- I genuinely did not expect to see a robot handling clothing like this within the next ten years at least. Insanely impressive
I do find it interesting that they state that each task is done with a fine tuned model. I wonder if that’s a limitation of the current data set their foundation model is trained on (which is what I think they’re suggesting in the post) or if it reflects something more fundamental about robotics tasks. It does remind me of a few years ago in LLMs when fine tuning was more prevalent. I don’t follow LLM training methodology closely but my impression was that the bulk of recent improvements have come from better RL post training and inference time reasoning.
Obviously they’re pursuing RL and I’m not sure spending more tokens at inference would even help for fine manipulation like this, notwithstanding the latency problems with that.
So, maybe the need for fine tuning goes away with a better foundation model like they’re suggesting? I hope this doesn’t point towards more fundamental limitations on robotics learning with the current VLA foundation model architectures
- AlphaGo showed that RL+search+self play works really well if you have an easy to verify reward and millions of iterations. Math partially falls into this category via automated proof checkers like Lean. So, that’s where I would put the highest likelihood of things getting weird really quickly. It’s worth noting that this hasn’t happened yet, and I’m not sure why. It seems like this recipe should already be yielding results in terms of new mathematics, but it isn’t yet.
That said, nearly every other task in the world is not easily verified, including things we really care about. How do you know if an AI is superhuman at designing fusion reactors? The most important step there is building a fusion reactor.
I think a better reference point than AlphaGo is AlphaFold. Deepmind found some really clever algorithmic improvements, but they didn’t know whether they actually worked until the CASP competition. CASP evaluated their model on new Xray crystal structures of proteins. Needless to say getting Xray protein structures is a difficult and complex process. Also, they trained AlphaFold on thousands of existing structures that were accumulated over decades and required millenia of graduate-student-hours hours to find. It’s worth noting that we have very good theories for all the basic physics underlying protein folding but none of the physics based methods work. We had to rely on painstakingly collected data to learn the emergent phenomena that govern folding. I suspect that this will be the case for many other tasks.
- Yudkowsky seems to believe in fast take off, so much so that he suggested bombing data centers. To more directly address your point, I think it’s almost certain that increasing intelligence has diminishing returns and the recursive self improvement loop will be slow. The reason for this is that collecting data is absolutely necessary and many natural processes are both slow and chaotic, meaning that learning from observation and manipulation of them will take years at least. Also lots of resources.
Regarding LLM’s I think METR is a decent metric. However you have to consider the cost of achieving each additional hour or day of task horizon. I’m open to correction here, but I would bet that the cost curves are more exponential than the improvement curves. That would be fundamentally unsustainable and point to a limitation of LLM training/architecture for reasoning and world modeling.
Basically I think the focus on recursive self improvement is not really important in the real world. The actual question is how long and how expensive the learning process is. I think the answer is that it will be long and expensive, just like our current world. No doubt having many more intelligent agents will help speed up parts of the loop but there are physical constraints you can’t get past no matter how smart you are.
- Here's one: Yudkowsky has been confidently asserting (for years) that AI will extinct humanity because it will learn how to make nanomachines using "strong" covalent bonds rather than the "weak" van der Waals forces used by biological systems like proteins. I'm certain that knowledgeable biologists/physicists have tried to explain to him why this belief is basically nonsense, but he just keeps repeating it. Heck there's even a LessWrong post that lays it out quite well [1]. This points to a general disregard for detailed knowledge of existing things and a preference for "first principles" beliefs, no matter how wrong they are.
[1] https://www.lesswrong.com/posts/8viKzSrYhb6EFk6wg/why-yudkow...
- http://spotthedrowningchild.com/
You should try this. I was a lifeguard for several years, and I think the key is that there are almost always signs a person can’t actually swim. They cling to a flotation device, they stand up to their tip toes in shallow water, they seem visibly uneasy in the deep. They’re the ones who are going to get in trouble, it’s comparatively quite rare for a strong swimmer to suddenly start drowning.
- First, I’m almost certain that this article was also partially written by AI. See for example this paragraph obviously copy pasted from Deep Research
“Overall, a more nuanced view of AI in government is necessary to create realistic expectations and mitigate risks (Toll et al., 2020)”
What a unique and human thought for a personal blogpost. Also who the fuck is Toll et al, there’s no bibliography.
Second the authors used Gemini to count em dashes. I know parsing PDF’s is not trivial but this is absurd.
- This shows just how completely detached from reality this whole "takeoff" narrative is. It's utterly baffling that someone would consider it "controversial" that understanding the world requires *observing the world*.
The hallmark example of this is life extension. There's a not insignificant fraction of very powerful, very wealthy people who think that their machine god is going to read all of reddit and somehow cogitate its way to a cure for ageing. But how would we know if it works? Seriously, how else do we know if our AGI's life extension therapy is working besides just fucking waiting and seeing if people still die? Each iteration will take years (if not decades) just to test.
- Please, I really don’t think you’re discussing this in good faith.
“And the fact is that the people of Gaza could end the conflict whenever they want. All they need to do is surrender and hand over the hostages”
So no, Israel decides how and when the killing ends and apparently that’s when “Gaza no longer poses a threat”. Who knows what that means but apparently it involves mass starvation, firing tank rounds into crowds, and destroying every hospital.
- I keep hearing the “the killing would end if Hamas would just release the hostages”. But the Israelis keep offering ceasefire terms that include full release of hostages but no permanent cessation of hostilities, only 60 days and not even temporary full withdrawal from Gaza.
Why do you think the Israelis want to keep their tanks in Gaza even after all the hostages come back? Why won’t they offer a full and permanent ceasefire? I think this hostage justification is just Israelis buying time so they can keep on doing what they actually always wanted, full ethnic cleansing of Gaza.
- Molecular dynamics describes very short, very small dynamics, like on the scale of nanoseconds and angstroms (.1nm)
What you’re describing is more like whole cell simulation. Whole cells are thousands of times larger than a protein and cellular processes can take days to finish. Cells contain millions of individual proteins.
So that means that we just can’t simulate all the individual proteins, it’s way too costly and might permanently remain that way.
The problem is that biology is insanely tightly coupled across scales. Cancer is the prototypical example. A single mutated letter in DNA in a single cell can cause a tumor that kills a blue whale. And it works the other way too. Big changes like changing your diet gets funneled down to epigenetic molecular changes to your DNA.
Basically, we have to at least consider molecular detail when simulating things as large as a whole cell. With machine learning tools and enough data we can learn some common patterns, but I think both physical and machine learned models are always going to smooth over interesting emergent behavior.
Also you’re absolutely correct about not being able to “see” inside cells. But, the models can only really see as far as the data lets them. So better microscopes and sequencing methods are going to drive better models as much as (or more than) better algorithms or more GPUs.
- Don Pinkel is not well known but he was a pioneer in the 60’s at St. Jude in Memphis in developing the first combination treatments that pushed the childhood acute lymphoblastic leukemia cure rate from effectively zero to about 50%.
https://www.smithsonianmag.com/innovation/childhood-leukemia...
- The problem with this machine-learned “predictive biology” framework is that it doesn’t have any prescription for what to do when your predictions fail. Just collect more data! What kind of data? As the author notes, the configuration space of biology is effectively infinite so it matters a great deal what you measure and how you measure it. If you don’t think about this (or your model can’t help you think about it) you’re unlikely to observe the conditions where your predictions are incorrect. That’s why other modeling approaches care about tedious things like physics and causality. They let you constrain the model to conditions you’ve observed and hypothesize what missing, unobserved factors might be influencing your system.
It’s also a bit arrogant in presuming that no other approaches to modeling cells cared about “prediction”. Of course, systems and mathematical biologists care about making accurate predictions, they just also care about other things like understanding molecular interactions *because that lets you make better predictions*
Not to be cynical but this seems like an attempt to export benchmark culture from ML into bio. I think that blindly maximizing test set accuracy is likely to lead down a lot dead end paths. I say this as someone actively doing ML for bio research.
- Apparently, the article for David Woodard, an American composer and conductor has been translated into 333 languages, including Seediq, a language spoken in Northern Taiwan by about 20 thousand people.
I am absolutely baffled as to why this is the case. I have to imagine some kind of "astroturfed" effort by Woodard or a fan to spread his name?
- It doesn’t know anything about where it “needs” to go. One of the weirder and more unintuitive things about molecular biology is just how fast everything moves inside a cell. The CRISPR molecule diffuses from one side of the nucleus to the other in a couple seconds and probably bumps into the entirety of the genome in a matter of minutes or hours. It’s very, very crowded inside cells, proteins and DNA and metabolites are constantly bumping into each other and are tumbling around at frankly incomprehensible rates. So, nothing needs to “know” where it needs to go, it simply gets pushed and jostled around until arrives there and then the attraction between the CRISPR’s RNA and the DNA takes over
- Germline gene editing is still considered risky and unethical. That is, editing cells that form eggs and sperm, thus changing the genome of some of the descendants of the edited person. This is somatic editing. These edits will not be inherited.
Somatic editing is becoming more common (see Casgevy) but there are technical hurdles that prevent its application to many cases.
I’ve never heard about modern people doing serious persistence hunting, except for a stunt that I read about years ago. I think it was organized by like Outside or some running publication that got pro marathoners to try and they failed because they didn’t know anything about hunting