This is precisely how I feel. LLMs are useful tools, worth evaluating and improving, but they are only one type of tool, one type of statistical model. In saner times, in my field, this was called "machine learning" and was integrated into many scientific tasks. Too much hype now
LLMs are extremely impressive tech, but we need to stop pretending that they extend far beyond their NLP scope.
An LLM is great at drafting an email. It however sucks at predicting over large biological knowledge graphs.
Agreed. Yet capital allocators divert endless resources into the idea of a "data panacea" to build a utopia that will never come, while starving most other domains of science and technology of the resources necessary to build a better future for a broad swath of humanity.
I believe we'll achieve ASI, but it will be both narrow and finite in capability.
We already have ASI in Go, Chess, some video games.
In those domains, infinite amounts of training data can be generated; and correctness of output is trivially and quickly validated. Imagine that: Absolutely perfect training data, available in unlimited quantities. This is the stuff of Singularity dreams.
Of course, these are also toy systems. Not like the real world. And even here, our superhuman systems aren't godlike. Give a human Go master 5 stones, and they will likely beat AlphaGoZero. SuperIntelligence, even narrow, isn't infinite.
Math as a domain is possibly as precise as Go, though orders of magnitude more complex and non-finite. If any important domain will show SuperIntelligence next, it's likely to be formal math. Even there, the gains will likely be incremental past those of humans, and comprehensible to humans, for the foreseeable future.
Coding, at first glance, shares many of the characteristics of these SuperIntelligence-plausible domains. Yet it is orders of magnitude more difficult to verify than even pure mathematics, and brings in the complexity of translating human language requirements to formal systems.
Most of human work and thought doesn't resemble these domains at all. The possibility space is infinite. Yet training data is finite or expensive. Verification is slow and subjective. Training data is mostly capped and also full of errors. SuperIntelligence may be possible in broader human domains, but we have no hard evidence that it is, and the hurdles are much higher.
It’s estimated that the Protein Data Bank (PDB) cost around $13B to create. Alphafold was only possible because of it. If we want ML to solve biology, we should be funding the creation of databases and the development of new assay technologies. ML is nothing without data.
@saintsoftness Many of us, even in STEM fields, still care. I cannot adequately explain how a love of non-science literature has helped me as a scientist, but it has. It's more than "critical thinking training" (though that's a valuable aspect). We will help you hold the line.
If you really think you can do something special...
Then live like it.
Move in the world like you're on a mission.
Start doing things that support who you really are, instead of settling for the smaller life.
The idea that sequencing more genomes would lead to better medicine and better health was a good hypothesis in 2000. But 26 years later, evidence has quite convincingly disproven that hypothesis.
The answer to most common chronic illnesses that plague us isn't written in genes. Personalized medicine likely cannot come from sequences of nucleic acids. There is more to life's dynamic nature.
Why do we cling onto that hypothesis/dogma like it is truth.
What does long-term thinking look like in practice? Introducing Long Now Labs, a collaborative space to test, prototype, and build long-term tools.
Lab Series 001 is a collaboration with the Protocol Institute to investigate three aspects of civilizational durability that are being radically reshaped by frontier technologies.
-> Lab 001.1: Book of Time - An open call to submit a concept for a new way of marking, experiencing, or making sense of time.
-> Lab 001.2: Epistemic Cycles - Seeking an individual or team to investigate historical patterns of technological disruption that broke down society's ability to discern truth.
-> Lab 001.3: Interspecies Protocols - Exploring the protocols needed to support interspecies ecologies.
If you are a designer, researcher, writer, or technologist interested in the deep future, we want to hear from you. Submissions are now open.
Learn more about Labs and how to apply: https://t.co/meZmRDc8t2
“Speculative bubble,” indeed. I’ve encountered many Theranos-like AI-centric biotechs that are VCs magnets. VCs funding garbage starves honest and non-hype driven startups of funds. It’s a game of VC “musical chairs” (or “hot potato”). Most capital allocation now is gambling.
I can guarantee that AI will not "cure cancer" - at least, not in any clean singular way.
Cancer is an umbrella term for many different diseases, affecting different tissues, with different pathologies and treatment pathways - each requiring different cures.
And this is before we discuss the inherent challenges faced by AI drug discovery, which are unlikely to be resolved anytime soon.
This "AI will cure cancer" narrative among AI utopianists, demonstrates a mixture of overconfidence, ignorance, and naivety - about both the capabilities of AI, and the applied domain of biology that they so naively delve into.
The hill I will die on - we have to rethink graduate training.
“Scientists are trained for a world where data speaks for itself. Where misinformation moves slowly. Where scientific expertise naturally rises above noise. That world is gone.”
https://t.co/hWv5M8SQjk
You need to be careful of overconfident scientists and CEOs in the AI/bio space.
Every conference I go to has someone boasting about an AI that is 99% accurate at predicting some bioactivity.
When you ask more questions, it turns out that they used a very easy dataset with a lot of bias, or have completely misrepresented their model's statistics.
I am a trained scientist in the AI/bio domain, and so can sniff out this bad behaviour from a mile away.
But how's about everyday people, investors, industry clients, civil servants, science communicators, and journalists?
Half of them wouldn't stand a chance against these slippery charlatans.
A lot of what you see online, about: "Wow omg this AI just SOLVED ageing in mice!" or "This new AI can design a personalised cancer vaccine just for you!" - turns out to be oversold bogus, when you dig deeper.
Protect your money from these people, and especially don't let them influence your health decisions.
most people operate on a model of gain, it's almost universal. their usual thought patterns revolve around questions like what do i get out of this? what do i win? what's in it for me, to make this move, start this thing, etc?
i think the inversion is more interesting & way more honest.
my operational model is closer to nothing to lose. especially when you're building a company from zero, you're operating in open territory, or even interacting with anyone new. the downside is almost always capped. the upside is infinite. i guess some ppl might see this as optimism framed another way but i think of it as pure math.
& once you internalize that asymmetry, it becomes a filter for everything. it tells you what actually matters & what's just fear. it will free you to take action.
the meanest thing you do to yourself is pretend you don’t want the things you want. shrink the desire before anyone can see it. call it unrealistic before someone else does. and then walk around with this low grade starvation you can’t name because you buried the appetite so deep even you forgot where you put it. wanting things is dangerous. I know. it opens you up to disappointment and to looking foolish and to reaching for something that might not reach back. want it anyway. the alternative is a life of pretending you’re full.