Google all create a mustard-gas-miasma of dark patterns and rent-collection. Really, it's our fault for not recognizing these threats during the dotcom boom.
Currently at my “peak” at Google (senior staff), after having worked at a healthy mix of startups/mid-cap/large-cap before, so I like to think I’ve seen a bit of everything and I’m starting to age in my mid 30s, I’m consistently the oldest around the office.
My next move will either be “retiring” and trying to start a solo business in super low-pressure mode (thanks to good savings), or find a right partner and be a cofounder at a typical venture backed startup in the Bay and shoot for the stars. Likely nothing in between.
So far I’m not too motivated about the second even if more than one person told me that I have the right attitude, practicality and perseverance to get stuff done (I don’t totally believe them, I know how hard it is to start one as I was in the front seat as early employee a few times). I am more looking forward to the first, somehow I am enticed about the idea of spending my days not talking to a single soul, except customers via email (can you tell I spend 70% of my time in meetings?!).
I am waiting for a unicorn I have a good chunk of equity in to go public (or bust) before taking a decision about what’s next. At that point, my FAANG compensation will either become an irrelevant drop in the bucket, or a necessity not worth risking for several more years.
So here’s my tinfoil hat theory. When I watch Tik Tok (and I am fully aware my feed is different from others), I become genuinely happy. I see people singing, creating music together via duets, showing homesteading skills, teaching things they know, telling funny jokes, etc. I never see dangerous challenges, though I am not a teen. Here’s the tinfoil hat theory: Tik Tok is what social media was meant to be, but unlike American social media giants that made their deal with the devil and drive growth via separating and polarizing people, Tik Tok makes people feel like the world isn’t a scary awful place with “them vs us” mentality, and so it poses a threat to our political culture which feeds off division.
All I know is that when I go on Facebook I just see angry boomers being quasi political, when I go on Instagram I am filled with envy, when I go on twitter I feel overwhelmed by the sheer flood of information and vileness, and when I go on Tik Tok I feel happy and a new ambition to pick up the guitar again.
Entrepreneurship is like one of those carnival games where you throw darts or something.
Middle class kids can afford one throw. Most miss. A few hit the target and get a small prize. A very few hit the center bullseye and get a bigger prize. Rags to riches! The American Dream lives on.
Rich kids can afford many throws. If they want to, they can try over and over and over again until they hit something and feel good about themselves. Some keep going until they hit the center bullseye, then they give speeches or write blog posts about "meritocracy" and the salutary effects of hard work.
Poor kids aren't visiting the carnival. They're the ones working it.
Linguists refer to these traits you're describing by classifying languages as being more or less "synthetic" or "analytic."[1] Broadly speaking, synthetic languages tend to encode information by modifying words (e.g. the articles "der" vs "dem" in German); analytic languages tend to encode information via adding words and word order (e.g. "the" versus "to the" in English.)
Real languages exist somewhere on a spectrum. Languages in the Indo-European family show an empirical historical tendency to become less synthetic and more analytic over time and via contact with other languages. German is more synthetic than English, but less synthetic than Russian, and much less synthetic than very old IE languages such as Vedic Sanskrit. English is more analytic than German, but less analytic and more synthetic than Afrikaans. Some non-IE languages such as Finnish are extremely synthetic.
This is an observation that goes back to at least Cicero.[1]
Cicero on the primary goal of oratory:
"As, therefore, the two principal qualities required in an Orator, are to be neat and clear in stating the nature of his subject, and warm and forcible in moving the passions; and as he who fires and inflames his audience, will always effect more than he who can barely inform and amuse them..."
Cicero describes the problem the OP reports:
"But let us return to Calvus whom we have just mentioned,—an Orator who had received more literary improvements than Curio, and had a more accurate and delicate manner of speaking, which he conducted with great taste and elegance; but, (by being too minute and nice a critic upon himself,) while he was labouring to correct and refine his language, he suffered all the force and spirit of it to evaporate. In short, it was so exquisitely polished, as to charm the eye of every skilful observer; but it was little noticed by the common people in a crowded Forum, which is the proper theatre of Eloquence."
Nuanced communication not working at scale, 2100 years ago.
CAR-T therapy is incredibly complex and incredibly cool.
One of the reasons CAR-T therapy has been so successful thus far with certain lymphomas and some leukemias is that there is a specific surface protein (CD19) which is expressed in all B-cells (the deranged lineage in the case of lymphoma) and is also not expressed by any other cells in the body. By engineering a patient's T-cells to target CD19, you create a highly sensitive and specific attack that recruits their own immune system to annihilate the entire B lineage population.
One problem we run into when trying this for other cancers (like, that don't come from B-cells) is that it's been really hard to find such a nicely specific surface protein, as well as an entire population of cells you can just annihilate and be survivable for the patient. Most surface proteins are expressed in varying degrees throughout various different organs in the body, so a CAR-T against it would cause a ton of off-target effects. In some early trials for certain cancers they encountered this with unfortunate side effects (including in some cases death). Nevertheless, there is lots and lots of research still ongoing in the field, which is super exciting, from trying out previously unknown targets, to figuring out how to better produce the T-cells, to enhancing the resultant immune response cascade, etc etc.
Yes, it's really fascinating stuff. Just to add to that, cells must present a continuous sample of their proteasome on the cell surface in the form of a short peptide 9-12 amino acids long. T cells will recognise when the cell is producing protein it shouldn't be (i.e. because it became cancerous or infected with a virus), even if only one amino acid is incorrect.
It learns this in the thymus, which has a bizarre gene called AIRE which switches on genes from all over the body, essentially creating a representation of the entire body in the thymus gland as a sort of sandbox before they are let out.
However, we don't have enough T cells to recognise every possible sequence of 11 amino acids (which would weigh around 1.5 tons), so T cells must be so-called 'cross-reactive'[1]. And therefore other factors must go into how T cells respond to abnormalities.
At least part of the adaptive immune system is implemented in the thymus. As an infant, the thymus makes examples of nearly every type of cell in the body and uses it as negative labelled examples to tune the false positive detector so it doesn't identify self as threat. From an information theory perspective, that's pretty extraordinary (you don't normally expect differentiated cells in an organ to act like cells from another organ).
Another part of the adaptive immune system randomly shuffles different regions of genes together to produce enormous diversity (searching for a rare example of something that "works"), then picks the proteins from those genes that work best and distributes them throughout the body.
It's possible to train your mind to do this. It's not just musicians who find their inspiration in dreams.
Albert Einstein and Salvador Dali practiced naps while holding an object over a plate. When they dozed off and dropped the object, the clang would wake them and they could better remember their thoughts from the dream-like state moments before.
Wrote professionally about taste in a previous life. It is related more closely than we expect to techne or competence from physical knowledge.
When we think of poor taste, we tend to think of symbols that are separated from their function and meaning, where instead of representing that, "I do this thing," something gaudy says, "I have this thing!" That's what crassness is, and it comes down to our relative apprehension of the real vs. the represented, where typically, something real is powerful independent of who is observing it, and the representation is not. It's whether something legitimately represents power. Taste may be an instinct for honest signals, which would seem like its own sort of intelligence.
Viewed this way, taste is the expression of what you percieve to be power based on your experience, good taste is the inverse of the distance between them, and poor taste is measured in the gap between what is affected and of-what it is the effect.
That difference between effect and affect is one of the sneakiest bits of the english language and perhaps even the culture's most cunningly set trap. Do not underestimate the value of good taste, it's an intuition about power.
Happiness should always be a factor, but I will argue that sanity permitting you should always test out new waters. Perhaps you might be satisfied at retirement with what you’ve got, but if you can forge new paths without sacrificing the clearly important parts of your life (sanity and family) then I believe that you owe it to yourself to do so. Not every half a mill job is fun but there are fun ones from what I hear.
Security researcher once told me that he sees social media as a distributed hacking attempt on the human mind.
I think it's a genetic algorithm. You try random stuff and when something works you clone and mutate and crossbreed it.