HNHacker News
TopNewBestAskShowJobs

inciampati

1,697 karma · joined June 29, 2017

submissionscomments
inciampati··on Medieval
Did they resolve the intense hardware bugs with the EP-133? I haven't been following after basically giving up on TE due to this (EP-133) being the second time they sold me a lemon.
inciampati··on Tomato nostalgia as I relive my Croatian island childhood
You need sun and local intensive agriculture to grow them well. They taste better the more hot and stressed they are. It's just not something available in the UK?
inciampati··on Tomato nostalgia as I relive my Croatian island childhood
Buy a cheap flight to south Italy and gorge yourself.
inciampati··on Building a Local Perplexity Alternative with Perplexica, Ollama, and SearXNG
I'm a happy customer. They provide a good service that saves me hundreds of tiny chunks of time during research. The instant ads hit, I'm out. Any distraction from the results is a time and mind suck which users like me will have no tolerance for. Thankfully, what they're doing is "only" as complex as driving searches with LLMs. And it probably doesn't even involve fine tuning, given that they're able to run new LLMs the same day they're released. I hope to soon be a happy customer of the natural open source replacement for this.
inciampati··on Creativity fundamentally comes from memorization?
Modeling is not memorization. It's more generic and can't allow you to reproduce the memorized information, only describe its underlying structure.
inciampati··on In the Beginning Was the Command Line (1999)
I read this during my university graduation ceremonies. It was hidden in a fold in my robes. It was the most fitting thing I could have done, as I immediately changed my life direction and focused on exactly the ideas outlined in this work. My goal: move the world from the command line. I've almost managed to.
inciampati··on Ask HN: Fast data structures for disjoint intervals?
Yup, a Vec with binary search is basically all these are. Tree rebuilding is "just" sorting the Vec. To the extent that's fast enough for small updates (merge sort with new elements) you can update rather often without issue.
inciampati··on Ask HN: Fast data structures for disjoint intervals?
You're looking for an implicit interval tree. The key idea is to store sorted ranges and simulate the interval tree via binary search. The best implementation I know of is coitrees: https://github.com/dcjones/coitrees, but you could also roll your own. cgranges is another good implementation: https://github.com/lh3/cgranges
inciampati··on Codestral Mamba
No, you need to provide as much information in context as possible. Otherwise you are sampling from the mode. "Write me an essay about cows" = garbage boring and probably 200 words. "here are twenty papers about cow evolution, write me an overview of findings" = yes
inciampati··on General Theory of Neural Networks
I love your hot take, but you forgot the nonlinear transformation which lets the "god equation" represent literally everything.

The post makes a nice point but it's not really surprising that everything can be modeled by an equation capable of universal approximation.

What I don't get is how genetic systems relate to this. They don't hook into it cleanly and the author just jumps right past them even though they're the most fundamental (biological) system of all those described.

inciampati··on The Illustrated AlphaFold
It's so, so complex! I confess I had a sense of this but had no idea. We don't even hear which MSA algorithm is used to align the protein sequences.
inciampati··on Former OpenAI employee quit to avoid 'working for the Titanic of AI'
Probably exhaustion of human created data. The plateau is coming and much smarter techniques will be needed to go beyond it.
inciampati··on Circos – Circular Visualization
It's a nice visualization system when you have a lot of very long one-dimensional things with nonlinear relationships. This is super common in genomics but in principle could be applied many places.

That said, circos plots are sort of cliche in genomics and I do see people tending to move away from them.

It suffers from an effect of having too many things all together, all at once, layered in ways that sometimes make it difficult to compare. This is especially true when you have data tracks that are circularly laid out around the plot.

inciampati··on Another Climate Truth Bomb
Unfortunately, we aren't lucky. 2024 is just continuing the same insane jump that began in 2023.

See https://climate.copernicus.eu/sites/default/files/custom-upl... and https://climate.copernicus.eu/copernicus-may-2024-streak-glo...

inciampati··on Anti-aging molecule successfully restores multiple markers of youth
The trick is, you want the telomeres to shorten all the way in cells that rapidly divide because this ends cell lines. And if you don't end proliferating cell lines, you get cancer. All of us and all of our tissues have cells that are basically pre-cancerous, they have driver mutations associated with cancer, and the only thing preventing them from running amok is the fact that there are limitations in the number of divisions that are possible. This is all based on information that's become visible via single-spell spatial transcriptomics. See Peter Campbell's work.
inciampati··on Welsh government commits to making lying in politics illegal
I'm not so sure that this article is unbiased. For example, it says that Biden was lying when he said that he was the first president of this century not to have service people dying anywhere in the world. And this has, in fact, been true for the past six months (three people died in combat in Jordan due to a drone strike), which is an extraordinary amount of time for the 2000s. The last time that there was a period without combat fatalities this long would probably be during the 1990s. I don't have any interest in political discussion but I'm pointing out that the article is more useful in showing how much of any political discussion is really about arguing for a certain truth or perspective.
inciampati··on ChatGPT just (accidentally) shared all of its secret rules
The problem is that transformers have a state space which includes attention over their input. Thus, it's very hard to manipulate.

I think you're right, but it will require moving to more recurrent architectures.

inciampati··on Anti-aging molecule successfully restores multiple markers of youth
Long telomeres do not lead to long life. Telomere length has an optional length that's not too short nor too long. Maybe activating TERT in elderly people won't hurt because they won't have time to develop cancer? But it seems like a bad idea based on our tendency to project absolutes: telomeres shorten with age, so obviously we should lengthen them to be young! Right?

Recent research has challenged the notion that longer telomeres are universally beneficial, revealing a complex relationship between telomere length and cancer risk. Having excessively long telomeres is indeed related to an increased cancer risk, primarily because it prevents the natural selection against potentially cancerous cells[1][3].

Key findings from recent studies include:

1. People with mutations leading to abnormally long telomeres have a higher risk of developing various cancers, including melanoma, thyroid neoplasms, and blood-related cancers[1].

2. Long telomeres allow cells to accumulate more mutations over time, as they can divide more times before reaching critical telomere shortening[1][3].

3. The telomere tumor suppressor pathway, which normally helps prevent cancer by limiting cell division, is less effective when telomeres are too long[4].

4. Families with mutations in the TINF2 gene, which regulates telomere length, show a higher incidence of various cancers, including breast, colorectal, and thyroid cancers[4].

5. Genome-wide association studies have identified both rare and common genetic variants that lengthen telomeres and are strongly associated with increased cancer risk[3].

These findings suggest that while short telomeres can lead to premature aging and organ failure, excessively long telomeres may provide a permissive environment for cancer development by allowing cells to proliferate beyond normal limits[1][3][4].

[1] Long telomeres may heighten cancer risks https://www.nih.gov/news-events/nih-research-matters/long-te... [2] Link Between Long Telomeres and Long Life Is a Tall Tale, Study ... https://www.nytimes.com/2023/05/04/health/long-telomeres-age... [3] Long telomeres and cancer risk: the price of cellular immortality - PMC https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6715353/ [4] Telomere shortening protects against cancer https://www.rockefeller.edu/news/29625-telomere-shortening-p... [5] Mice with hyper-long telomeres show less metabolic aging ... - Nature https://www.nature.com/articles/s41467-019-12664-x

inciampati··on A Person of Compute
Yeah :/ was looking for more with a bold title like that.
inciampati··on I've stopped using box plots (2021)
They really help when you're working with huge numbers. It's just a different kind of density plot. A vertical histogram can be nice too. Or you can use color and overlay a few regular old histograms. Go wild.
inciampati··on AI Hype is completely out of control – especially since ChatGPT-4o [video]
I tend to agree with the caveat that we are experiencing exponentially diminishing marginal returns on investment (energy, FLOPS, currency). Yes it's going to get a lot better and smoother to work with. Agents will become secret and someday very soon no one will want to hype that they are using the tech. But we will need another revolution to get the kind of phase change we experienced with the introduction of LLMs. Imo the biggest thing that will lead to improvement now is open models that will get reused and remixed in creative ways.

edit: to be clear, I'm saying that we are running into scaling "walls" that make hard extrapolation based on increased investment senseless.

inciampati··on Gene therapy restores hearing to children with inherited deafness
How perfect can your reaction be? Can you get to 99.999999999999%?
inciampati··on The Intellectual Obesity Crisis (2022)
Key points (summarized):

> You're drowning in low-quality, pointless content designed to hijack your attention. It's making you distracted, confused, and unable to think straight.

> Be mindful of what you consume. Spend more time creating, less mindlessly scrolling.

> Your mind deserves better than the junk clogging your feed. Cut it out before it rots your brain.

That's literally it. I guess the piece is trying to be ironic by using the same style of junk information that it's telling us to watch out for?

inciampati··on 1 in 4 U.S. Teachers Say AI Tools Do More Harm Than Good in K-12 Education
You have to start from scratch. The training data has many examples of almost directly copied blocks of text and code and the model will tend to just repeat itself after several iterations.
inciampati··on New attention mechanisms that outperform standard multi-head attention
No, you can make multiple passes across the data set. There's no reason that you have to use quadratic attention. I point to the shape of our own brains, which are limited in size. Yet, through multiple passes over information and random access, which we guide, we're able to process very large systems.
inciampati··on New attention mechanisms that outperform standard multi-head attention
To be frank, the long-context techniques that you're describing are still extremely limited in the length of context to consider, only a million-token order, and extremely expensive to apply.
inciampati··on Feynman's Garden
You eventually get to the solution if you keep thinking long enough. The trick is to not stop thinking in the background about it. That's the process. I think that hours or days, like the author mentions, even seems very short. Many of the kinds of things that I have experienced, which yielded to this kind of process, took months to years. Sometimes big problems can be broken down into many small ones and only take those minutes to hours to days to complete. But still, the big problem takes a long time to work through.
inciampati··on To the brain, reading computer code is not the same as reading language (2020)
To me reading code is reading math. It feels exactly the same. Same pattern about holding lots of things in mind at once.
inciampati··on Llama3 implemented from scratch
One 3-layer (1 hidden layer) neural network can already approximate anything. You don't even need to stack them.
inciampati··on Llama3 implemented from scratch
Magical thinking. Nature uses gradient descent to evolve all of us and our companions on this planet. If something better were out there, we would see it at work in the natural world.
← PreviousPage 5 of 20Next →